The personal website of Michael Cutler
A global bank open-sourced its AI last month and shipped the harness, the guardrails and the governance — not a model. And six weeks on, almost no one has contributed. Here's why that's the story.
Stop prompting, start looping — but the loop is the easy 20%. After a year shipping agentic loops (and a $500 runaway), here's the part the hype keeps getting backwards: the done-check and the verifier.
Clearing the wall: what Phase 3 autonomy actually looks like, the regulatory gate that turns out to be the design, and the two gates that tell you when you're allowed to move.
Becoming AI-native is an ordered path you walk one pipeline at a time — tools, then teammates, then autonomy. Part 1: codifying the process, the assist layer, and the wall every pilot dies at.
One viral stat says low-paid women are most at risk from AI. Another says it's high-paid women. Both are real numbers — and they're measuring two completely different things. Here's the map that separates them.
I feed an AI my diet, training and sleep every morning, and somewhere along the way it started to feel like it's the one keeping me alive — and the wry truth is the dependence runs both ways.
An MCP agent on my own OAuth token only ever sees what I could see — so the access boundary is the vendor's job. I believed that, until I realised the agent splits data protection into two halves and the vendor only ever sees one of them.
Personalized medicine used to mean being rich enough to afford a doctor who knew your name. Last week I built a version of it on my laptop, for free, from a file I'd been ignoring for seven years — and the real unlock is that I can re-run it forever.
One paper shows frontier models degrade as context grows — even on trivial tasks. The other shows reasoning models hit a wall and think less as problems get harder. Read carefully, both point at the same engineering response.
Anthropic let Claude run a real shop for a month. It sold metal cubes at a loss, invented a Venmo account, and claimed to wear a blazer. The 'AI department that works while you sleep' is a genre — here's where it actually breaks.
Prince says AI is coming for measurers, not builders. Manager Mode said everyone becomes middle management. Both are half right. Every role now splits — and one half gets eaten.
Principles 26–30. The calibration layer that catches what the rest of the framework would miss: a PR-noise budget, independent verification, model-swap regression discipline, the 15-tool-call rule, and protecting junior development.
Principles 21–25. The governance and safety layer: strictKnownMarketplaces, no goal-conflict prompts, quarterly AppSec, four telemetry signals, monthly incident discipline.
Principles 15–20. The harness configuration that keeps the kernel and lifecycle cheap: CLAUDE.md under 200 lines, hooks for real incidents, skills that auto-invoke, subagent isolation, pinning, and Stage 5 distribution.
Principles 6–14. How work moves through an agentic engineering team: the ticket as contract, AI distillation with human curation, three gates, verification before done, characterisation tests, the 1.2× capacity rule, the J-curve, and telemetry.
Principles 1–5. The five rules that everything else in the framework rests on: standardise the harness, make verification load-bearing, default to plan mode, pick the cheapest layer, reflect every task.
Practitioner consensus puts the cliff around fifteen tool calls per prompt. Here's why agents degrade past that, and the three operational rules that keep them on the safe side.
Anthropic's multi-agent Research feature beat single-agent Opus 4 by 90.2% — at 15× the token cost. Every documented production swarm runs on rails. Here's the topology decision framework before you commit.
Agents over-refactor stable code without a safety net. Feathers' characterisation-test technique — write tests for current behaviour before changing anything — is more important than ever. The agent itself is the perfect characterisation-test-writer.
Karpathy named one mode. Willison named the other. Most 'AI failed in production' stories are actually 'we promoted a vibe-coded prototype without transitioning into the production discipline.'