Senior Product Manager / Builder
AI Products · Enterprise SaaS · Agent Harnesses
Featured
The same model feels smart in Claude Code and flat behind a raw API. The difference is a stack of product decisions.— The Harness Is Where Your Taste Goes · 9 min read →
By the question they answer
Q1.Why does my AI product not feel as smart as Claude?
You own the data, the domain, the users. Claude Code with one skill file still does the job better. Here's why, and what is actually left to own.
The same model feels smart in Claude Code and flat behind a raw API. The difference is a stack of product decisions. Most of them, lately, are yours.
Q2.How do I build a coding stack as a PM?
Working artifacts · tracked on GitHub
Point it at a monument and it explains what you're actually looking at. Built in a weekend at the Google DeepMind vibe-coding hackathon.
Demonstrates
The multi-model setup behind every build on this site. Not a listicle, the actual working system.
Demonstrates
The working build behind the A Layer Down essays. If the harness is where your taste goes, this is mine: designed in the open.
Demonstrates
A decade in product · the distance collapsed
Most of my career I built by proxy: discovery, judgment, specs, handoff. Then the distance between idea and artifact collapsed. Now I follow product decisions past the interface, into the system itself.
Where the judgment comes from
Where AI met the enterprise: end-user adoption and AI-based work intake.
First AI products shipped: Copilot for recruiters, and revenue owned directly with Referrals.
The founder year: 1.3M visitors, every lesson self-funded.
First products owned end to end.
Where the code discipline came from.
The useful conversations start below the feature
Open to conversations about AI products, harnesses, enterprise systems, and where product work now meets engineering.
