Keep codingbefore AGIreplaces me.
Until then, building software that turns frontier models into products people actually rely on.
// currently -
Start hereI Cloned Myself Into Slack So I Could Go SkiingRead the essay→
Recent
writings.
- One Claude Session Cost Me $1,880. It Was Reading Four Novels to Write One Tweet.1.48 billion tokens in one week, 99.6% of them the model re-reading conversations I never closed. Transcript forensics, the cache math nobody does, and the fix that took my bill from $600 a day to $125. Compression tools would have saved almost nothing.→
- I Cloned Myself Into Slack So I Could Go SkiingAnthropic's Claude Tag put an AI teammate in Slack, behind a Team/Enterprise plan. I had been running the same idea since March, for a simpler reason: two weeks of Easter skiing and no one to cover for me. So I built a stand-in CTO. It's open-source now.→
- Everyone Just Shipped an AI That Watches Your Screen. Almost None of Them Run on Your Machine.At Build 2026 the PC quietly became an agent that sees everything you do. The part nobody's pricing in: your screen is the most sensitive surface you own, and the default architecture streams it to someone else's cloud.→
- Three Tools Want to Cut Your AI Token Bill. They're Fixing Three Different Leaks.Caveman, Headroom, and the proxy I built all promise cheaper agents. Put side by side, they attack three different points in the pipe, and only one of them is aimed at the tokens that actually cost money.→
- Everyone's Burning Millions on AI Coding. I Tried to Fix the Bill, and Built the Wrong Thing.I spent days building a proxy to cut my AI coding bill. The benchmark deleted it, and showed me the one setup change that actually cuts cost by an order of magnitude. It wasn't the model.→
Real systems, running in production.
Voice AI Receptionist→Self-hosted AI voice agent for live operational workflows, turning phone conversations into structured actions and dispatch data.Active client workHouseparty→Local-first multi-agent orchestrator for Codex and Claude Code, with persistent terminals, shared memory, and remote supervision.Open sourceNorse3→A real-time defense layer for AI-powered support. It moderates every message in the live path, enforces policy inline, and gives operators a console to investigate incidents.Case-study readyHannah→Cache-aware proxy for Codex and Claude Code that cuts token waste, improves cache efficiency, and makes session costs measurable.Open sourceAshlock Trading→Systematic trading research platform spanning live options, intraday crypto microstructure, and statistically rigorous strategy validation.Active research + live options
The better the models get, the more the engineering around them matters. The hard part keeps moving up the stack.
VoiceReal-time voice agents, live on the phone.
AgentsInfrastructure for long-running coding agents.
SafetyGuardrails that let models act safely.
MarketsQuantitative trading research and systems.