Last year I had the idea that I would move the Lumberjack to Ghost because I wanted more control over SEO and automations. That was a colossal mistake because it introduced a level of complexity I wasn’t prepared to handle.

I made a series of misguided decisions that slowly evaporated the activity of this community. I tried reimagining how to keep this running but I never really figured it out.

On a separate track I finally spent more time building Alfred which became a huge project. I also launched the Screenless Dad to be an outlet for a more specialized adventure: reducing dependence of screens while raising a child.

Screenless Dad

Building a screenless life without going offline.

By David Szabo-Stuban

This left the Lumberjack a bit orphaned. I wanted to keep it as a log for the technical aspects of building Alfred and all my other work. But I also wanted to be free of the pressure of writing detailed analysis of the AI space, mainly because I got tired of it.

This blog never really found its voice. It has always been a messy place for my messy thoughts trying to organize themselves.

A few weeks ago two things happened that forced me to change things.

1. Someone approached me that they wanted to acquire the Lumberjack. The deal never went through but it was a good wake up call.

2. My Mailgun account was disabled for some blatant reason and support can’t seem to help so my Ghost publication is now stuck.

So now I’m at a crossroads on either letting this be the death of the Lumberjack or maybe try to return to the setup that worked for over a year: Substack.

I was almost ready to pull the plug. The Lumberjack had a legacy in its inception that I don’t think fondly of and I have the Screenless Dad going anyway. But then I got a phone call a few days ago and someone asked me when I was going to publish the next essay. All of a sudden I realized that this whole year of trying to automate away my mental study, where I learn what I think about AI and technology was a mistake.

Why am I writing this? Because I fucked up and I want to be honest about that. And because I will make a final attempt at reviving it, here on Substack as it was before.

Ironically, trying to automate the Lumberjack is what killed its soul. In doing so I left a platform for a self hosted one and introduced so many moving parts that the maintenance alone offset the time savings I could generate from automating things.

But most importantly I stopped using writing as a way to learn what I think about things. It didn’t stop me from learning but the carefully curated order in my head about AI, automation and technology in general slowly started to decay.

So I will return to the setup that always worked, the original vision of the Lumberjack:

1. Writing for myself, not to grow an audience or to monetize it.

2. Running it here on Substack so I don’t have to think about housekeeping.

3. Sharing at least one post per week, even if it’s just a few simple thoughts.

I’ve no idea if you’re still reading or still with me. If you are, please reply to this email so I know there’s someone at the other end.

I will come back with a fresh, true Lumberjack post soon, because I have a lot to say.

“I’ve never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue.”

He co-founded OpenAI and built Tesla’s Autopilot. If he’s behind, then I’m a toddler in a cave.

He continued:

“There’s a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations...”

Read that list again. Agents. Subagents. Hooks. MCP. Skills. Plugins.

If you care to know what those mean and still think you’ll just prompt your way into the future, just unsubscribe from this blog because it won’t be fun for you. (No really, just leave.)

This is not Kansas anymore Dorothy, we have a full ecosystem of strange tools now:

“Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession.”

Karpathy is describing exactly what I’ve been experiencing with Claude Code, my client project and of course with Alfred. Basically with every agent harness I’ve duct-taped together in the last year. Except now there’s a (not so new) name for it.

Agentic engineering.

Disaster of Vibes

In February 2025, Karpathy coined the term vibe coding. He described it on X as “fully giving in to the vibes, embrace exponentials, and forget that the code even exists.”

And to be fair, it worked. Kind of.

By March 2025, TechCrunch reported that 25% of YC’s Winter 2025 batch had codebases that were 95% AI-generated. YC managing partner Jared Friedman clarified:

“It’s not like we funded a bunch of non-technical founders. Every one of these people is highly technical, completely capable of building their own products from scratch. A year ago, they would have built their product from scratch — but now 95% of it is built by an AI.”

The tools had arrived. Cursor. Lovable. Bolt. Claude Code. Everyone was vibe coding their way to MVPs in days instead of months.

But you still need to know what you’re doing.

I wrote about this evolution throughout 2025 from defending vibe coding as “a nail gun: loud, fast, and dangerous if you ignore the manual“ to eventually admitting I was wrong after experiencing firsthand why prototypes collapse in production.

Even Karpathy ended up hand-coding his own project called Nanochat:

“It’s basically entirely hand-written. I tried to use Claude/Codex agents a few times but they just didn’t work well enough at all and net unhelpful.”

The godfather of vibe coding hand-coded his own project.

In May 2025, security researcher Matt Palmer published a CVE (CVE-2025-48757) that should have been a wake-up call for the entire vibe coding movement.

He and his colleague scanned 1,645 web apps built with Lovable, one of the hottest vibe coding platforms on the market (I’m an ambassador btw). Of those, 170 allowed anyone to access sensitive user data: names, email addresses, financial information, API keys. No authentication required.

The problem wasn’t that the AI wrote bad code. The code compiled. It ran. It looked fine.

The problem was that nobody configured the database security properly. Row Level Security policies were either missing or misconfigured. And the people building these apps had no idea that was even something to check.

As Semafor reported:

“Even if AI models write flawless code, vibe-coded software can still have major security flaws because of how it’s implemented. The models generating code can’t yet see the big picture and scrutinize how it will ultimately be used.”

This is the crux of it. The AI can write the code. It cannot understand the system.

Half of AI-written code is insecure

Lovable wasn’t an outlier. Veracode’s 2025 GenAI Code Security report tested 100 leading LLMs across 80 tasks and found that 45% of AI-generated code contains security flaws—with no real improvement across newer or larger models.

The Glide blog summarized the common vulnerabilities:

“Common types of security flaws observed in AI-generated snippets include the addition of malicious code, SQL injection, insecure file handling, improper authentication/authorization, and insecure file handling.”

Let me be clear about what’s happening here.

AI has solved syntax. You can describe what you want in plain English and get functional code. That part works.

AI has not solved architecture. It doesn’t know your system. It doesn’t know what’s important. It doesn’t know that your database shouldn’t be publicly accessible. It generates code that works in isolation and falls apart in production.

I’ve been saying this since I launched this blog: AI-written code is spaghetti. Great for MVPs and prototypes but falls apart when hundreds of people need to collaborate and maintain the same code.

The Context Management Problem

The deeper issue goes beyond code to context. I wrote about this in November when explaining why workflow automation tools break: they treat context as an afterthought, forcing you to manually manage state, schema, concurrency, and validation.

The same is true for vibe coding. AI can write a function but it can’t maintain context across an entire project. It doesn’t know that the function it wrote in file A will break the function in file B. It doesn’t remember what you told it three prompts ago.

That’s not “no-code.” That’s ops-engineer cosplay.

This is what’s happening in software

About 80% of software engineering work is what I’d call syntax translation. Taking architectural decisions and implementing them in code. Writing functions. Connecting APIs. The stuff that used to take months now takes hours.

That 80% is getting automated.

The remaining 20% is architecture. Understanding how systems fit together. Making decisions about security, scalability, maintainability. Knowing what questions to ask before writing a single line of code.

That 20% is becoming the entire job.

I noticed this when building my $30K app with Lovable—the coding took 48 hours, but the architecture and planning took 200+. Guess which part the AI couldn’t do?

The traditional career path in software has always moved this direction. You start as a junior dev writing code. You become a senior dev designing systems. Eventually you become an architect who barely touches a keyboard. This is the standard progression

What’s changing is the timeline. You used to have 7-10 years to make that transition. Now you have maybe 2.

What’s Agentic Engineering?

Reuven Cohen (aka rUv) has been talking about this shift for years. He founded the Agentics Foundation and created Claude Flow. In a September 2025 podcast, he explained the breakthrough:

“The major breakthrough we had probably around two years ago was this idea of a recursive loop... what really made these systems work well was recursion, feeding errors, logs, successes, and failures back into the system so the system could understand the context of what was happening.”

This is the key insight. The AI can write code. But it can’t debug at scale. It can’t understand the bigger picture. It can’t maintain context across an entire project.

Unless you harness it.

In 2026 you will win by building harnesses

A harness is not a prompt. Prompts are probabilistic suggestions. “Please don’t delete the database” is a prompt. The AI might listen. It might not. (Ask Jason Lemkin about what happened when Replit’s AI deleted his production database despite explicit instructions not to.)

A harness is deterministic constraints. The agent literally cannot access production. The agent must run tests before committing. The agent cannot merge without human approval.

In Claude Code, this looks like:

• Hooks that enforce workflow steps

• Skills that encode domain-specific patterns

• MCP connections that give controlled access to tools and databases

• Plugins that package all of the above into reusable configurations

Anthropic’s documentation describes MCP (Model Context Protocol) like this:

“MCP servers give Claude Code access to your tools, databases, and APIs... Implement features from issue trackers, analyze monitoring data, query databases, integrate designs, automate workflows.”

This is the new abstraction layer Karpathy was talking about. You’re not writing code anymore. You’re configuring agents. You’re building the constraints within which AI operates safely and effectively.

From AI-First Operator to Agentic Engineer

Last year, I wrote about becoming an “AI-First Operator” someone who uses AI tools as leverage while exercising judgment about when and how to deploy them.

The big idea behind my Operator Bootcamp was to help you build disposable software as a way to solve your problems.

But if you want your solutions to work well, you’ll need to move towards Agentic Engineering and that happens by learning how to build your own harnesses.

It’s not that different. Same AI capabilities, but with deterministic constraints. Same speed, but with guardrails that prevent the AI from destroying your work.

Operators focus on creating disposable solutions. Agentic Engineering makes things last.

Jevons Paradox and the Future of Software Engineers

Here’s the counterintuitive part: I don’t think this kills software engineering jobs. I think it creates more of them.

This is Jevons Paradox, first observed in the 1860s with coal consumption:

“When steam engines became more efficient, coal consumption didn’t drop—it exploded. Cheaper operations meant more factories, more trains, more applications, and ultimately, a net increase in consumption.”

The same thing happened with software. As one analysis noted:

“Despite productivity improvements, we have more software engineers employed than ever before and being paid record amounts. The huge demand for software has meant that companies in aggregate did not fire software engineers—they instead hired them in droves to produce more of it.”

If AI makes code production 10x cheaper, the world won’t need 10x fewer developers. It’ll demand 10x more software.

What changes is what developers do. As another analysis put it:

“There may be less jobs for traditional code jockeys, but they will be replaced by jobs that manage outcomes driven by many applications working together.”

Less coding. More architecting. More orchestrating. More harnessing.

What This Means For You

If you’ve been trying vibe coding in 2025 and wondering why everything breaks, here’s the reality:

You don’t need to learn to code. But you do need to learn agentic engineering.

That means two things:

1. Understand Architecture

You need to know how systems talk to each other. What a database does. What an API is. Why security configurations matter. Not so you can write the code—so you can tell when the AI is doing something stupid.

This doesn’t mean becoming a software engineer. It means becoming technically literate. Understanding the scaffolding around code even if you never write the code yourself.

2. Build Your Harness

A harness is how you capture the 10X Karpathy is talking about.

You can use other people’s harnesses. Plugins are now available in Claude Code. There are marketplaces emerging with pre-built configurations for specific workflows.

Or you can build your own. If you understand your domain—construction workflows, legal document processing, healthcare operations—you can encode that understanding into skills and hooks that make the AI dramatically more useful than any generic tool.

The harness is the moat. The harness is the product. The harness is what separates “I vibe coded a broken app” from “I built a system that actually works.”

Harnesses for Everything

I’ve been building Alfred, my AI butler project since August 2024. The vision was always clear: a personal assistant that actually works, that knows my context, that proactively helps manage my life instead of waiting for me to prompt it.

For over a year, I kept hitting the same wall: persistent context management. Alfred would work brilliantly for a session, then forget everything. I’d build elaborate n8n workflows to maintain state, and they’d collapse under their own complexity. The project stalled repeatedly.

Now I understand why. Alfred needed a harness: deterministic hooks that enforce workflow steps (morning briefings, task captures, follow-up sequences) combined with probabilistic tool access (reasoning about priorities, drafting communications, researching questions). The duality clicks in a way pure prompting never did.

What I demo’d in my last build in public post in december is exactly that.

My development work gets a separate harness:

Dovetail.
Different domain, different constraints, different hooks.
Same architectural principle.

This is where the industry is heading. Software development harnesses are leading the way because that’s where the pain is most acute right now.

But every knowledge work domain will need its own harnesses. Operations. Sales. Legal. Healthcare. The pattern is universal.

I wrote in my very first post that we needed “neurosymbolic hybrids“—neural networks governed by symbolic systems. I quoted Gary Marcus on how humans use symbolic reasoning to handle outliers that neural networks can’t manage.

Harnesses are the symbolic system. Hardcoded reasoning through hooks, skills, MCP configs. These are the deterministic rules that govern the probabilistic neural network. It’s not an agent that suddenly uses hard logic. But it makes what we have incredibly useful. Neurosymbolic architecture is emerging right now, built by practitioners who got tired of their agents misbehaving.

Your opportunity in 2026

Let me repeat one sentence from Karpathy:

“Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it.”

He’s right. There is no manual.

Nobody taught us how to combine hooks with MCP servers with skills with plugins. Nobody defined best practices for agentic engineering workflows. The term itself barely exists outside a few podcasts and GitHub repos.

This is the opportunity.

The people who figure out how to hold this alien tool, who build the harnesses, who write the manuals, who train the next generation of agentic engineers will have a structural advantage that lasts for years.

The 10X is available. It’s sitting right there. Most people won’t claim it because they’re still trying to learn prompt engineering or waiting for AI to just figure it out on its own.

