One essay a month. Things that only get written when someone has run the show.

The two camps killing AI adoption

and the boring middle that wins

The thesis essay. The contrarian argument that owners are split into two losing camps — "AI is a fad" and "replace everyone with agents" — and that the actual winning move is in the middle. Why most strategy work right now is reactive instead of considered. What twenty years of operations changes about how you read the AI hype cycle.

Your context should outlive your vendor

evoctx, the memory layer we open-sourced — and the hostile review that shipped it

The launch note for evoctx, our open-source MCP memory server. Why every vendor’s memory feature is a moat pointed at you, what a local typed store looks like instead, and the part we think matters most: the hostile security teardown we ran on our own product before shipping it — written by the same assistant that helped build it.

What 20 years in telecom ops taught me about scoping AI projects

Specifically: the failure modes that look new but aren’t. Scope creep is scope creep, whether you’re building a billing system in 2008 or an agent in 2026. The patterns are older than the tools.

Available when it’s ready, not when a calendar tells us to.

Why I left dev to run a business, then came back — and what that gap is worth in 2026

The biographical essay. Less about the founder, more about what happens to your judgement when you spend a decade doing the thing instead of building the tool that does the thing. Why that gap is the moat now, and why it wasn’t five years ago.

Available when it’s ready, not when a calendar tells us to.

When Claude is wrong

the kinds of mistakes only an operator catches

A working document, updated as new examples appear. The taxonomy of model failures that look right but aren’t, organised by the operational background required to catch them. Worked examples from our own build log.

Available when it’s ready, not when a calendar tells us to.