Update, 16 October 2026. This was the first post, and I leave it as I wrote it on 26 July 2026. The mission it describes has since changed its centre. This space is no longer led by a transition into AI Engineering, or by the code as the argument. It is my personal, independent commitment to learn in public with the people and machines of this era, turn what I live into language, and give back what I find — for as long as I can sustain the tools and infrastructure behind it.

I write it as an ongoing book: each article a situated chapter, what I thought at one point of the journey rather than a final word; the Library, the books I learnt from. The harness is my workshop, not my subject. I believe harness engineering is the new modern art: language is the material, constraints are the form, models are the instruments, and the human–machine relationship that emerges is the work. That is my conviction, and you are free to reach your own.

What follows is where I started.

I'm a software engineer in the middle of a transition — from cloud and distributed-systems architecture into AI Engineering — and I decided to do it in the open.

Most "how to become an AI Engineer" content is one of two things: PhD-in-ML gatekeeping, or hype threads. Neither helped me. So I'm writing the thing I wish I'd found: the real transition, with the trade-offs made explicit and the code open for you to take.

And this space is more than a blog: it's the workspace where I run that transition in the open — building AI agents, sharing the artifacts as I make them, and collaborating on open source as I go. I'm not here as an authority; there's no master of AI yet — the field is too young. I write as a peer, learning alongside you, experimenting out loud and showing what works and what doesn't.

What I'm not going to do

I won't claim years of agentic experience I don't have — the tech is new, my hands-on AI work is months old, and I'll say so. I won't sell you a shortcut. The substance I bring isn't the AI hype; it's ~two decades of SDLC and distributed systems — the unglamorous production rigor (evaluation, observability, architecture, cost, failure modes) that most AI work skips. AI is the newest layer on that, not a fresh start.

The commitment

• Honest over impressive. Real trade-offs, including what didn't work. If I don't know something, I'll say so. A pitch that can't survive a hard question isn't worth writing.
• Show the machine, not just the output. Everything here is public and reproducible — the code is the argument.
• Reader-first. The goal is to help you build better. Whether it helps my own repositioning is a by-product, not the point.
• Sustainable, not performative. I have a full-time job. This grows on weekends, at a pace that survives a bad week — a cadence that lasts beats a burst that burns out.

If you're on the same path

Your background isn't a deficit to overcome. It's the thing that makes you rare in this lane. Build in public, ship small, and let the artifacts do the arguing — and if any of this is useful, take it, improve it, and build on it in the open.

That's the deal. Let's build.