Subscription billing foundation, end to end, in 48 hours — against a 2 to 3 week industry baseline.
Context. The next product launch was blocked on a revenue path that did not exist. No billing, no schema, no deploy story.
Move. Stood it up end to end: payment-processor integration, schema migrations, 17 automated tests across 4 critical paths, deploy automation, and an on-call runbook. The agent fleet wrote the migration, the route, and the tests in parallel while I reviewed the synthesis.
Result. Live in 48 hours. The revenue path for the next launch was unblocked, with tests and a runbook from day one.
5 to 10 agents in parallel. Roughly 3x throughput. 60+ PRs merged in six weeks.
Context. A four-person team trying to ship at the pace of a much larger one. Solo coding could not keep up, and AI assistance one prompt at a time did not either.
Move. Built an OS-agnostic agent fleet: an orchestrator that routes a task to the specialist whose intent matches highest, runs independent work in parallel under git-worktree isolation, and synthesizes a single go/no-go. Around 50 in-house skills, branch-safety hooks, and a routing contract tested in CI.
Result. Roughly tripled per-engineer throughput. 60+ pull requests merged in six weeks, frontend to infrastructure, at an average open-to-merge time under 18 hours. The fleet is how every feature ships, not a side experiment. The engine that built this site and its video is open source.
270+ production workflows at ~99.5% job-success, thousands of loan applications a week.
Context. Credit-union lenders needed lending automations that could not silently drop or double-process an application.
Move. Built and ran 270+ production workflows on a multi-tenant Node/AWS pipeline: retries, idempotent re-runs, per-client data isolation. As a Solutions Engineer I read the angry customer email, reproduced it in code, and shipped the patch.
Result. Thousands of loan applications a week at about 99.5% job-success across 3+ credit-union clients, with customer-issue resolution cut from days to hours.