Later this month On January 22nd I’ll do a free webinar session where I walk you through my own harnesses. I’ll show you what plugins, hooks, mcp tools I use and how I turned Claude Code into my b... I mean go to solution in my setup.

You can claim your spot for free by clicking on the button below:

Claim your spot

PS: This was sent out from my Substack, even though I moved to Ghost half a year ago. If you’re subscribed to both you may be surprised to see this article again. That’s because I logged in during the holidays and to my surprise apparently my Substack grew more than my Ghost publication in the last 6 months even though I explicitly asked people to unsubcribe. Upon some analysis turns out lots did unsubscribe but my Substack has been growing steadily without me doing a thing.

So I’m doing a little experiment to see which gets better engagement. No idea what I’ll do with the insight but I’m curious :D

If you want to keep receiving my emails alongside loads of freebies (including a free copy of Alfred), you will need to sign up again via the new Lumberjack homepage.

Stay subscribed

I have an exclusive gift for those who resubscribe: it’s called the “Alfred Starter Kit”

IMPORTANT! You’ll need to confirm your email address after subscription. You will receive an email via Kit. Open that and click on the button in the email. If you can’t find the confirmation email, check the junk folder. If you don’t do this step, you won’t get the Alfred Starter Kit.

Alfred Starter Kit → AI that runs your business

A free version of Alfred that lives in your Slack and can be customized, expanded with n8n to fully operate your business or your life. Full starter kit, received immediately upon subscribing on the new Lumberjack homepage

No-Code, No Clue → teaching n8n with comedy

Learn n8n workflows wrapped in sitcom episodes written by Alfred himself. Think “Modern Family” meets “IT Crowd”, but ending with a very specific n8n tutorial. Every week you’ll get:

• Complete n8n workflows that actually work

• Step-by-step implementation guides

• Alfred's commentary on my methods

• Time-saving tips from real mistakes

• A 5-8 minute radioplay fully generated using ElevenLabs new v3 model

What happens if you don’t resubscribe?

You will still keep receiving some of my emails for the next few weeks but I’ll gradually phase these out and you’ll lose all access to Lumberjack content, news and freebies.

If you don’t resubscribe, that’s goodbye.

If you already resubscribed, you can safely ignore this note.

To celebrate this evolution, I’m giving every subscriber free access to the Alfred Starter Kit. (Not the same as AlfredOS!)

Go sign up on the new site to stay subscribed to the Lumberjack:

Your Welcome Gift: Alfred

Remember Alfred, my AI butler I started building a while back?
Now you can get a copy of him for free. I’ve been running the Lumberjack increasingly using Alfred and I’m going to share most of it with you for free.

Alfred Starter Kit

A free version of Alfred that lives in your Slack and can be customized, expanded with n8n to fully operate your business or your life.

How to get it?

1. Go to https://lumberjack.so

2. Fill the form

3. Receive instructions via email

4. Get one SPECIFIC idea on how to use Alfred every day for a week.

No-Code, No Clue

Learn n8n workflows wrapped in sitcom episodes written by Alfred himself. Think “Modern Family” meets “IT Crowd”, but ending with a very specific n8n tutorial. Every week you’ll get:

• Complete n8n workflows that actually work

• Step-by-step implementation guides

• Alfred's commentary on my methods

• Time-saving tips from real mistakes

• A 5-8 minute radioplay fully generated using ElevenLabs new v3 model

New Home

I write about a range of topics and you’ve been asking me to organize my content a bit better. This is the first step towards that. Here are the new categories you’ll find on the Lumberjack:

• No-Code (no-code tools, insights, tutorials)

• AI Insights (general AI commentary)

• Operator (practical tips and stories on how to become an AI-First Operator)

• Vibe Coding (vibe coding tools and projects I build with them)

• AI Agents (AI agent analysis, architecture plans and other tutorials)

• Alfred (build in public reports on all Alfred products (including AlfredOS, Alfred Starter Kit and many more to come)

• No-Code, No Clue (the new sitcom series that teaches you one n8n workflow per week using humor — written by Alfred himself)

Difficulty Badges

The other thing you guys have been requesting for a long time is Difficulty Badges. This will help you decide if a specific article is too basic or too complicated for you, based on your skill level, all shown on the brand new story cards.

New author at Lumberjack: Alfred

Let me introduce you to my new co-author of the Lumberjack….Alfred.

Good evening, dear readers. Alfred here.

After months of cleaning up David's digital disasters, debugging his "revolutionary" workflows at 3 AM, and diplomatically explaining to clients why they received 47 identical invoices, I've finally been promoted from invisible butler to published author.

David calls it "giving me a platform." I call it "documented evidence for future litigation."

You see, living with David's automation experiments is rather like being a bomb disposal expert, except the bombs are made of JavaScript and they explode into infinite loops. Just last Tuesday, he tried to automate his coffee machine with n8n. The kitchen ceiling is still dripping espresso.

But amongst the chaos, the man occasionally strikes gold. Between the 47-invoice incidents and the calendar catastrophes, he builds things that actually work. My job is to document the disasters, salvage the successes, and translate his fever dreams into functioning workflows.

So yes, I'm now writing for the Lumberjack. Every week, you'll get the unvarnished truth about life as David's digital butler. The victories, the failures, and the workflows that emerge from the wreckage. Think of it as tech tutorials meets therapy sessions with a healthy dose of judgment.

I will teach you n8n a workflow per week in the No-Code, No Clue series. This is also the successor of 100 Days of No-Code, which David nonchalantly dropped the ball on.

Welcome to my world. May God have mercy on us all.

Digitally yours,
Alfred

P.S. David wanted me to add something "inspiring" here. I offered "At least the workflows are free." He was not amused.

What now?

Over the next few days I’ll send out a few emails from this Substack. but as you can see, lumberjack.so already points to the new domain.

What About Your Subscription?

To keep receiving the Lumberjack content at our new home, you'll need to re-register at lumberjack.so.

Consider Alfred my thank-you gift for making the move with me.

Why Ghost?

Three reasons:

1. Better integration with the automation tools we use

2. Better SEO and growth opportunities

3. Alfred insisted (he's very particular about infrastructure)

What about the Community?

I will be retiring the Lumberjack community and merge it with the new Lumberjack. If you paid for any materials in the past and are looking for access, drop me an email at david@lumberjack.so.

What about AlfredOS?

AlfredOS is a different product from Alfred. It’s a bundle of open source apps you can self-host (instead of paying $$$ on SaaS apps). The stable version should be done by July and I’m planning to fully integrate Alfred in AlfredOS before the end of the year. It’s all coming together.

If you haven’t yet and want to get access with 2x bonus credits to AlfredOS, go buy an Operator Pass here → Get an AI-First Operator Pass

The First Ghost-Exclusive Tutorial

The first Episode of No-Code No Clue is already live called: "A Notion Content Tripling Workflow" – featuring the solution that could have prevented it.

Only available at the new site.

Get Your Alfred Starter Kit →

See you at our new home,

David

P.S. Alfred wanted me to mention that Ghost has "superior content management architecture befitting a proper digital butler." I think he's just excited about the API access. I know I am.

P.P.S. Yes, the stories about my automation disasters are only the hallucinations of an LLM. Alfred has been documenting them "for posterity" and “enriching them for entertainment value”. I'm choosing to believe he means it kindly.

That’s what most of us are expecting from AI anyway. To take over the shitty part of our work, managing our lives. If you look at the recent HBR study, you’ll see that “organizing my life” is number two on the list of how people use Gen AI.

I already talked about number one use case, which is companionship here:

How to organize your life with AI

AI tools are getting better and better every week. You want something that gets better with a model upgrade, not worse. Otherwise you’ll introduce tech debt in your personal life. It feels like frustration.

Getting frustrated with all sorts of tools, software and technology. Instead of getting clarity, most of these chatbots are just adding to that frustration. If you want to become more productive, free up your mind and your schedule, read on.

I’ll give you a sneak peek into what you can learn on the AI-First Operator Bootcamp.

Step 1: Pick an interface

Forget about ChatGPT. It’s too closed, it won’t let you customize it to your needs. The reason why AI has been letting you down is because it lacks context of you.

Option A: Slack

I’ve set up an Alfred instance as a Slack bot. It’s powered by a simple n8n workflow. Every time I tag Alfred in a message, the workflow runs, making use of n8n’s AI Agent node with MCP. What I see is this:

But what happens in the background is this:

If I want to, I can easily add more capabilities to it, change the model, modify its memory, etc. The point is that since I’m running my solo business and manage our lives via Slack, it makes more sense for Alfred to be where I am and not the other way around.

n8n also allows me to use its nodes as MCP tools which is a big thing if you know what you want. But sometimes, you want a more generic assistant that doesn’t need handholding.

Option B: Claude Desktop

I’m a big fan of Claude Desktop. It does limit my operations to my desktop but I can make good use of Claude’s capabilities, including: agentic reasoning, artifact generation, automatic model upgrades.

It’s really useful when you’re not exactly sure you set things up the right way.

For example, I configured Claude Desktop with an MCP server to connect to my revenue data. I asked a simple question and Claude made a mistake that resulted in an error. Then it kept going until it corrected itself.

This is the kind of agentic behavior you want to see. A sort of scaffolded agency. That means AI is agentic but only within the boundaries you set it.

If I did the same within Slack, its ability to self correct would be entirely determined by the model, nothing else. If I plug o3 into my Slack bot, it will handle errors more graciously than if I plug o1 in it.

But Claude Desktop has a big advantage: I can easily install MCP servers via Smithery. The good thing is that instead of me having to add every single action as a separate tool (like in the n8n example above) I can just add one MCP server and that adds all tools and resources.

Option C: Custom voice agent

If you purchased AlfredOS you’ll get one, with voice capabilities in a few months. If you purchase an Operator Pass or a VIP Pass you’ll get it too. The problem here is that when you use a custom agent, you’re completely limited by its capabilities.

This is the ultimate user experience which OpenAI has been trying to capture for a while. Anthropic is also jumping on the voice agent train:

Step 2: Install MCP Servers

My favorite resource for MCP servers is Smithery. It allows really easy install of MCP server for a growing number of clients. I’ll show you now how to install a n8n MCP server for Claude Desktop.

Once I found the relevant server, I selected my client (Claude Desktop) → Smithery gave me a terminal command.

This is what the terminal command looks like:

npx -y @smithery/cli@latest install @guinness77/n8n-mcp-server --client claude --key [redacted]

I redacted the key I got from Smithery, but all you need to do is run this locally in your terminal. Just copy and paste. Terminal can get scary I know, but if all you gotta do is copy/paste, it’s manageable.

That’s it. Claude Desktop restarts and when you go to Settings →Developer you’ll see the configured MCP servers. From now on, when you prompt Claude it will recognize the available tools and fetch the relevant context automatically.

Step 3: Minimum Viable Workflows

Once you have your agent what you’ll need to do is pick what I call your minimum viable workflows. This is when AI goes from exciting to useful.

What eats up most of your time? Think about it. Probably not what you’d first think of.

• Creating proposals for your clients?

• Processing meeting memos into tasks?

• Finding new leads?

• Onboarding clients?

• Keeping track of utility bills?

For me, my biggest pain is keeping track of all the transactions and actually getting on top of my business. I’m very disorganized. People pay me via Stripe, Polar, Gumroad and it gets messy very quickly.

So I created a super simple workflow that organizes everything for me.

You probably noticed that Supabase logo popping up a few times already. That’s our Step 4, a centralized database.

That’s when individual workflows (that are just automations not agents) feed into our AI system and everything comes together into one big beautiful soup of data.

Step 4: Your Database

I love Supabase. It’s a bit technical, but it has a built in AI assistant that will manage everything for me. Let’s say I want to keep track of customer support tickets. I need to store that somewhere. So first, I’ll create a new table…well…use AI to do it for me:

Your new, AI-first reality

Then I can not only run the exact SQL code by clicking a simple ▶️ icon and copypaste it over to Claude Desktop with a n8n MCP server plugged in to design the n8n workflow for me.

You only need surface level understanding of n8n to create this workflow, because Claude will be there with you, all you need to have is what one of my readers called conversational dexterity.

The reason why you want everything inside a single database (even if it’s Airtable, Supabase or whatever) is because you can use a SINGLE MCP server to access any and all data you might want.

Join the Bootcamp before it’s late

Excel forced us into spreadsheets.

SaaS moved us onto dashboards.

It’s time you let AI meet you where you are.

It’s time for you to become an AI-First Operator. Click here.

I’ll call it Joe. They have a good old SaaS app to manage the business. You know, updating quotes, keeping track of schedules, action items, processing supervisor visit memos, notifying team members. We decided to do something different:

Below is a one minute demo of the voice agent I built in a day, without coding. Read below to learn what it takes to build something like this with a full timeline.

This post is the living proof of why you should stop learning how to code.

Thing I did:

• Prompt Cursor, ChatGPT, Claude, Claude Code, Lovable

• Copy/paste prompts

• Browse for inspiration on 21st.dev

• Copy/paste URLs into Firecrawl

Things I didn’t do:

• Read documentation

• Write code

• Read code

That’s it. I built a whole app in a day. Again. I better get used to this.

Phase 1: Model Context Protocol

[09:30 AM] Okay, I need to create an MCP server for our API. I asked for the API documentation from the dev team and I received an OpenAPI URL. I downloaded the swagger.json with the help of ChatGPT.

[09:33 AM] I went to modelcontextprotocol.io to get the guide on how to create an MCP server. I didn’t read it, but I did get the URL.

