Vibe Coded · Part 10 — Agentic Live

March 2026

This past weekend, I built a kanban board for my agentic projects. Nothing fancy — a rough board for tracking features and tasks. But the goal wasn't the board itself. The goal was to connect an AI agent to it so the agent could work through the backlog autonomously and develop the system itself.

For the last few years, once a year, I've tried to build a fully automatic software pipeline — one that doesn't need a real developer. It's never worked. Until now, I think it might.

The Full Circle

Fifteen months ago, on December 23rd, 2024, I generated 10 websites in parallel for $2 and asked LinkedIn which one was best. Today, I'm building systems where AI agents pick their own tasks from a backlog, write the code, test it, and deploy it.

The distance between those two points is enormous. But looking back, the path was surprisingly straight:

AI as tool: "Generate this for me." (Dec 2024)
AI as pair programmer: "Help me build this." (Feb-Apr 2025)
AI as junior developer: "Build this while I review." (May-Aug 2025)
AI as team member: "Handle this area of the project." (Sep-Nov 2025)
AI as autonomous agent: "Work through this backlog." (Dec 2025-now)

Each step required more trust and less control. Each step was terrifying the first time and obvious in retrospect.

The Question

"Have I coded myself out of a job?" is the question I keep coming back to.

The honest answer is: I don't know. And that uncertainty is itself remarkable. I've been building software for over 20 years. I've survived every previous technological shift — from mainframes to web, from web to mobile, from on-prem to cloud. Each time, the old skills remained relevant even as new ones became necessary.

This time feels different. Not because AI replaces developers — the last 15 months have shown me that humans are still essential for judgment, creativity, and the "what should we build?" question. But because AI is compressing the value chain of software development so rapidly that the economic model changes.

When building costs 10x less, you don't need 10x fewer developers. You need 10x more ideas. The bottleneck shifts from execution to imagination. And that's a different job than the one I was trained for.

I wrote in my most recent post: "I honestly believe there's a probability that we'll build something that means I don't know if what I'll be working with in 5 years is what I do today."

What Claude Code Taught Me This Week

Even now, 15 months in, the tools keep surprising me. This week's observation: "Claude Code is like all other developers. It'll find a way to solve its goal — the only thing is whether the road there was good enough."

There's a motorcycle metaphor I used: remember those videos of motorcyclists weaving through traffic at 300+ km/h? Is that "good coding" or is it reckless? The result — getting from A to B fast — looks impressive. But the path matters. Code quality matters. Architectural decisions matter. Not because the AI can't get things done, but because the next AI session needs to understand what the last one built.

That's the real argument for code quality in the age of AI: not human readability, but AI readability. Code that's clear, well-structured, and follows consistent conventions isn't just easier for humans to maintain — it's easier for AI agents to extend.

The Agentic Live Community

The biweekly Friday sessions have become something I look forward to. A small group of people watching me code live, asking questions, sharing their own experiments. It's not a polished production. It's a workshop. Sometimes I build features. Sometimes I debug for 45 minutes. Sometimes the AI does something incredible and we all just watch.

The format works because it's real. No slides, no scripts, no carefully curated success stories. Just a developer and his AI agents, building software in real time, with all the messy reality that entails.

I've also started streaming directly from the terminal — anyone can watch what Claude Code is doing in real time via a tool I've been building. The radical transparency isn't for show. It's because the best way to learn agentic development is to watch it happen.

What I Believe Now

Fifteen months of daily AI-assisted development have crystallized some beliefs:

The technology was ready a year ago. Most people are waiting for the next model, the next tool, the next breakthrough. They're missing the fact that what we have today is already transformative. The bottleneck is human adoption, not AI capability.

Rules beat intelligence. A cheap model with excellent rules and context produces better results than an expensive model with vague instructions. Invest in your ADRs, your conventions, your project documentation. That's where the leverage is.

Build over buy. The economics of custom software have changed fundamentally. Organizations that recognize this early will have a massive advantage.

The team matters more than the individual. My personal productivity peaked months ago. The next frontier is making entire teams productive with AI. That's an organizational challenge, not a technical one.

Judgment is the last human skill. AI can code, test, deploy, and debug. What it can't do is decide what should be built, who it's for, and whether the tradeoffs are acceptable. That judgment — informed by experience, empathy, and business understanding — is what makes a developer valuable in 2026.

What's Next

The kanban board agent is running. It picks tasks, writes code, runs tests, commits if they pass. It's crude. It's slow. It makes mistakes. But it's autonomous. And with each iteration, it gets a little better.

Am I coding myself out of a job? Maybe. But I think I'm also coding myself into the next one. Whatever that turns out to be, I'm confident it'll involve thinking about what to build, not how to build it. It'll involve judgment, not keystrokes. And it'll involve trusting an AI to handle the implementation while I focus on the vision.

That's the journey from December 23rd, 2024 to March 8th, 2026. From $2 websites to autonomous agents. From curiosity to conviction. From vibe coding to agentic development.

The code writes itself now. The question is what to tell it to write.

This is the final post in a 10-part series documenting my journey from traditional software developer to agentic AI developer, December 2024 – March 2026. All dates and quotes are sourced from 155+ LinkedIn posts during this period.

If you're at the beginning of this journey, my advice is simple: start. The technology is ready. You don't need the perfect tool, the latest model, or a comprehensive plan. You need curiosity and a willingness to feel uncomfortable while a machine does something you used to do by hand.

The rabbit hole is open. Jump in.

Series Index: