The builder track, end to end

I run software companies without a software engineering background. I have never been employed to write code, and I still don't write it. AI coding agents are my entire engineering capability. My work is everything around the code: deciding what gets built, designing the architecture, and building the verification and governance systems that catch the machines' mistakes before users and clients see them.

This page is the honest account of how that happened and what it makes me.

The journey

In May 2025 I was using AI like everyone else: a chat window and a lot of questions. I started tracking my own capability against a 10-level ladder I maintain for exactly this purpose, reassessed every few months against hard evidence from my own repositories.

• Mid 2025: first real integrations. AI connected to actual business systems, first deployments, first things that broke and had to be fixed.
• Late 2025: production systems. Multi-agent architectures, real users, the first platform serving a company rather than a demo.
• Early 2026: products with external users, client work prepared and delivered with AI, revenue. A consulting practice running on AI-generated materials.
• Mid 2026: the platform stage. A shared engine powering a client-delivery platform across dozens of engagements. Teammates shipping daily on infrastructure I built: in the five months to August, the team operations repo logged 2,800 commits and only 29 were mine. Scheduled agents doing real work, with health monitoring I designed because no engineering department was going to debug them for me.

Fifteen months, level 2 to level 8 on my own ladder. The honest caveat: it is my ladder, scored against my own evidence, not an industry benchmark.

The cohort this puts me in

There is a specific, small group forming right now: people with no engineering past who reached platform-level output using AI as the whole engineering department, and then kept those systems alive through incidents, security scares, and maintenance, with no engineer to fall back on.

Three things define membership:

1. No code written professionally before AI tools existed. Not a rusty engineer, not a technical product manager. The pre-2024 version of this person could not have built software without hiring.
2. AI as the entire engineering capability, not an assistant to it. The human never becomes the coder.
3. Production systems other people depend on and operate, sustained over time. Launched apps are nearly free now; operated platforms are not.

This group could not have existed before roughly 2024, because the substitution it runs on was never available before. Historically there were two ways to get engineering done: pay engineers or spend years becoming one. AI created a third: direct the work and verify it. Very few people have pushed that substitution all the way to running platforms.

The filter is not building. The filter is surviving: the silent pipeline failure you diagnose yourself, the data leak your own safety gate catches, the scheduled job that dies quietly and looks identical to one that succeeded. Every one of those is the moment where most projects by people like me end. Mine didn't, and that is the actual credential.

There is no job title for this yet. "Founder" undersells it, "developer" is false, "AI user" describes a billion people. When a real group exists before its label does, that usually means it is early.

What the work actually is

Since I can't out-review a senior engineer line by line, I compensate with systems. Decision records instead of memory. Verification loops instead of code review. Pre-push gates that scan for leaks instead of trusting myself to notice. A registry where every scheduled job must announce that it finished, so that a job that died looks different from a job that succeeded. The engineering skill is orchestration and verification, not syntax.

The rest of this site is that method written down: the Compilation volumes cover how knowledge compounds, and the Factory volume covers the agent workforce and its safeguards. This page is just the part where I say plainly who is writing.

Where this goes

The next stage is making the platform observable to itself: cost, usage, and quality measured, not just gated. And publishing the patterns that kept these systems alive, starting with the ones I had to invent. If those patterns help other people build the same way, that is the last rung of the ladder.