Three-agent harnesses: Anthropic’s blueprint for long‑running full‑stack AI development

The wiki-as-memory pattern is one answer. Not the only one, but a structurally honest one.

Read with a highlighter. There are a lot of pages. Most of them earn their place.

This is what governance as infrastructure looks like in practice.

A control plane is active — it governs what can be used, by whom, under what conditions, and with what audit trail.

That gap is "harness engineering," and it's where the real work of building production-grade agents happens

Agent observability is a distributed systems problem, and OpenTelemetry is the backbone

That works for focused problems, not agents