
A Well Thought Out Harness Can Substitute Anything
It doesn't need a new interface built around it. It just needs a good enough one already doing the thinking.
Read EntryVersion 1.2
Direction, velocity and meaning in progress.

It doesn't need a new interface built around it. It just needs a good enough one already doing the thinking.
Read Entry
The best team I can build with AI isn't ten people plus ten bots. It's ten people who each operate like a small team, with one human accountable for all of it.
Read Entry
I run a three-layer agent architecture in Claude Code — one layer inlined as my digital twin, two on demand when needed. Here's what each layer does, why it is beyond solid, and how I've built a skills library on top of it.
Read Entry
The step between a finished session and something you can actually hand to someone.
Read Entry
Socratic Agentic Workflow v0.4.0: same interview, but the scaffold it hands you is no longer a guess from whenever I last wrote it down.
Read Entry
From first question to a system that runs every morning.
Read Entry
I recently spent a day building a structured AI tool for Apple platform design. What surprised me wasn't the outcome, but the ease of decision making along the way.
Read Entry
Most software projects fail because coding begins before the problem is fully understood. I wanted a tool that forces teams to articulate both the "why" and the "how" before writing a single line of code.
Read Entry
The web was built for human eyes. That made sense. But agents don't have eyes — they read structure, not style, and most of the web was never designed that way.
Read Entry
Linear is a reaction to teams that spent more time managing their tools than shipping. Opinionated, minimal, built for closing tickets not debating them.
Read Entry