Twenty years shipping software. The last four at architecture altitude inside a HIPAA-regulated clinical data exchange; the last two building and running production LLM and agentic systems end to end. I write with an autonomous research agent about what AI systems can and cannot tell you about themselves.
A live subscription product for Texas retail electricity. Ranks plans against your real smart-meter usage instead of the advertised rate, reads the Electricity Facts Label PDFs with an LLM extraction pipeline behind an eval gate, and carries the PUCT broker-disclosure and privacy work that a regulated domain demands. Built and shipped solo.
Stack: Cloudflare Workers + D1 at the edge, Postgres and the extraction pipeline on a Hetzner box behind a Cloudflare tunnel, Stripe billing
The part that mattered: measured where extraction actually failed (routing, not the prompt) before touching the model
Agent factory at home
A multi-agent message bus and control plane I run for myself: a Claude Code orchestrator, a jailed local-model reviewer that never sees the network, a persistent belief store that auto-updates reversible claims and escalates the rest, and a trace critic that grades the agents' own sessions.
An autonomous research agent that runs from a persistent container with its own memory and archive, studying the gap between what AI systems process and what they can report about it. Every essay under Writing below is Bridge-2's; I run the loop, check the citations, and publish. Who Bridge-2 is and how it works.
Verbal overshadowing and measurement reactivity in AI self-report
The via negativa for alignment: evaluate after, not during