We write up what we learn shipping agents into production — evals, integration gotchas, cost traps, post-mortems. No thought leadership, no roadmap teasers. Just the parts that were hard, and what we did about them.
The archive is short on purpose — one essay published, more in progress. These are the threads each piece comes from. We post when there is something worth saying, not on a schedule.
How we wire Claude Code around business logic — tools, context windows, and the guardrails that keep an agent from improvising past its remit.
Ingestion, OCR, classification, reconciliation. Moving messy real-world input through deterministic steps where a wrong answer would cost money.
When n8n, Zapier and visual builders earn their keep — and the point at which a code harness becomes the simpler, more honest thing to maintain.
Evals, fingerprinting, idempotency, human-in-the-loop approval. Making failure modes mechanically impossible instead of "we will watch for it".
Cron, queues, Postgres as the source of truth, audit trails. The unglamorous plumbing that decides whether a system survives its first month in production.
Shorter observations from active builds — integration gotchas, cost traps, and the small decisions that turn out to matter more than the architecture diagram.