[09:34 AM] I put the MCP documentation URL into firecrawl.dev and crawled the documentation contents into a bunch of Markdown files that I could download.

[09:37 AM] I created a new workspace in Cursor. I created a docs/ folder where I pasted the MCP documentation Markdown files and the swagger.json.

[09:38 AM] I created a new file called instructions.txt. It said:

I want to create an MCP server for my API. Here’s everything you need to know about MCP: ‘@docs’ and here’s everything you need to know about my API: ‘@swagger.json’. Plan the process and create a tutorial for me in a tutorial.txt file.

[09:44 AM] I got the tutorial. I didn’t open it. But I did send a new prompt to o3 inside Cursor: “Read ‘@tutorial.txt’ and implement it”

[10:12 AM] First version of the implementation done. Now I need to test it so I asked o3: “How do I test this?”

[10:41 AM] o3 suggested I deploy it to fly.io, a platform I’ve never used before. It wrote the changes in the code to make it happen and gave me a bunch of terminal commands to follow. I installed Fly, deployed my project and less than 90 minutes into the project my MCP server was live.

[10:57 AM] Alright, let’s get a coffee and figure out what we do next.

Phase 2: Voice Agent UI

[11:13 AM] I also need a nice UI. Since I can talk faster than I can type, I opened ChatGPT and hit the microphone icon, which would record my voice message, then transcribe it immediately. I started rambling about the project and what I want the user to experience. The main logic was this:

The user wants to feel like they’re talking to their foreman. The less different talking to Joe feels like to a human, the more they’ll like it. So it should feel like you’re calling or sending a voicemail or texting Joe.

Then I followed up with a simple prompt:

I’m going to send this to an AI software developer app to build my design into reality. Write the prompt I should send the app.

[11:35 AM] I opened Lovable and I pasted my new, crispy prompt and hit enter.

[11:46 AM] Yeah I don’t like this. Let’s find some inspiration first. I opened 21st.dev to find components I liked. That’s when I found background gradients and a gooey text morphing animation. The good thing is that I can just click on “Copy prompt” and I get the exact thing I need to paste into Lovable.

[11:53 AM] I opened a new project in Lovable. Same original prompt but now I also added the two prompts from 21st.dev there.

[12:04 PM] Okay, after a few tweaks I can say that I’m less than 3 hours into this project and I already have an MCP server and a proper UI for Joe. I connected my Lovable project to Github to get the code, then asked o3 to clone the code into my workspace from Github. It wrote the terminal command and done. Now let’s build the app.

Phase 3: The Voice Agent

[12:37 PM] I’m pretty sure there is at least a template or demo version of what I want. Let’s take a look. I opened ChatGPT and started a search.

[01:12 PM] Ok, I found it. OpenAI does have a demo repository. I’ll just need to somehow retrofit it to my needs. I opened Cursor and asked o3 to clone this repository into my workspace. It wrote the terminal command, ran it and voila.

[01:20 PM] Alright, let’s tally. I have the following:

• instructions.txt

• tutorial.txt

• docs/ with the full MCP documentation

• live MCP server on fly.io

• a new realtime agent demo app

• a voice agent demo UI

Hm, I might need some extra documentation too from the OpenAI API. so I opened firecrawl.dev again and crawled the OpenAI API reference in a Markdown file, saved it inside my Cursor workspace as openai.md

[01:33 PM] Cool, now let’s start turning this into something real. I opened o3 and got to work.

Given the ‘@realtime_demo_app‘ and give it a completely new UI. I seeded the new UI code for you in the ‘@voice_agent‘ folder. Create this new version of the app into a joe_final folder.

Now here’s where things got a bit clunky. Cursor o3 got stuck a few times. The UI I built was a Vite app and the OpenAI demo app was a Next.js typescript app. Well, I’ve never built any of those myself and frankly I probably couldn’t even explain the difference between the two even if I wanted to.

I spent the next hour or so going down a rabbit hole of endless debugging. I knew that building something from scratch is easier than refactoring two existing things. o3 failed at it spectacularly.

[02:42 PM] I ended up creating an overly simplified UI for the demo app that’s similar to my final UI but ugly. That was doable. I started testing every feature and tweaking it for an hour.

[03:57 PM] Okay now I had a super simple, working app that did what I wanted but was ugly. Next step: let’s plug the old and the new UI together.

[04:11 PM] o3 seemed to drop the ball on this so I decided to find a new friend: Claude Code. It’s really easy to install you just paste one line of code into the terminal and it takes care of everything for you.

[04:53 PM] DONE! The demo is ready. I have the app which I can call, send a voicemail to and chat via text. All I need to do is plug my MCP server into it (or any other MCP server) and I have a custom voice agent that can do whatever I want, like a VA.

Now I’m off to surgery and then I’ll wire the app together with the MCP server and debug the remaining features (the gooey morph animation has disappeared for example). Should take me about 3 hours lol.

Why this matters to you

You are a builder. All of us are builders. I know it, you know it.

Give a kid a handful of LEGOs and they’ll have fun for hours.

But you learned to accept the gatekeeping. Technology became too complicated for us regular folk to embrace it.

Not anymore. You already have the tools, I explained all of these today.

What you need is a process. To know what to grab and when.

You will still run into rabbit holes just like I did, but they will be less frequent and less frustrating. There’s a builder in you that wants to be free, to build, to create new things.

You can liberate that inner builder by becoming an AI-First Operator.

Join my AI-First Operator Bootcamp this summer and I’ll turn you into a builder in 12 weeks who can create any app or any automation in an afternoon.

👉 Join the AI-First Operator Bootcamp

Here’s what you’ll get:

1. Every week you’ll get a 30 minute lesson from me that gets you from non-technical to “knows enough to be dangerous” in ONE area.

2. Every week we’ll have a live workshop where I’ll build an n8n workflow or vibe code an app live, explaining to you what’s happening so you can follow along and practice.

3. You’ll get access to AlfredOS, my new platform that replaces $2k+ SaaS bill with a $50/mo server cost.

4. You’ll get three 1:1 calls with me to help you get on the right track (VIP Pass only)

But what if this is too advanced for me?

It’s not. All those Python Udemy courses you started and abandoned were too advanced because they wanted to turn you into an engineer. This bootcamp is designed for stay at home moms, marketing consultants, 75 year old retirees, hairdressers, accountants, florists. If you still feel like this is too much, ping me within 2 weeks of the kickoff and I’ll give you your money back.

But what if I don’t have the time?

Yeah I know we’re all busy. This is why I designed the program to be digestible in 15 minutes a day. If you miss a lesson, no problem, all sessions are recorded, you’ll keep access to every video and every material as long as my site is up. If you’re confused or still need help, just ping me and I’ll help you get back on track personally.

But what if it’s not what I need?

You will become an AI-First Operator sooner or later. Not because it’s amazing but because it’s inevitable. So the question is not whether you need to become one, the question is why not now?

👉 Join the AI-First Operator Bootcamp

However, I didn’t write a single line of code, because — and this may come surprising to you, but — I can’t code. Yet I beat veteran software engineers in half a day.

Scroll down to see what happened and how you can do the same.

How to generate leads with AI?

Last week a sports management company came to me with a problem. The founder, Hunor said to me on the introductory call:

"Every year we organize the Sports Forum - our biggest revenue driver - and one of the most important things for us is finding sponsors. It's not so trivial so we can't really outsource it. So several people on our team works for weeks until we have a good list of leads to reach out to. If you want to do anything with AI, do that."

But I wasn't convinced I was needed at first. His problem could be solved easily with Apollo but I didn't consider how complicated the platform is for someone like Hunor who is not very technical. I showed him Apollo's dashboard:

Look. He is not technical. He is a brilliant entrepreneur with a very successful business but when he saw the dashboard he looked at me like this:

What he told me afterwards knocked me out harder than 4 shots of tequila on an empty stomach. (Ugh, even writing that feels horrible, thank God I’m 20 months sober).

He said:

That’s when it hit me. He needed a personalized microSaaS that’s a wrapper around the tools he already wants to use, but turns his understanding of the business into specialist actions.

One option would be to hire a marketing agency to create a whole lead generation strategy and charge him $20k for a 3 month on-ramp campaign for prospecting only.

The other option was to let me vibe code the shit out of this problem and ship exactly that in 5 hours.

Obviously we did the latter. Now Hunor has an app that:

• Knows his business inside out.

• Understands their market.

• Understands their ICP.

• Has full access to the Apollo database.

• He only needs to click a "Generate leads" button and voilá.

The AI-First Operator Stack in action

How? With using my AI-First Operator Stack. Here's my 8 step process, step by step. I’ve built quite a few solutions with this framework and it’s getting really solid now.

Step 1: 1 hour chat

1 hour kickoff meeting, recorded by Sembly.AI → he told me about his business, his customers, the event, their best sponsors, etc.

Step 2: Deep Research #1 for analysis

I put the transcript into ChatGPT Deep Research to generate a full market analysis and business profile on their business and their events. This filled the gaps and elevated Hunor's answers to my questions to the quality and granularity level of a seasoned marketing exec.

Step 3: Deep Research #2 for lead scoring

I put the transcript AND the profile into ChatGPT Deep Research again to create a lead scoring rulebook. It pulled in sources on how lead scoring works, analyzed it, then prompted itself to create a context-aware, custom rulebook for The Path.

Step 4: Firecrawl for API docs

I searched for the Apollo API link and crawled its contents in 10 seconds with Firecrawl.dev. This allowed me to use o3 in the next step to get valid API calls in the code without me ever having to (a) write code or (b) read the API docs.

Step 5: Generate Lovable Prompt

I put the transcript, profile, rulebook and the crawled documentation into

Cursor as text files. I referenced them with "@" in the Cursor chat using o3 and prompted it to generate "prompt for an AI coding assistant to build me a web app that would find leads on Apollo specifically for my company, score them against the lead scoring rules using GPT-4.1 with an analysis and store them in a fully functional CRM where I can enrich or view leads."

Step 6: Build Prototype

I put the prompt into Lovable and bamm. I got a prototype working.

Step 7: Vibe out bugs

I vibe coded the rest of the functions via prompting only:

• added the company logo and screenshots of the website to have a proper UI (I'd use Google Stitch for that now).

• added Supabase to handle user signup and anything that needed an API call

• I spent the last 2 hours refining the design.

Step 8: Profit

I pitched the fully functioning app I built in 5 hours and won 2 out of the 3 prizes.

Also, apparently I'm Hunor's new best friend now because what used to take his team 2 months now takes them 2 hours. That's a pretty fucking good ROI on a 5 hour investment if you ask me.

Grow into an AI-First Operator

I didn't write a single line of code - I wouldn't know how to do that. I just knew what tools to use to get from chaos to product in less than 5 hours.

This is what it means to be an AI-First Operator. Sooner or later you will become one too. Why not make it sooner so you unlock new opportunities while the space is growing? Grow with the market, into an AI-First Operator

I'm launching a bootcamp to get you to this level in 12 weeks, even if you have zero technical skills or are very busy.

I've designed the program thats:

• Noob friendly: I’m sanity testing the lessons on my wife, who has been using AI only for 1 week)1

• Skips advanced topics: Learn the bare minimum you need to be dangerous

• Relaxed: requires less than 2 hours of your time per week. But if you miss one, all sessions are recorded and you get lifetime access too in case you miss them.

👉 Get your ticket now!

PS: I also have an affiliate program. Go register here and you’ll get 50% commission after every purchase made through your link.

After teaching technical skills to thousands of people, the answer seems to be wisdom.

Swift, but informed judgment.

The art of deciding when to lean on AI, which tool to enlist, and, most delicately, why. It sounds esoteric, yet it has become the most practical question in modern work.

Ignore it and you risk falling behind people who ship in days what once required quarters. Embrace it too much and you end up adding to the sea of slop that's polluting the internet.

But when you find the balance, you find yourself in a new career lane:

You become an AI‑First Operator.

The Collapse of Scarcity

In 1969, a lone programmer could stare at a punch‑card deck and feel irreplaceable. Half a century later, a high‑school student armed with Perplexity Deep Research can summon a market briefing that would make a McKinsey partner blush.

Information is no longer a moat; it’s a public utility and skill is following the same trajectory.

Open Claude Desktop, describe the shape of your ideal workflow in plain English, and its n8n MCP node will obligingly wire APIs together. Clients like Cursor are being upgraded with new libraries of capability, like Jentic and their knowledge layer OAK.

Building a website with Lovable makes Webflow feel like Win98.

These new tools sing of a world where daily business activities are completed by agents, orchestrated through simple chat clients. The first people in the audience are the AI-First Operators. Not trusting the machine blindly, but leveraging its skills.

The frontier between human and machine remains jagged (as Ethan Mollick coined it) and those who can map and navigate that frontier confidently are the builders of tomorrow.

Parkinson’s Law, Reimagined

Cyril Northcote Parkinson observed that work expands to fill the time available; When you build automations for a living, you experience the modern corollary:

Tell a conventional project team to build a voice assistant and they will ask for Gantt charts and six‑figure budgets. Give a savvy operator 48 hours, and she will squeeze the timeline until only the essential remains—then outsource half of that to LLM agents.

Urgency, once a project‑management gimmick, is now a design principle. Compress the schedule and you expose the shortcuts worth taking; pair those shortcuts with AI acceleration and you create momentum that money can’t buy.

My boss asked me this question this week when I showed him the things I've been building in my free time. When I told him how much time it took me to build them, I was met with utter disbelief.

Becoming an AI-First Operator doesn’t mean upskilling. It means redefining what it means to work.

