AI-assisted delivery, on a platform I own end to end

Evidence

1. Designed and built an agentic software development lifecycle that orchestrates AI coding agents across the delivery pipeline, backed by MCP servers over documentation repositories so agents work from current, authoritative sources rather than from memory.
2. Built “gotcha”: an agent skill that runs after every plan execution, captures what failed and why, and writes it into a structured knowledge base that later planning loads — so the same dead end is not walked twice.
3. Made the guardrails machine-asserted rather than documented. The knowledge index and its entries are checked for parity in both directions, by a script whose selftest deliberately mutates the repo to prove the check can fail.
4. Prototyped the entire workflow on my own projects first, and adopted it at work only once it had earned its place.
5. Tiered the models against a flat subscription, with a session precondition gate that catches silent fallback to metered billing before it costs anything.

Numbers

The full timeline, the stack and what I am building now are on thefront page.