MANIFESTO #funny
A quick follow up to hiring in 2026
Engineering Heresy
Challenging conventional wisdom in AI and software engineering. Deep explanations, mental models, and practical heresies for building better systems with less bullshit.
Latest episodes
I'm Tired of Broken Engineering Hiring, So Here's How I'll Hire Instead
Eng hiring in 2026 still runs on ATS filters and trivia theater. What I'll hire for instead: philosophies that survive hard problems, proven in a
I Pay My Agent Harness 4× So I Never Have to Babysit Parallel Workers
It reviews its own security, checks its own quality, and writes its own docs so I can launch a dozen of them and step away. I benchmarked it
I Kept Blaming the Model. The Fix Was Never in the Model.
I swapped models three times to fix one flaky test. None of it helped. Every reliability lever in AI-assisted development lives in the layers you
agentic-os v2: The Tool You Installed Isn't the One Running Now
Since the last release I wrote up, agentic-os got about 47% smaller, grew a set of hooks that fight its own context drift, turned them on by
The Day a Frontier Model Got Switched Off. Security Is Now Metric #1.
A government just switched off a frontier AI model overnight. Why security not velocity is now your top engineering metric.
I Got Tired of AI Code Review Noise, So I Built a Ratchet
My four LLM reviewer agents kept re-discovering the same findings, run after run. The fix: tier the checks, ledger the noise, and promote what
AI Made Engineers Faster. It Also Made Teams Slower to Integrate.
AI made every engineer faster and your team slower to ship. The velocity is on the dashboard. The collaboration tax that pays for it isn't, and I
Modern Claude Code: The Complete `.claude/` Anatomy
You're using maybe a third of what `.claude/` can do. Here's the complete modern anatomy organized by directory and scope precedence and the
I've Been Gatekeeping the Magic. Here's Everything.
Senior engineers have been quietly building AI scaffolding that juniors never get access to. I just open-sourced mine.
Build Your AgenticOS: Watch It Run
Seven posts explaining the system. This one shows it a real engineering task, start to finish.
Build Your AgenticOS: Hooks Automate the Invisible
Hooks wire your AgenticOS to session events. They're the automation layer that removes the work you'd otherwise have to remember to ask for.
Build Your AgenticOS: Worktrees for Parallel Agents
Parallel agents stomping files is an isolation problem, not a parallelism problem. Here's the worktree pattern that fixes it.
Build Your AgenticOS: The CLAUDE.md Constitution
CLAUDE.md is where your AgenticOS rules live. It governs every session, every agent, every project.
Build Your AgenticOS: Memory That Survives Sessions
The layer that makes your AgenticOS learn, one non-obvious fact at a time.
Build Your AgenticOS: Specialist Agents
The constraint is the feature. One agent definition file is more reliable than any prompt you've ever written.
Build Your Own AgenticOS, Part 2: Shapers, The Intake Layer
Intake quality is the ceiling on execution quality. Here's how to build a shaper that scopes briefs before any agent touches a file.
Build Your AgenticOS: Start with Skill Files
Skill files are the atomic unit of a Claude Code AgenticOS. Get this layer right and every agent you build downstream becomes composable.
Build Your Own AgenticOS: The Complete Map
An AgenticOS is a composable, version-controlled layer that makes AI behaviour consistent, reviewable, and delegatable and you can build it in a
You Don't Get to Pick CA
CAP says pick two, but partitions aren't optional. So you're really choosing Consistency or Availability *during* a partition. And PACELC is the