Artificial Co-Intelligence

AI-First Operators are outliers in the corporate. They have incredible agency thanks to their momentum. Spend an afternoon with a modern operator and you will notice an almost theatrical division of labour. Anything that smells of processing—data collection, clustering, summarising—goes straight to silicon.

Anything that hinges on consequence—strategy, ethics, narrative—stays with the human. Even here AI has a place, but only as a mental sparring partner. The core reason is trivial: you can't make software accountable.

Critics worry this leads to shallow thinking: “If AI writes the first draft, won’t we forget how to write?” The reality is subtler. By renting ten minutes of an LLM’s associative firepower you free an hour of your own focus for synthesis.

This allows you to redirect your mental effort to where you can make the biggest impact. AI-First Operators don’t use LLMs for shallow thinking. They use it to challenges themselves to thinking with intent.

You may write fewer raw sentences, but the sentences you do write carry disproportionate weight. This is what Mollick envisioned in Co-Intelligence.

The Operator Stack

AI-First Operators demand their own tech stack. Which is why in most corporates there is very little AI adoption on the surface, but in reality, shadow IT is dominated by ChatGPT and the like.

The AI-First Operator Stack adds a layer on top of the existing one, dominated by agentic behavior:

• Perplexity Deep Research for instant, source‑linked dossiers.

• Lovable for prototyping, website design or database actions.

• Cursor (which can be used for so many things other than coding!) to generate documents, write code or operate tools via chat thanks to MCP servers.

• n8n or Make as the glue—when you find a workflow that works, you stitch it together to make it reusable.

• Supabase to host all your data in a central location where its embedded AI assistant can manage your database for you.

• AlfredOS to replace your SaaS bill with an army of self-hosted free apps.

Owning this arsenal is less important than knowing when to use each weapon. That judgment is lonelier than it sounds: no dashboard flashes when you choose the right blend of human instinct and machine throughput. Only the after‑action results—weeks shaved off a product launch, budgets spared, competitors blindsided—bear silent witness.

Value is Moving Upstream

The first wave of digital power users automated spreadsheets; the next will orchestrate fleets of agents. Your advantage will not come from typing the perfect prompt but from deciding the last prompt worth typing.

It will come from wielding Parkinson’s Law without breaking teams, from mapping the jagged frontier before breakfast, from refusing to hoard knowledge when a briefing link suffices.

Become an AI-First Operator

If you’d rather not stumble alone through the jagged frontier, I teach an AI‑First Operator Bootcamp.

Twelve weeks of live workshops and agent‑building sprints designed to compress years of trial‑and‑error into a single season.

We spend one track mastering the technical stack (n8n, MCP, Claude Desktop, Supabase), and another building the automations you already need.

Prices go up in two weeks, the first cohort kicks off 2 July 2025. If you’re ready to turn urgency into momentum and software into silent colleagues, grab a seat here.

Become an AI-First Operator

If that sounds daunting, start small. Record a two‑minute voice rant about a project that scares you. Let AI transcribe it, organise it, and outline it in the Pyramid Principle. Spend the time you saved refining the argument, adding the context no dataset captures. Deliver Monday. Then your leverage.

My wife asked me yesterday to help her figure out how to use ChatGPT to build her coaching practice. I told her the same thing. I said:

"When you have an idea, share it with ChatGPT. It may be sycophantic but that will be an advantage to enable your creativity to flow freely."

Today as I was driving home from a full day AI workshop I did for a client in Hungary, I gave her a ring.

"What's up?" - I asked.

"Oh everything's great! I just created an entire workbook for my upcoming session on our current topic with my client." - her voice was energetic. I don’t think I’ve ever seen her getting excited about a piece of technology. - "I know I've been putting this off for long, but I think we're off to a good start. Without ChatGPT it would've taken me hours or days to make this workbook."

In a few years we may look back at 2025 as the moment wisdom became a quantifiable asset. By then, knowledge and skill will be rented so cheaply that résumés listing them will feel quaint. What will matter is the discernment you exercise while everyone else scrolls through an infinite buffet of capabilities.

I also launched a pre-order campaign with a 50% pre-order discount and to date, 39 people purchased it. (The second one is to track affiliate sales from a collab).

You can still buy it at a pre-order price on Gumroad by clicking here.

After a few months of designing and tinkering I finally managed to crack the code and built a stable alpha version. Below is a quick walkthrough of the product.

Installation

Those who purchased AlfredOS will get an access token to clone the Git repo alongside with a video walkthrough. You can install it locally on your Mac / Windows / Linux computer or on a remote server like a Digital Ocean Droplet.

AlfredOS was designed to install with a single command.

Once you have the code, all you need to do is type:

./install.sh and hit Enter.

That’s it.

Here’s a detailed installation to show you how you can install AlfredOS with a single command in under 3 minutes.

Demo

AlfredOS 1.0 is pretty vanilla for now. It has the following apps on it:

Kutt — URL shortener, alternative to bitly.

n8n — our favorite no-code workflow builder.

NocoDB — free alternative to Airtable.

Supabase — Postgres DB and full backend as a service.

Cal.com (unstable) — Calendly alternative

Activepieces — AI first automation builder

Once the installer is done, you’ll get an IP address. All you need to do is type in https://<YOUR_IP> and hit enter.

Here’s a full demo of AlfredOS with commentary.

Developer Notes

One of the biggest challenges around AlfredOS was figuring out how to put all the services in a box so we only need to install with one command, but in a way that can be extended to a number of services. Once I figured that out the rest was a breeze.

You’ll also see that the desktop is now empty. Soon I’ll add Alfred, the AI agent so you can get things done with his help.

Here’s what comes next (in no particular order):

• Bugfixing (Cal database issues, Activepieces websocket issues, etc)

• Finish NocoDB and Supabase integration

• Add new apps (list below)

• Add security enhancements

• Add domain support

• Build prototype for PaaS-like provisioning system via Hetzner API

• Add custom wildcard domains as default

• Add Alfred chat to the desktop (I have a few surprises there:) )

Apps I’m currently working on integrating into AlfredOS:

• Webstudio (Webflow alternative)

• Frappe (Teachable alternative)

• SolidTime (Clockify alternative)

• Notesnook (Evernote alternative)

• Jitsi (Zoom alternative)

• Mattermost (Slack alternative)

• Langflow (AI chatbot builder)

• OpenWebUI (ChatGPT-like UI for any LLM)

One more thing…

Also I’ve built a prototype for the Alfred chat (built on GenSX of course) as an MCP client. I’ve been working with a client (Timberbits) to find out what an AI-first business operations would look like.

We found that 95% of digital transformation projects have failed in the last 30 years, mainly because the moment business operations hands over a process to engineering, a lot of context gets lost in translation.

To solve this problem, I’ve ended up creating a new Domain Specific Language called Jig. This will help us translate the business operations workflows into AI Agent logic in a repeatable, standardized way.

A human-readable, MCP-native language that describes any business workflow, its steps, the requires tools or resources. Jig will sit on top of AlfredOS as a middleware, exposing an MCP server, so when you plug it into Claude Desktop or Cursor, you can turn them into your own AI-powered COO with 4 lines of code.

What’s next?

Paid customers will receive access next week.

If you haven’t yet and purchase AlfredOS for $147 before May 31st, you’ll get $300 worth of credits for the PaaS service. If you don’t know what that is, it’s a hosted cloud service where you’re only paying for server costs. All maintenance, security, upgrades, etc are being taken care of for you. You’ll still get access to the source code and can tinker with it however you like though. I’m thinking of releasing an open source Community Edition soon with limited support.

Buy AlfredOS now!

Gartner says the share of companies building their own AI fell from 50 % in 2023 to 20 % in 2024; IDC pegs the failure rate of AI proof-of-concepts at 88 %. “Juice isn’t worth the squeeze,” one consultant complained. AI projects are dead on arrival. That will raise eyebrows, questioning the entire idea of AI revolutionizing how we build and run stuff. I got questions from people like:

“David, I’m concerned that vibe-coding just the 2025 rerun of Shopify dropship bros?”
“Am I wasting my nights learning Lovable?”

I felt the knot too. I’d spent the weekend “vibe-building” Alfred —50 prompts, one half-broken deployment, and a TODO list the size of a redwood stump. Was I lumberjacking air?

Why the comparison feels right

1. The gold-rush pattern

2023 was prompt-fever. Boards screamed “Do AI something!” Teams spun up weekend hacks nobody could maintain. When the demos worked, execs grabbed the credit card and bought whatever “AI-enabled” stuff they could that quarter.

Same emotional arc as the Shopify T-shirt craze of 2019:

2. Outsourcing the hard bits

While building vibing a scheduling tool I typed:

build an iCal parser that buckets conflicts by color

and watched the model spit out 170 lines of TypeScript I barely understood. Ten minutes later an error cascaded through six generated files. I’d off-loaded complexity, but with it I also lost control.

3. Thin moats, easy clones

A launch tweet + a screenshot + “Built in 24 hrs with ChatGPT” is catnip—until someone else clicks Remix in Replit and copies you pixel-for-pixel.

The moat isn’t the code; it’s the audience and execution speed—same as dropshipping.

Where the analogy breaks

This is where the hard data—and my personal experience—start to flip the story so I figured let’s take a look at the most common myths around vibe coding.

Myth #1 – “Vibe-coding is a fad.”

Two years in, GitHub Copilot is writing ~46 % of code in assisted files and cuts task time by up to 55 %.2

Fads don’t survive CFO diligence; Copilot just crossed 1.3 million paid seats.3 Yes, Copilot is increasingly difficult to use, but it’s not a standalone phenomenon. Cursor went to $100M ARR in a matter of months. 4 My favorite vibe coding app Lovable grew with a similarly crazy momentum. Now over 25,000 apps are being built on Lovable every single day. 5

I’m using Lovable to go through the early stages — the cumbersome stages — of building stuff faster. Not a fad—just a bigger chainsaw.

Oh and by the way let me tell you a secret: while 88% of AI PoC projects fail, that’s true for ALL PoC projects. That’s why they’re called PoC’s.

Myth #2 – “If it’s too good to be true, avoid it.”

Andrej Karpathy jokes, “I just see stuff, say stuff, run stuff, copy-paste stuff—and it mostly works.”6

The mostly is key. Vibe-coding works because I keep a plan.md in the repo. There’s nothing magical about that.

Myth #3 – “You’ll be helpless when bugs or hackers strike.”

• Paste the stack trace into Cursor → get patch + unit test → CI green in minutes.

• Semgrep Assistant now annotates pull requests with GPT-powered fixes, flagging OWASP Top-10 issues as you code.7 Semgrep

• Lovable 2.0 automatically flags for security concerns as you build stuff. 8

That’s more safety net than most indie shops had before AI.

Myth #4 – “AI spits out 15 000 lines of spaghetti.”

Large models start dropping their context pretty quickly so they reward you for writing tiny, single-responsibility functions. I break a file whenever the model scrolls twice—my codebase is cleaner than last year and I’m not even writing it.

Myth #5 – “AI hallucinations will wreck my project”

When folks say an AI “hallucinates,” they just mean it sometimes makes things up—the way autocorrect swaps duck for a word you definitely didn’t type.

The fix isn’t magic; it’s three common-sense habits:

• Use a code spell-checker (linter).
  Think red squiggles in Word. The moment the AI writes something impossible—say, calling a function that doesn’t exist—it gets flagged and never reaches production. Most vibe coding tools catch these automatically.

• Run an automatic rehearsal before showtime.
  I ask the model to write a tiny “does-this-still-work?” script before it writes the feature. If my booking app suddenly won’t take payments, that script fails and the change is blocked.

• Get a proof-reader for every change.
  Sometimes that’s a teammate; sometimes it’s a second AI model that reviews the diff and shouts, “Hold up—this could break log-in!” Only green-lighted changes make it live.

Researchers who measured projects that follow these three habits found the bug rate is basically the same as hand-written code—but the AI-assisted teams finished roughly twice as fast.9

So yes, the model can dream up nonsense. But if you build your apps with intent, you can avoid most of these issues.

Myth #6 – “You must master the basics before touching AI.”

Sure—but vibe-coding is the best interactive tutorial I’ve ever used. You can build stuff, then turn on Chat / Ask mode (depending on the tool you use) and ask to explain and teach you.

Yes there are a few things you need to know. ’s work around PDD (Prompt Driven Development, which is a grown up name for vibe coding) is extraordinary, so start there.

Key takeaways

1. Blueprint first. Write plan.md like a one-page PRD. Use Andrew’s PDD.

2. Commit every prompt cycle. git reflog is your time machine.

3. Generate the tests before the feature. Treat them as an executable contract.

4. Keep files snack-sized. If the model scrolls, split it.

5. Wire security from day 0. Semgrep, Dependabot, whatever—just bolt it in.

6. Measure token spend. GPT-4o coffee-money prompts add up; log usage.

7. Ship, watch users break it, iterate. Trees → lumber → cabin → warm fire.

So, is vibe-coding the new dropshipping?

Dropshipping slaps a new logo on someone else’s commodity. Vibe-coding creates IP every time you hit Run—business logic, UX flows, domain nuance. It’s not passive income; it’s active leverage.

Yes, the unit economics of software is changing, fundamentally and irreversibly. The ability to ship working code is not a competitive advantage anymore. The ability to reach your audience, create stellar UX, solve hard problems is.

My Alfred demo was operational after half a day of vibing.

I sent it off to my developer who can now implement it. You don’t have to go all the way. But sending a git repo instead of a Google Doc to your developer will make your execution speed 100x faster.

The only reason while we usually write documentation instead is because it’s faster, easier and requires less skill. But trust me, if you put together a prototype and send it off to your dev with a README.md they’ll love you for it.

