Essays, ideas, and thinking out loud.
Building a data science agent that was actually good meant building two layers of memory first. The orchestration I thought was the product turned out to be scaffolding — and better models ate it. What was underneath generalized to any agent at all.
Sam Altman says we're entering the fast fashion era of SaaS. He's right. GPT is fast fashion. But fast fashion only works if the factory is solid. Nobody's building the factory.
A short story about a father, five copies of his daughter, and the cost of choosing which parts of someone to keep. Plus the AI production stack behind the film.
Client isolation, reproducible workflows, audit trails — the infrastructure that makes agent-driven data workflows actually work in production. Architecture decisions from building a 26-agent swarm that handles multi-client data without bleed.
On the range of Blackness, the absurdity of flattening it, and why that logic applies to everything we build.