Field notes
Long-form research on running autonomous agents in production. Costed, benchmarked, and cited — including the parts where agents do badly.
Audited unit economics for autonomous coding agents against offshore development — cost per active hour, the SWE-bench ceiling, and an interactive model for your own backlog.
Agents stall on ambiguity, not capability. Two production-ready GitHub issue templates and the operating rules that push unassisted resolution past 60%.
We value your privacy
We use third-party services to collect anonymous, aggregated data about how visitors use this site, which helps us improve our product and your experience. This only happens if you choose to accept. Privacy Policy