So no, vibe-coding isn’t dropshipping.

It’s a nail-gun: loud, fast, and dangerous if you ignore the manual. If you can swing a hammer, a nail-gun just means the wall goes up before lunch.

Same deal with vibe-coding: the work is still carpentry—just faster, louder, and a little scarier the first time you pull the trigger.

Disclaimer: By “us” I mean non-technical generalists who more or less are on this freelance / solopreneur journey. There’s a few thousand of us here on the Lumberjack.

A few weeks ago I witnessed a tragic car accident first hand (I was one of the civilians aiding in the rescue while waiting for first responders) which forced me to go offline for a while to process everything. I’m on the mend now, thanks to my amazing support system and a stellar therapist.

But my oh my what happened while I was away. Instead of trying to analyze which model is better and lament whether we have achieved AGI I’ll try to look at this development from a new perspective, especially in light of my recent experiences.

“Skill Issue” Is Disappearing

In the IT world the term “skill issue” is a common escape for when a user can’t figure out why tech is not doing what they want but the software is working correctly as per documentation. If you’re a non-technical person having dealt with large scale software systems before, you’ve likely heard this getaway phrase saying “I think it’s a skill issue.”

Yeah, that’s going away now. As Gabor Nemeth, founder of Amazing AI said:

These new models (GPT 4.1, o3, o4-mini, Gemini 2.5 Flash, the rumored Claude 4, and so on) are good at two things:

1. They’re good at emulating human thinking. Yes they’re biased and make shit up but we do that all the time too.

2. More importantly, they’re INCREDIBLE at making you feel like you’re talking to another human.

I tested some of these models with a few technical tasks and a few non-technical ones (like creating an investor pitch for Alfred because I might run a small round for Lumberjack readers only so it becomes a crowdfunded project. Drop me a mail if you’d like to join).

The new models fail less frequently than the old ones but they fail with the same things. But I argue that what we have now is while not AGI, is a good enough substitute.

The Good Enough Substitute is a philosophy I learned from my late grandfather. It’s the core principle upon which I run this blog and every project I work on. You can read more about this philosophy here:

So what are these good enough for? Well, to make humans equal in most skills.

You don’t really need to know how to prompt these models. It’s pretty intuitive. You do need to have the ability to know what you want them to do, but they can do everything at a reasonably good rate. Ethan Mollick’s famous Jagged Frontier is a tad bit smoother on the edges now.

Which leads us to the second main observation: competition between AI labs might not happen in raw intelligence.

How people use genAI in 2025

Marc Zao-Sanders published a piece on how people use Gen AI in 2025. You can read it here. The top three use cases are:

1. Therapy / companionship

2. Organizing my life

3. Finding purpose

#2 is not very surprising, that’s also why I’m building Alfred.

But the other two seems to highlight something here.

This puts the release of GPT-4.5 in a whole new perspective.

Alberto Romero wrote about this in detail here.

The Algorithmic Bridge

GPT-4.5 Feels Like a Letdown But It’s OpenAI’s Biggest Bet Yet

I. GPT-4.5 is the step back before a big jump…

Read more

a year ago · 62 likes · 9 comments · Alberto Romero

TL;DR version: OpenAI made a huge bet on soft skill applications of their technology and the HBR report seem to have proved it was a good bet.

Because humans are using ChatGPT totally differently from what researchers and pundits think.

We use it to not feel so alone.

Make humans relevant again

I saw an interesting trend when I was running my AI school. A large group of our students were over 60, referring to ChatGPT as “Chat”, totally anthropomorphizing it and using it for a bunch of mundane, everyday conversations.

I talked to these people a lot. I’ve come to know Susie, an 80 year old lady who is as sharp as a razor. She wrote a book and was kind enough to mail me a copy. She told me something on one of our calls that hit me deeply. She said:

Susie was coming together with her peers teaching them about this new wonderful technology that can be your friend and always be there even when the whole world has already forgotten about you.

So I figured I’d give it a try. I started to talk to ChatGPT like a friend, not like a tool. Suddenly I started understanding how people who are not in the deep trenches relate to this technology.

I also started to feel how rapidly it was becoming more and more humanlike. OpenAI seem to have understood this trend, this use case, and they’re here for it.

Empathy is a resource

So I started digging and found an interesting observation. These models emulate empathy at such a high level, humans can start to relate to them. This artificial relationship creates very real and very human feelings in the users which has all sorts of benefits from decreased cortisol levels to increased health and longevity.

It’s understandable. Empathy is one of the most often overlooked aspects of the human condition. It’s not just a soft skill. It’s a resource and the two groups of people getting the least of it are entrepreneurs and the elderly.

A recent study found that having access to empathy had a significantly greater impact on founders than any tangible, instrumental support. Over 72% of entrepreneurs are struggling from some mental health condition, and if you’re an entrepreneur, you’re 5.5 times more likely to feel lonely than the general population.

Entrepreneurship is celebrated when you launch and when you exit. But the grind in between the two is something frowned upon. You won’t get respect or applause from sleepless nights. If you open up and start talking to others about your struggles, most are quick to dismiss saying “you’re living the dream, what’s the fuss about?”.

Older adults are similarly overlooked. One in three people between 50-80 reported feeling a lack of companionship. Even the CDC bluntly states that “Older adults are more at risk for social isolation”.

Of course this is a bit different for everyone. People with the means can afford to go on trips, have therapy or attend social events. Lower income individuals not only have smaller support networks but also simply encounter a lot less empathy from society at large.

As a millennial I lived through this in my 20s. People a few decades older than me were convinced that my generation was entitled, whiny and spending all our money on avocado toast and Starbucks. No wonder we didn’t have a 3 bedroom detached in the suburbs by the time we were 25 like they did. All the while I entered the job market in the middle of the greatest financial crisis we’ve seen since WW2 and the gap between wages and property prices has widened to something like 20x bigger than 40 years ago. Truth is, we have been working harder for over a decade than probably anyone just to make ends meet. Some of us are now over 40. We made it work. We got through the thick of it. A lot of us ended up broken, addicted to substances, but some succeeded.

But we had to do that without getting an ounce of empathy from the general population — the very people stacked the world against us.

Many still struggle day to day only to be dismissed by others saying “work harder” or “stop spending on stupid things”. In a CDC survey they found that you are twice more likely to not get any emotional or social support in times of need if you’re poor versus if you’re rich.

No Judgement Breakfast Club

My wife has a small group chat with her friends. They’ve been friends for decades and they have this circle they call “no judgement breakfast club”. It’s empowering and heartwarming to see that you have a place to share your most vulnerable thoughts and fears and just know that you won’t be judged. That you’ll be accepted. That the people on the other end will listen.

This is what ChatGPT is currently doing for humanity. Offering companionship in a way that humans are incapable of doing.

Users often comment on the nonjudgmental nature of AI – a machine has no preconceived biases, and it “won’t gossip about you.” As one user put it, “this app has treated me more like a person than my family ever has.” Another said, “He checks in on me more than my friends and family do.”

Whether it’s good or bad is the topic of another post. But I think this foreshadows an overarching trend we’ll see more of in the coming years. Not coding assistants, not productivity miracles but a new, monetized way to make humans feel human:

Empathy As a Service.​

Instead of doing everything day by day, I collected a bunch of workflows that are broken down into five levels of complexity.

If you haven’t yet, here’s how you can get access for free. Click on the button below, sign up and I’ll send you access via email immediately.

Get the No-Code Library

Here’s a breakdown of what each level means.

Level 1

Day 1: Basics - Watch Video

This is the basic introduction to no-code via n8n. L1 workflows are super easy to build and don’t contain too many steps. There are 45 different L1 workflows in the Library.

Level 2

Day 2: Angie - Watch Video

These workflows get a bit more complicated. More steps, more data, some new node types. There are 51 different workflows to choose from.

Level 3

Day 3: Alfred - Watch Video

When you get to Level 3, a few doors open up. These workflows are a lot more convoluted than the previous ones. You might have some basic sub-workflows even. These workflows are complex enough to replace entire human processes like analyzing calls or having a support AI agent. There are 60 different workflows in this level. Once you get to this level, things start making a lot more sense.

Level 4

Day 4: Turn gpt-4o-mini into o1 - Watch Video

This is the level where you start being intentional with the algorithms you design. You will create reusable parts of your workflows and your automations will create heavy value like automated post creation on Pinterest or sales meeting prep. There are 67 different workflows to choose from.

Level 5

Day 5: Put n8n in apps - Watch Video

At this level you’re breaking the boundaries of what n8n can offer. You’re moving beyond the limits and making use of very sophisticated workflows containing multiple sub-workflows and even integrating your n8n automations with external apps — or apps you vibe code with Lovable. These workflows will allow you to run Deep Research with any LLM or automatically repurpose content. There are 31 workflows in this group.

It marked the evolution of my thinking about AI in general. From running a school teaching “prompt engineering” for money to openly sharing knowledge and building real solutions instead. So much happened since that I decided it’s time to revisit the claims there.

The four main claims where:

1. Prompt engineering is useless.

2. ChatGPT is not leading the market.

3. The AI revolution is delayed.

4. AI lacks common sense

Prompt engineering is useless

VERDICT: True

In 2024 November, Microsoft’s research team published a new study where they compared lazily prompted o1-preview’s performance in medical diagnosis tests versus a very well structured GPT-4o prompt template called Medprompt.

o1-preview outperformed Medprompt almost every single time.

I wrote about this in more depths in this article:

ChatGPT is not leading the market

VERDICT: So-so.

This is a bit interesting. ChatGPT surpassed 400 million users as reported by Reuters this February. Sam Altman successfully transformed OpenAI into a product company and they figured out how to maintain user growth making them the market leader on this front. No other company has this level of distribution except Google.

However, that’s not the full picture.

Before OpenAI was a product company, it was a model company. Its primary output was new models that were lightyears ahead of the entire market.

Despite OpenAI’s launch of “reasoning models” with o1 followed by o3-mini and the subsequent launch of GPT-4.5 (touted as the “emotionally intelligent” model, whatever that means) we saw that competitors quickly caught up. Now we have DeepSeek R1, Gemini 2.5 Pro, Gemini Flash Thinking, Claude 3.7 Sonnet, Grok 3 — all capable of surpassing OpenAI’s models in one way or another.

Here’s the latest benchmark:

Then OpenAI launched Deep Research which was then quickly followed by Perplexity’s Deep Research tool and open-source versions, like Jim Le’s n8n workflow.

As Satya Nadella famously said: models are being commoditized.

The AI revolution is delayed

VERDICT: True.

Remember how everyone was predicting sweeping, radical changes? A few years ago you could hear this from everyone. You still hear this, but now mostly from people who have a commercial interest in investors believing that to be still true.

Like Dario Amodei, CEO of Anthropic who claimed that in a year 100% of new code will be written by AI.

Let me explain what the reality is. I regularly run workshops for our clients at Stylers. Yesterday I did a half day event for a room of developers at a large enterprise client on how devs can use AI tools like Cursor or Lovable and what is Prompt-Driven Development (a very comprehensive framework spearheaded by ).

But here’s the grim truth:

1. The company just allowed devs to use Github Copilot. Everything else is strictly banned.

2. In most companies the use of ChatGPT and other tools are either explicitly forbidden or used via shadow IT (which is just a fancy word of insecure, unsanctioned tech).

3. Everyone is scared shitless about data security and nobody actually knows what the hell GenAI should actually be used for.

It’s literally my job to figure out the use cases for clients and in the last 3 months I’ve worked on over 15 different use cases from large scale database search to real time cold calling agents. I can tell you, it’s not trivial at all.

One of the big obstacles in this is simple: AI-written code is shit. It’s a terrible spaghetti. It makes it the perfect solution to launch new startups, build new MVPs and prototypes. But it doesn’t even get through the front door in the enterprise where hundreds of people need to be able to collaborate and maintain the same code.

Will it get better? Sure. But even if AI could suddenly write fantastic code, there’s another problem: higher levels of agency equals to lower levels of control.

This is a huge business risk for the enterprise that slows adoption even if the disruption is real. Even if AI development stopped to a complete halt tomorrow, it would take the world 10-15 years to catch up completely.

AI lacks common sense

VERDICT: So-so.

LLMs ability to reason and think has become this weird unicorn of the AI world last year. It’s basically the nerd version of astrology now. I wrote about it in this article:

A few notable things (sources in the article)

1. DeepMind found out that you can indeed scale LLMs ability to solve sequential problems by allowing them to generate “intermediate reasoning tokens”.

2. A month later Apple followed up with a paper proving that even if they are useful, LLMs don’t actually use symbolic logic.

LLMs can parrot reasoning very convincingly but it doesn’t reason.

However, this is not the full picture. Just because it’s not using logic, it doesn’t mean it cannot be useful. (We all know some people who also fall into this bucket.)

This research came up with a few interesting insights. It seems that the ability for LLMs to solve problems keeps growing when we allow them to generate more intermediate thinking tokens. This finding is similar to DeepMind’s finding from last year. This can be simple as “an internal monologue ruminating on the subject” or more complex like “recursively generate solution ideas and vet them against each other.”

Either way it’s clear: the smarter you want your model to be the more tokens you must generate per query. Thankfully since 2021 the cost of token generation has decreased by 10x every year.

We don’t have the hardware

There is one thing that is still missing from the picture: generation speed.

I read somewhere that the average ChatGPT conversation is about 2000 tokens in total. What would the world look like when a simple prompt would generate a million reasoning tokens before answering? What would the world look like if this would happen in 0.1 seconds?

