Hello, world

A short note before the real work starts.

What this is

A working notebook on building AI systems that survive production — agent reliability, orchestration, and the kind of boring governance that keeps things working when the model changes underneath them.

Written for the engineers and CTOs who have to ship the hard, expensive parts.

What you should expect

A new piece roughly every two weeks. The mix:

• Essays — opinionated takes on what actually breaks in production agent systems.
• Deep dives — long-form on a specific problem: eval design, audit trails, latency budgeting, identity in agent runtimes.
• Project logs — notes from shipping the AI Control Plane, the flagship project this site partly serves as a journal for.
• Reviews — short, focused readings of papers or systems worth your time.

Every piece will have at least one of: code I wrote, a diagram I drew, or a concrete example from a real system. No pure-opinion pieces.

What you should not expect

• Beginner-friendly AI tutorials. I write for peers.
• Hype takes or “10x AI hacks”.
• Anything specific to Alpian Technologies, where I work. Strict separation — personal projects and general industry topics only.

The small thesis

If I had to compress my view on the next two years of building with AI into one sentence, it would be this:

The model is the cheap part. The runtime around it is the product.

Most of what comes next on this site is a version of that argument, applied to specific subsystems — identity, latency, audit, governance, human review.

What’s next

The first long piece — Why Most Enterprise AI Agents Will Fail in Production — lands within the next two weeks. Subscribe to the RSS to get it as soon as it does.