Everything I Learned About Harness Engineering and AI Factories in San Francisco (April 2026)

A field report for founders, CTOs, CPOs, and engineering leaders describing how teams in San Francisco build with AI agents today: definitions (model, harness, agent), hot product and engineering takeaways (faster implementation, shifted PM role, verification over advice), a seven-layer minimal AI factory architecture, concrete harness artifacts (AGENTS.md, SKILL.md, plugins), tooling and runtime patterns for parallel agent sessions and isolation, review/verification practices, metrics and cadence for adoption, and a practical playbook for the first 30 days of adoption. The piece emphasizes treating models as task-routing components, preventing recurring failure modes via deterministic checks, and balancing automation with human review and product judgment.