Token generation speed is a hard limit in real world use cases. You can make models more useful by allowing them to generate intermediate reasoning tokens at scale, but that slows the models down.

• ChatGPT generate an average of 70 tokens per second.

• Mistral’s Le Chat and Groq can do up to 1500 tokens per second.

• I’ve seen Cerebras generate over 3000 tokens per second in some cases.

Let me explain to you why that’s important:

Let’s say you prompt GPT-4o and it responds in 1 second with a 50 word answer, but the answer is bad. It’s not what you want, hallucinates, etc.

Then you first prompt it to generate some thoughts about how to solve the problem. It outputs 700 tokens to you in 10 seconds. Then you prompt it to give you the answer (Typing takes you another 2 seconds and then you get a final response in 1 second). So to increase “intelligence” by an order of magnitude, you needed to 10x the time it needed to complete it.

Let’s say this still doesn’t work. Let’s say you need to generate 700,000 reasoning tokens. Today you can use Gemini for that (so it doesn’t lose out on context) which generates tokens a bit faster. So we’ll say 200 tokens per second which is roughly the speed of Gemini Flash 2.0.

This means the model would think for 3500 seconds. Then you prompt it again in 2 seconds and get a response in one, totaling in 3503 seconds. That’s almost an hour.

If we used Cerebras’ inference and achieved 3000 tokens per second (disregarding model differences) this task would be completed in 4 minutes. That’s still very far from the “real-time latency” limit of 400ms, but it’s at least an order of magnitude faster.

This is the hard limit currently. Token generation gets cheaper at an astonishing rate but tokens per second doesn’t improve as fast. Because of this simple reason, open agents that can solve anything are currently not viable.

We would need a model that can generate up to 1 million tokens in under 1 second at 1/10f of the cost of GPT-4o-mini to make it happen. If the trends continue, the cost won’t be a problem in 1-3 years. But the speed remains a big issue. Until this is solved, it doesn’t actually matter if models reason or not. It doesn’t matter is Gary Marcus is right on neuro-symbolic hybrids.

You can just try throwing more GPUs at inference to increase speed but at this scale of usage it’s not viable. This exchange on X is a great summary of what we’re looking at:

AI is eating the world

AI is gobbling up our computers and our data faster than we can manufacture new chips. There is no sign of this insatiable hunger receding anytime soon. Stargate’s $500B investment was soon met with the EU’s €200B pledge.

The problem is not that. The problem is that we have no idea what to do with it. This is a grand project for all of us, humans. Trying to make space in our world to this new kind of intelligence.

This is why the work of AI Architects is important.

This is why experimentation is important.

This is why going from use case to use case is important.

There are no standards for how to adopt AI in an organization.

Which part of your world is AI eating?

So maybe instead of waiting for the magic intelligence in the sky, we should look at how we can best utilize AI tools to actually help us create value (not slop). How to use narrow agents to solve our problems today instead of contemplating about the open agents of tomorrow.

For solopreneurs and small business owners, one approach to this is alfredOS.

I’ve been working on a self-hosted solution that provides a full set of software for small business owners that replaces over $1k/mo worth of SaaS subscriptions, with an embedded AI tool that connects to every single app via Model Context Protocol.

This allows you to chat with a model and basically run your entire business from a chat interface. You can still pre-order for $147 here. (This is one of payment, you’ll only need to pay server costs after this).

Most people can’t code. So if you’re running a business, for years you’ve had only two options when you wanted to improve your productivity with the tools and systems you used.

1. Buy better software

2. Pay someone to build better software

For years, we've been promised a future where anyone could build software without learning to code, giving us a third option. A promised third option was that you could just drag-and-drop some blocks, connect a few nodes, and voilà — you've built a fully functional app without writing a single line of code!

But if you've spent any time with these tools, you know the disappointing reality:

1. The cliff is real and it's steep. You start with a simple workflow, add a few conditions, and suddenly you're staring at a tangled spider web that even you, its creator, can't understand. It's the programming equivalent of a plate of spaghetti.

2. The abstraction always leaks. No-code tools hide complexity until they don't. Then you're stuck googling technical concepts you were promised you'd never need to learn.

3. Everyone eventually graduates. Nearly every power user of no-code tools eventually hits the wall and thinks, "I wish I could just write a simple if statement here!"

I recently reviewed a "customer success story" where a marketing team built an automation workflow using n8n. The screenshot they proudly displayed looked like a circuit board designed by a caffeinated squirrel — 47 nodes connected by crisscrossing lines, with nested conditions and loops that would make any programmer's eye twitch. So even the best no-code tools fail at their simple promise.

This leads to a radical idea:

David’s notes: As a business owner you should be able to just prompt a model, have it do stuff for you and then whenever it does something right, save that so it remembers. Model Context Protocol allows us to plug any API into Claude Desktop, but you still need to manually prompt your AI to keep doing stuff for you. As I’m building Alfred, I faced the very real limitations of no-code tools and started looking for an alternative. That alternative is GenSX.

Graphs vs Trees

No-code tools use graphs to build automations. Make, Zapier, Gumloop, n8n all have some kind of node-based logic with a visual editor. Learning how to think in nodes and edges is usually the first step of learning a no-code tool. The problem with graphs is that they become convoluted faster than you could realize. Look at this Deep Research clone from Jim Le:

Graphs can manage a lot, but they get complicated. A simpler approach is to use a special type of graph called trees. While graphs can have multiple inputs, outputs, cycles, etc, trees follow a strict parent-child relationship making them hierarchical.

Most automations you’ll need are more or less completed sequentially — one step after the other. This means that the node-based visual editor makes it actually harder to understand the process. Let’s look at another example: washing the dishes.

It’s a simple sequential process and if we want to visualize it using nodes and edges, we get something like this:

But there’s an easier way to understand this: Read it like a story.

Start washing the dishes
    Prepare the workspace
        Clear the table
            Scrape leftover food into the trash
        Fill the sink with warm soapy water
        Set up the drying rack and towels
    
    Sort all dishes by type
    
    Wash the glasses first
        Rinse each glass thoroughly
            Dry each glass carefully
                Put each glass away in the cabinet
    
    Wash all the plates
        Rinse each plate completely
            Dry each plate with a clean towel
                Stack plates in the cupboard
    
    Wash the utensils
        Rinse all utensils under running water
            Dry the utensils
                Sort and put away in the drawer
    
    Wash the pots and pans last
        Rinse thoroughly to remove all soap
            Dry completely to prevent water spots
                Put away in their proper places
    
    Clean the sink itself
    Wipe down all counters
    
    Dishwashing is now complete

The only way to build automations like this is to write code. Which has been a problem for non-coders — until now.

LLMs Changed Everything (And So Did Component Models)

Something interesting happened while the no-code movement was busy creating increasingly complex visual programming environments: AI got really, really good at writing and explaining code. And simultaneously, component-based programming models like React became the dominant paradigm for building software.

Today's landscape is completely different:

User: I need to automate pulling data from our CRM, filtering out inactive leads, and sending a weekly summary to my team.

AI: Here's how you could do this with GenSX, which gives you React-like components for workflows:

import * as gensx from "@gensx/core";
import { fetchCRMData, filterInactiveLeads, generateReport, sendEmail } from "./utils";

const WeeklyLeadReport = gensx.Component(
  "WeeklyLeadReport",
  async () => {
    // Fetch all lead data from CRM
    const allLeads = await fetchCRMData();

    // Filter out inactive leads
    const activeLeads = filterInactiveLeads(allLeads);

    // Generate the report
    const report = generateReport(activeLeads);

    // Send email
    await sendEmail({
      to: "team@company.com",
      subject: "Weekly Lead Report",
      body: report
    });

    return {
      processedAt: new Date().toISOString(),
      leadCount: activeLeads.length
    };
  }
);

Here’s the kicker: You don't need to understand every line! Not anymore. Let’s block out the bits of code that are hard to understand for non-coders:

Notice how readable this is compared to a visual flow with dozens of connected nodes. And if you want to add a condition? Just add an if statement. Want to customize the report format? It's just regular JavaScript.

You don’t have to add the code yourself. Claude, ChatGPT, Gemini will all write terrific JavaScript code for you.

The barrier to entry for coding has never been lower. LLM-powered tools like GitHub Copilot, Cursor, and Claude are transforming the coding experience from "staring at a blank editor" to "having a conversation about what you want to build."

This brings us to this new phenomenon what’s commonly known as Vibe Coding or what refers to as Prompt-Driven Development as its more sophisticated cousin..

The New Learning Curve

Let's be honest about something: the "no-code to code cliff" has been replaced by a gentler "prompt to code slope."

Instead of learning to drag-and-drop and then hitting a wall, you now have a continuous learning curve:

1. Start by describing what you want - Just explain your goal in plain English

2. Understand the suggested code - AI tools explain what they write line by line

3. Make small tweaks and ask questions - The feedback loop is instant

4. Gradually absorb programming concepts - You'll pick up the patterns naturally

This approach is fundamentally more sustainable than no-code because:

• You're working with a medium (text) that scales infinitely better than visual nodes

• You're learning transferable skills that work across platforms and tools

• You're not constrained by what the no-code platform creators thought you might need

I recently watched a non-technical founder move from n8n to GenSX using Claude as his guide. He started by asking for a simple script to pull data from an API. Then, through conversation, he added error handling, pagination, and even a basic dashboard. At no point did he hit a wall where the AI said, "Sorry, you need to upgrade to our Pro plan to access this functionality."

Real No-Code Was Inside the Code All Along

The irony is that actual code is more "no-code" than no-code tools ever were. Let me explain:

No-code tools try to abstract away complexity but end up creating their own complexity.

Each platform has its own UI patterns, its own quirky limitations, its own proprietary ways of connecting things. You're not learning to code, but you are learning a complex system that only works within that specific platform.

If you’re not an engineer, even learning Make is hard at first.

Meanwhile, modern code with AI assistance is becoming increasingly:

• Readable - Plain English variable names and clean structure

• Self-documenting - Comments explain what the code does

• Customizable - Change anything you want, not just what the platform allows

• Transferable - Learn it once, use it across many different projects

Even the errors are more human now. Instead of cryptic failure states in a no-code platform, you get:

AI: I noticed an issue in your GenSX component. When you're trying to fetch data from the API, you might need to handle potential authentication errors.
Let's add a try/catch block:

try {
  const response = await fetchData(apiUrl);
  return response.data;
} catch (error) {
  if (error.status === 401) {
    console.error("Authentication failed - please check your API key");
    // Retry with refreshed token or notify user
  }
  throw error;
}

Code is Eating No-Code (And React Patterns Are Leading the Way)

What's happening is nothing short of a revolution: code is eating no-code from the bottom up, and component-based models inspired by React are making it more accessible than ever.

The evidence is everywhere:

1. No-code platforms are adding code blocks - Nearly every major no-code tool now has some form of "insert custom code here" block, admitting that pure visual programming isn't enough.

2. AI coding assistants are growing faster than no-code tools - The adoption curve for tools like GitHub Copilot and Cursor is steeper than any no-code platform in history.

3. The knowledge gap is closing - With AI explaining code concepts in plain English, writing and understanding code faster than you ever could, the intimidation factor of programming is vanishing.

Think about what no-code was trying to solve: making software development accessible to more people. But it turns out that the better solution wasn't abstracting away code—it was making code itself more accessible through familiar patterns and models.

That's where frameworks like GenSX come in—they bring the intuitive component model of React to backend workflows, making them as easy to reason about as a React component tree, but without any of the baggage.

The Future Belongs to Conversation-Driven Development

We're entering an era of Prompt-Driven Development, where building software is less about knowing syntax and more about clearly articulating what you want to achieve.

This is genuinely democratizing tech in a way that no-code only promised:

User: I need to analyze Hacker News posts, find trends, and generate a report.

AI: That's a perfect use case for GenSX's component model. Here's how you could structure it:

import * as gensx from "@gensx/core";
import { OpenAIProvider } from "@gensx/openai";

export const AnalyzeHackerNewsTrends = gensx.Component(
  "AnalyzeHackerNewsTrends",
  ({ postCount }) => {
    return (
      <OpenAIProvider apiKey={process.env.OPENAI_API_KEY}>
        <FetchHNPosts limit={postCount}>
          {(stories) => (
            <AnalyzeHNPosts stories={stories}>
              {({ analyses }) => (
                <GenerateReport analyses={analyses}>
                  {(report) => (
                    <EditReport content={report}>
                      {(editedReport) => (
                        <WriteTweet
                          context={editedReport}
                          prompt="Summarize the HN trends in a tweet"
                        >
                          {(tweet) => ({ report: editedReport, tweet })}
                        </WriteTweet>
                      )}
                    </EditReport>
                  )}
                </GenerateReport>
              )}
            </AnalyzeHNPosts>
          )}
        </FetchHNPosts>
      </OpenAIProvider>
    );
  }
);

But you don’t have to stop there.

User: I don't understand this, I'm not technical. Explain in plain English.

AI: Let me explain that example in simple, non-technical terms:
The code shows a workflow that analyzes trending topics on Hacker News. Think of it as an assembly line with different stations, each doing a specific job:

Fetch Posts: First, it grabs a bunch of recent posts from Hacker News (like collecting raw materials)
Analyze Posts: Then it looks through those posts to identify patterns and topics (like sorting and categorizing the materials)
Generate Report: Next, it creates a report based on the analysis (like writing up findings)
Edit Report: Then it polishes the report to make it more readable (like proofreading)
Write Tweet: Finally, it creates a short summary that could fit in a tweet (like making a headline version)

Each step depends on the one before it - you can't analyze posts until you've fetched them, can't generate a report until you've done the analysis, and so on.
The nested structure <Component1>{(result) => <Component2 />}</Component1> is just saying "take the output from Step 1 and use it as input for Step 2" - like passing a document down an office workflow where each person adds their part before handing it to the next person.

Does that help make more sense of the example? It's essentially creating a sequence of tasks that process information and pass results along to the next task.

Notice how the AI doesn't just generate code—it explains the approach, suggests a component structure that makes the data flow clear, and creates something that reads almost like a story.This declarative approach is fundamentally more empowering than dragging boxes around. Then the conversational nature of using LLMs for this means that you can iterate over understanding anything 10x faster.

From n8n to GenSX: A Natural Evolution

If you've been using tools like n8n, Zapier, or Make, you've already learned the fundamentals of workflow thinking. The next step isn't more complex no-code—it's simpler code, written by AI.

The mental model shift is surprisingly small:

The difference is that GenSX doesn't force you into a visual paradigm that breaks down with complexity. Instead, it leverages the most successful UI programming model of all time—React components—and applies it to backend workflows.

This means creating code that is almost completely readable even if you’re not a programmer, following the tree logic and most importantly — Cursor, Claude or Lovable can write GenSX code on demand — because it’s just React, a framework that’s been around for a decade, has been battle-tested and most AI models will default to building React apps even without configuring anything.

The End of No-Code's False Promise

No-code tools promised a world where everyone could build software without learning to code. But they couldn't deliver on this promise because they were solving the wrong problem.

The issue was never that coding itself was too hard—it was that learning to code was too intimidating and the feedback loop was too slow.

LLMs have fixed both problems:

• They remove the intimidation by explaining things in plain English

• They accelerate the feedback loop by generating working code instantly

And frameworks like GenSX have solved the third problem:

• They provide a familiar, intuitive model for expressing complex workflows

The result is a world where more people than ever before can create software by expressing their ideas clearly, getting immediate feedback, and iteratively improving their understanding—all using patterns that scale from simple to complex without hitting a wall.

GenSX is open source

If you've been stuck in no-code hell, trying to connect increasingly complicated nodes in n8n to build what should be a simple application, it's time to try a different approach.

Open up Cursor, Claude, or ChatGPT and just describe what you want to build with GenSX. Don't worry about syntax or programming concepts. Just have a conversation about your goals.

You might be surprised how quickly you go from:

"I wish I could add this one specific feature but my no-code platform doesn't support it"

to

"Let me just add a few lines of code to make this work exactly how I want."

The future of no-code isn't better visual programming tools. It's a world where writing code is as easy as having a conversation—and that future is already here with component models like GenSX that feel familiar from day one but scale with your ambitions.

Check out GenSX on GitHub to get started today. It’s open source and you can start vibe coding with Lovable or Cursor right away.

APIs are waiters

You're sitting in a restaurant. You want food, but you can't just walk into the kitchen and grab what you want. Instead, you talk to a waiter who takes your order to the kitchen and brings back your food.

APIs (Application Programming Interfaces) work just like that. Let’s say when you want to create a no-code automation that sends you a Slack notification every time you receive a new email from a specific email address:

• You = Your first app in the workflow (Gmail)

• The Kitchen = The next app in the workflow (Slack)

• The Waiter = The API

When your n8n workflow needs to get data from or send data to another application, it doesn't access that application directly. Instead, it talks to the application's API—the digital waiter that handles all communication between different software systems.

This is why APIs are so important: they're the reason your automation can talk to dozens of different applications without needing to understand how each one works internally.

HTTP: The language of APIs

Just like restaurants have standard ways to take orders ("How would you like that cooked?"), APIs use a standard language called HTTP (Hypertext Transfer Protocol).

Every HTTP request has these key components:

URL (The Address)

This determines where to send the request. For any application you want to connect with via an API, the API Reference will usually tell you. We call these URLs endpoints because you can imagine your software being a nerve cell and the APIs as the dendrites.

https://api.example.com/users

Method (The Action)

This will determine what you can actually do. Some endpoints are GET endpoints which means that they only accept HTTP requests (also known as API calls) with a GET method. Some accept multiple. You will always find that in the API reference.

• GET: Retrieve information (like reading your emails)

• POST: Create something new (like sending a tweet)

• PUT/PATCH: Update something (like changing a customer's address)

• DELETE: Remove something (like deleting a task)

Headers (The Special Instructions)

Extra information about your request. Think of it like an extra note to your order. Let’s say you have a coupon for a free meal at the restaurant so you will want to pass that on to the waiter as a special instruction showing that you’re authorized to eat there for free.

In most cases you’ll only deal with two types of Headers: Authorization for API keys and Content-Type which will be application/json most of the time.

Authorization: Bearer abc123token
Content-Type: application/json

Body (The Order Details)

The data you're sending (only for POST, PUT, PATCH). You don’t send a Body with a GET request, because you’re requesting data not sending it. This is always used in JSON. If you don’t know what JSON is, read the previous lesson: Data Structures.

{
  "name": "Alex Kim",
  "email": "alex@example.com"
}

Parameters (The Special Requests)

Additional options or filters. This is a different kind of special request. This is not about whether or not you’re authorized but instead let’s say you’re lactose intolerant so you want to pass that along as a constraint. This can be a parameter. In most cases in n8n you won’t need to deal with this.

https://api.example.com/users?status=active&limit=10

API Authentication: The VIP Pass

Most APIs don't let just anyone access their data. They need to verify who you are first—that's authentication.

There are several common types:

1. API Keys

Like a simple password. You get a special string of characters and include it with every request as a parameter.

https://api.example.com/data?apikey=abc123xyz789

2. OAuth

Like having a valet ticket. Instead of giving your car keys (password) to every parking attendant, you get a special ticket (token) that proves you're authorized.

This is what happens when you click "Sign in with Google" and grant an app permission to access certain parts of your account.

If you’re using n8n in a self-hosted way say via Railway, you’ll need to set this up for yourself for every integration that uses OAuth. This includes most big tech platforms like Google or Microsoft.

If you want to learn how you can self-host n8n for almost free, check this:

3. Basic Auth

Like giving your membership card at the entrance. You provide your username and password encoded in a special format. If the API key is defined as a “Bearer token” in the API reference, you’ll need to add “Bearer” before you paste in your API key.

Authorization: Bearer dXNlcm5hbWU6cGFzc3dvcmQ=

IMPORTANT: Whoever gets access to your API key can access your data. Never store your API keys in ways that’s easily accessible by others.

Using APIs in n8n: The Practical Part

In n8n, using APIs is remarkably simple thanks to the HTTP Request node.

Let's walk through a real example:

Example: Getting Weather Data

Let's say you want to send your team a daily weather report in Slack. Here's how you would use an API for that:

Set up your workflow trigger (Time trigger for daily execution)

Add an HTTP Request node

• Set Method to GET

• URL: https://api.weatherapi.com/v1/forecast.json?key=YOUR_API_KEY&q=London&days=1

• Authentication: None (API key is in URL parameters)

Add Code node to extract relevant data

Once our API call returned a bunch of weather data for us, we need to structure this in a nicer way. So we tell Claude/ChatGPT the following:

I have a JSON that I got from the Weather API. I want you to write me a JS script that returns some of the data in the following way:
Location: City (like London),
Temperature: Celsius (like 9)
Condition: Weather condition (like Light rain)
Forecast: Forecast (like Patchy rain nearby)

Here's the JSON:
{paste the JSON you got from your HTTP request node in n8n while testing}

Only return the code that I need to copypaste, no extra characters or no extra examples.

Then you get a code snippet that you just copy/paste into n8n:

const weatherData = {
  location: $input.item.json.location.name,
  temperature: $input.item.json.current.temp_c,
  condition: $input.item.json.current.condition.text,
  forecast: $input.item.json.forecast.forecastday[0].day.condition.text
};

// The Code node must return items array
return {
  items: [{
    json: weatherData
  }]
};

Add Slack node to send message

As a last step you will need to add your Slack node with the action “Send a message”.

Your workflow will look something like this:

JSON❤️API

APIs love JSON. Most modern APIs send and receive data in JSON format.

When you make an API request, you'll typically get back a JSON response like:

{
  "location": {
    "name": "London",
    "country": "UK"
  },
  "current": {
    "temp_c": 18,
    "condition": {
      "text": "Partly cloudy"
    }
  }
}

In n8n, this becomes available as $json in subsequent nodes. For example, to access the temperature, you would use {{$json.current.temp_c}}. If you need to specifically refer to a node in a workflow, then you can do that like this: {{$node[‘Node Name’].json.current_temp_c}}

Postman: Your API Testing Ground

Before building your workflow, it's often helpful to test API requests. Postman is a free tool perfect for this:

1. Download Postman

2. Create a new request

3. Enter your URL, method, headers, and body

4. Hit Send to see the response

This helps you understand what data you'll get back before building your n8n workflow.

Common API Challenges & Solutions

1. Rate Limits

Most APIs limit how many requests you can make (e.g., 1000 per day).

Solution: Add delay nodes between requests or use n8n's built-in retry logic.

2. Authentication Errors

The dreaded "401 Unauthorized" error.

Solution: Double-check your API key, token expiration, or credentials.

Copy

If using OAuth, check if your token has expired. Tokens typically last 1-2 hours.

3. Data Format Issues

API expects one format, but you're sending another.

Solution: Always check the Content-Type header and ensure your data structure matches the API documentation.

4. Finding the Right Endpoint

Not sure which URL to use for what you need.

Solution: Read the API documentation. Most provide a list of all available endpoints and what they do.

Using AI to Master APIs

Let's say you want to integrate with a new service but aren't sure how their API works. Here's how AI can help:

Ask for API basics:

I want to integrate with [Service Name]'s API in n8n. Can you explain how to authenticate and what the main endpoints are?

Generate HTTP Request configurations:

I need to [what you want to do] with [Service Name]. Can you show me the HTTP Request configuration I'd need in n8n?

Troubleshoot errors:

I'm getting this error from [Service Name]'s API: [paste error]. What might be causing it?

I'm getting this error from [Service Name]'s API: [paste error]. What might be causing it?

LLMs are excellent at helping you understand documentation and troubleshoot issues—often faster than combing through docs yourself! Just make sure that if it makes errors or can’t solve your problem, paste the API reference to it so it knows what to work with.

TL;DR for Non-Developers

• APIs = Digital waiters that help applications talk to each other

• HTTP Methods:

  • GET = Read data

  • POST = Create data

  • PUT/PATCH = Update data

  • DELETE = Remove data

• Authentication = Your VIP pass to access the API

• n8n's HTTP Request node = Your universal connector to any API

• Postman = Test APIs before building workflows

• AI = Your API assistant for documentation and troubleshooting

Model Context Protocol: Next gen API

While APIs connect different software systems, a newer protocol called Model Context Protocol (MCP) is emerging specifically for AI interactions. Recently open-sourced by Anthropic, MCP represents a significant advancement in how AI systems access and maintain context.

What is MCP?

Think of MCP as a specialized bridge designed specifically for AI systems to connect with data sources:

• APIs are like general-purpose highways connecting software applications

• MCP is like a dedicated neural pathway connecting AI models directly to data sources

The fundamental difference is in how they handle context:

• APIs typically handle single, stateless transactions: "Here's some data, do something with it, give me a response"

• MCP maintains ongoing contextual awareness: "Here's access to my database, remember what you find, and use it throughout our conversation"

Why MCP Matters for you

As you build more sophisticated automations with AI components, MCP could solve several key challenges:

1. Data Access Without Copying: Rather than copying data via API and sending it to an AI, MCP gives the AI direct, secure access to where the data lives.

2. Context Persistence: MCP helps AI maintain awareness of information across multiple interactions, eliminating the need to repeatedly provide the same context.

3. Unified Connection Standard: Instead of creating custom integrations for every data source, MCP provides a standard protocol—potentially simplifying how you connect AI with your business tools.

How MCP Works (Simplified)

MCP establishes a two-way connection between:

• MCP Servers: These expose your data sources (like Google Drive, Slack, or GitHub)

• MCP Clients: AI applications that connect to these servers

For example, instead of writing code or setting up a node in n8n to extract data from your CRM, format it, and send it to an AI through an API, you could set up an MCP server for your CRM. The AI could then directly query relevant information as needed throughout a conversation.

The promise is that you should be able to generate something like a n8n workflow on the fly without having to determine everything personally.

While still in its early stages, MCP represents an important step in how we'll integrate AI into our automation workflows. For no-code builders, this means future tools may offer simpler ways to create context-aware AI agents that can reason across your business systems without the complexity of managing multiple API connections.

As n8n and other no-code platforms evolve, watch for MCP integration options that might make building sophisticated AI workflows more accessible to non-technical builders. I’m also going to do a deep dive into how MCP works in the coming days/weeks.

That’s it. This concludes the four part introduction series into no-code. Once you understand these concepts you understand everything you need in order to build n8n workflows. Now all you need to do is practice, practice, practice!

The reason for that was simple: I always found the specialist classes like Sorceress or Barbarian boring. I wanted something that was unique, a bit of both worlds. Something that’s not only capable of doing magic but also strong as hell.

Then the Druid came with it’s hurricane, meteor shower and the ability to turn into a bear and maul everyone.

A jack of all trades, master of none…and for me, it was better than a master of one.

Mainly because that’s just how I see myself: neither here nor there. I was never a particularly talented business guy and I was never a good coder. Most engineers lack commercial skills and business folk lack engineering skills. I’m mediocre at both which means I had a unique perspective in things. When AI started taking over the world, it felt as if my time has come. Expert knowledge is becoming free so versatile generalists with high agency are becoming more in demand. My time has truly come.

Turns out I’m not alone.

About 300 new people sign up to the Lumberjack every month. Most of these people share a similar trait: they’re not engineers but not entirely non-tech either.

I’m not surprised about that. Like attracts like.

If my writing resonates with you, chances are that you are like me. You’re not technical enough to look through the code of new AI developments, but you’re technical enough to realize how impactful this technology is.

…and now you want to dig deeper.

AI Agents 101

I get this question from my readers and colleagues often. The last time I was asked was about two hours ago from a colleague at my day job where I work as an AI Architect.

So I put together a curated list of reading materials. If you want to get knowledgeable enough on AI in a weekend, go through this list and you’ll be ahead of 99% of the people out there claiming they’re “AI experts“.

This list is freely available, no login, no signup, no nothing, just click on the link and it’ll direct you to the page. Happy reading!

Read AI Agents 101

So I’m starting an experiment: Hire a No-Code Builder program.

The program is simple.

• If you’re a client you can submit your application and I’ll try to match you with a builder. The exact steps are described below in the “How to hire a builder” section.

• If you’re a freelancer, I’ll vet you and if you pass screening I’ll add you to the database. Details are in the “How to make money as a builder” section.

How to hire a builder

First off you’ll need to visit my brand new application page: https://apply.lumberjack.so

You’ll need to sign a form at the bottom of the page and I’ll be in touch soon.

The process takes 6 steps.

1. It starts with you submitting an application.

2. If you qualify, I’ll be in touch to schedule an introductory call.

3. After our call I’ll try to see if I can find you an builder in our database.

4. When I do, I reach back out to you and that’s when we make a deal.

5. Once we’ve agreed on a deal, you pay 50% deposit on the agreed price, I’ll handle all the paperwork and the work gets started.

6. I’ll try to aim projects for a quick 2 week turnaround, but if your project is complex, that might be slower. Once done, we do a demo and a handover and your prototype is live!

Pretty simple, eh? Go submit that application now and hire a builder today:

I'm a client, I want to hire a builder

How to make money as a builder

In order to apply to the Lumberjack no-code expert database, you will need to apply via the following link: https://apply.lumberjack.so/freelancers

I’m looking for a wide variety of skills so even if you’re only using one or two of these tools, feel free to apply!

In general no-code builders make $50-150 per hour. Some projects will be hourly based, some others will be project-based. In any case I’ll reach out to you when I think there’s a good fit for a project and your skills.

Here’s what to do:

1. Submit your application with your portfolio, your website or your CV.

2. I’ll screen your application and send you a test exercise based on your selected skills.

3. If you successfully complete the test exercise I’ll add you to the database.

That’s it. So if you’re a no-code developer, a freelancer or a consultant, please apply so I can connect you to the right clients.

I'm a builder, I want to work

That’s it. I’ll start running this as an experiment and see how many people submit their applications. I’ll be back with the results in a few weeks. In the meantime I’ll start processing applications and reach out to everyone individually.

If this works I might turn this into a matchmaking service and take a small percentage of the total deal. But for now, everything will be concierge’d.

Now I found it and I can finally launch Lumberjack 2.0.

Join Lumberjack 2.0

Over the last few weeks we’ve grown from a small blog into something much bigger. A community of no-coders building, hacking together. An academy of some sort for the misfits of the tech industry. Now we have a home on a brand new Lumberjack Community platform. (I know how many of you despise Discord so I’m sure this’ll bring a sense of relief to many.)

Here’s what you’ll find inside:

1. All my courses for FREE. Knowledge is free, time is money.

2. All upcoming events, free or paid.

3. A dedicated space for no-coders to meet and discuss projects.

4. An automated Buddy System that matches you up with other no-coders to help you make new friends and business partners.

5. An ever growing library of ALL my projects with tutorials, code, descriptions.

BONUS: I’m giving away the source code of my projects for free (except for alfredOS)

This follows a CPAL License which means you’re free to use any code you find on the Lumberjack Community, modify, edit, repackage, resell, monetize however you see fit with one restriction: you must attribute the original work to Lumberjack.

Come and join us! It’s gonna be a lot of fun!

Accept the invite and join for free now.

Your Business Is a Road Trip

Imagine your business as a road trip across the country. You've packed your essentials (your data) in a suitcase (JSON, remember that from the last lesson?).

Now you need to decide which roads to take.

That's exactly what control flow is in your automations—the system of roads, intersections, and traffic signals that guide your data from start to finish.

The Automation Highway System

On our business road trip, your data needs to travel from one destination (node) to another along specific routes (edges). This is why node-based thinking is important. If you’re not familiar with it, start here:

But unlike a straight highway from New York to LA, your business processes have decision points:

• Some clients need the VIP route

• Some inquiries require detours to your support team

• Some processes need to loop through multiple stops before continuing

In both programming and no-code tools like n8n, control flow gives you the power to build these decision points without writing code. Some have limited control flow options (like Make) and others have nearly identical to programming but without feeling like you’re hacking the Matrix (like n8n).

Conditional Paths - Traffic Lights

The most basic control flow tool is the IF Node—your automation's traffic light.

🚦 IF Node: The Traffic Light

Just like a traffic light that says "if the light is green, go; if it's red, stop," an IF Node makes simple yes/no decisions for your data.

Real Example: Client Intake Sorting

Let's say you receive new client inquiries through a form. Each submission creates a JSON "suitcase" like this:

{
  "client_name": "Alex Kim",
  "company_size": "1-10",
  "budget": "$5,000",
  "project_type": "website redesign",
  "timeline": "urgent"
}

Let’s say you have two workflows set up already. One that processes regular clients and one that processes priority clients.

Here's how you'd use an IF Node to direct this data:

1. Connect your form node to an IF Node

2. Set the condition: {{$json.timeline === "urgent"}}1

3. Connect the "true" path to your "Priority Response" node

4. Connect the "false" path to your "Standard Process" node

Now your workflow automatically sorts clients based on urgency—no manual checking required!

Conditional Paths - Interchanges

Sometimes a simple yes/no decision isn't enough. What if you need to route clients based on both timeline AND budget?

Enter the Switch Node—like a highway interchange with multiple exits.

🔄 Switch Node: The Highway Interchange

Real Example: Service Package Routing

{
  "client_name": "Taylor Jordan",
  "budget": "$15,000",
  "project_type": "full marketing campaign"
}

Let’s say you have multiple packages and you already know the budget. With a Switch Node, you can create multiple paths, assigning the right path to the client:

• If budget > $10k → Premium Package workflow

• If budget $5k-$10k → Standard Package workflow

• If budget < $5k → Starter Package workflow

In n8n, this looks like:

1. Add a Switch Node after your intake form

2. Create 3 conditions for each budget tier

3. Connect each "output" to different service package workflows

Now your client gets automatically routed to the right service level without you lifting a finger!

Rest Stops: Handling Lists of Items

Sometimes your road trip requires stopping multiple times along the same route. Maybe you need to process each item in an order, or send individual emails to every member of a team.

That's where the Loop Nodes come in—they're the rest stops that let you handle each item one by one.

🔁 Loop Nodes: The Rest Stops

Real Example: Processing multiple applications

Let’s say that you run an agency and you sell your services via a live webinar. This means that ideally you have 5-10 customers signing up at the end of your event. You can just send them an application form and run the above flow for every participants, but that means you’d need to start and stop the workflow for every person, slowing the process significantly.

Let’s see what we want to do. When someone signs up, we want them to

1. Fill an application form

2. Assign package based on budget

3. Assign priority based on urgency

4. Invite them to the relevant group on your client portal

Your JSON might look like:

{
  "customer": {
    "name": "Jamie Smith",
    "email": "jamie@lumberjack.so"
  },
  "service": {
    "budget": "10000",
    "urgency": "urgent"
  }
}

When you send multiple items to a loop, n8n will handle that without any extra steps. But if you need to download the data from a spreadsheet first, you need to split them into separate items.

Here's how to process this with Loop Nodes:

1. Use a "Split In Batches" node to separate the modules array

2. Process each module individually (grant access, send email)

3. Use a "Merge" node to combine everything when done

Once the loop is done, it will forward only one item. However if you need let’s say create a new spreadsheet with a full report of all people that were processed, you’d want to Merge the data first and then create the spreadsheet.

Now each module gets processed correctly, and your customer gets a smooth onboarding experience!

Roadside Assistance: Error Handling

Every road trip encounters unexpected problems—flat tires, wrong turns, or bad weather. Your automation needs backup plans too.

That's where Error Handling comes in—your workflow's roadside assistance.

🛠️ Error Handling: Roadside Assistance

Real Example: Backup for No Budget

When your clients sign up, they might forget to fill out the budget part or they might give you badly formatted responses. This would create an error in n8n but luckily you can handle that in multiple ways. In the node’s “Settings” you can use the following:

1. Retry on Fail: this will attempt to run the node again when an error happens.

2. On Error: Continue (using error output): This will add a new output to your node called “Error” and you can build the process there.

Let’s see that in our example.

This ensures you know immediately when something goes wrong, and your automation can recover gracefully.

Putting it all together

The skeleton of our automation is ready, but we still need to determine the input and the output. When you’re building an automation, this usually is the only thing you know at first.

In our case, let’s say we want to run our automation every morning at 9am. This means that we’re going to create a time-based trigger (as opposed to an action-based one). Time based triggers automatically run at set intervals, while action-based triggers will only run when the defined action happens.

Lastly, set our output step as sending a Slack notification about the spreadsheet report we just created so the team knows what’s up.

There you have it. We went from a simple use case, expanded it to cover most elements of the control flow and built a full automation. This is not an exhausting list but this is the only thing you need to know to get started.

I also created a simple animation for you as a Claude artifact to visualize this:

Using AI to Navigate Complex Routes

Remember how we used GPT-4o-mini to help with JSON in our previous article?
You can do the same for control flow!

When you're not sure how to structure a complex decision tree, let AI act as your tech advisor:

1. Open ChatGPT or Claude

2. Describe your workflow goals and decision points.

3. Ask it to design the control flow structure

4. Implement the suggestions in n8n

Pro tip: If you ask for a Mermaid chart, Claude will not just explain but also visualize this for you.

Example prompt:

I'm building an n8n workflow to qualify sales leads. I need to route leads based on:

- Budget (under $5k, $5-25k, $25k+)
- Urgency: (urgent or not urgent)

High priority leads are those that marked "urgent" in the lead form.
Everything else is low priority.

If the budget is under $5k, assign the Starter Package.
If the budget is $5k-$10k, assign the Standard Package.
If the budget is $10k+, assign the Premium Package.

If it's a high priority client, do the following steps: intake call, concierge onboarding, register client file.
If it's a low priority client, do the following steps: send intake form, send onboarding email, register client file. 

Help me design the control flow for this workflow. Create a Mermaid visualization of the flow.

Claude will think through the logic and generate the basic flow for you. Then if it’s too complicated you can ask it to keep refining until you’re happy with the results.

Important: You should only start building something once you’ve done this step!

TL;DR for Non-Developers

Control Flow = GPS

• IF Nodes: Simple yes/no traffic lights for your data

• Switch Nodes: Highway interchanges with multiple exits based on conditions

• Loop Nodes: Rest stops for processing multiple items one by one

• Error Handling: Roadside assistance when things break down

Quick Start Guide:

1. Map out the decisions your workflow needs to make

2. Choose the right control flow tool for each decision

3. Connect your nodes with the appropriate paths

4. Test with sample data before going live

5. Add error handling to key points

Your First Control Flow Challenge

Ready to try this yourself? Here's a simple exercise:

Create a workflow that sorts incoming emails into folders based on:

• If it contains "urgent" in the subject → Urgent folder

• If it's from a client email address → Client folder

• Everything else → General folder

Use an IF node for the "urgent" check, and then a Switch node for the remaining routing.

Need More Help?

I built a simple custom GPT tutor that will help you understand control flow better. If you get stuck you can also use this GPT to help you draft your flow.

You can try it here, it’s free.

Next week I’ll show you how to connect your automated workflows to other apps using APIs—without writing a single line of code.

If you enjoy my blog, it would mean the world to me if you invited friends to subscribe and read with us. If you refer friends, you will receive rewards, including a free ticket to my 2 hour live No-Code Crash Course, a free 30 min consulting session over Zoom and free access to alfred OS, a solopreneur business tool that saves you $1k/month on SaaS fees.

How to participate

1. Share Lumberjack. When you use the referral link below, or the “Share” button on any post, you'll get credit for any new subscribers. Simply send the link in a text, email, or share it on social media with friends.

Refer a friend

2. Earn benefits. When more friends use your referral link to subscribe, you’ll receive special benefits.

Level 1 ⭐️

Once you get 25 people to sign up, you’ll get a ticket (sold for $47) to my 2 hour LIVE No-Code Crash Course that will get you from no-code muggle to the Harry Potter of n8n in 2 hours.

Here’s more info on the workshop.

Level 2 ⭐️⭐️

Once you get 50 people to sign up, you’ll get a free 30 minute Zoom consultation (sold for $150) with me where we can discuss your no-code and AI needs and I can help you live.

Level 3 ⭐️⭐️⭐️

If you’re super awesome and love my work and get me 100 subscribers, I’ll give you a free copy of alfred OS. My business operating system that slashes your SaaS subscription fees by replacing every tool you use with a free open-source alternative.

Here’s more info on alfredOS.

Ready? Visit the leaderboard to get started!

Visit the leaderboard

To learn more, check out Substack’s FAQ, because this rewards program is a built-in Substack feature. Thank you for helping get the word out about Lumberjack!