Simon Willison’s discomfort with not reviewing every line of agent-written code captures the real enterprise blocker. The issue is not whether agents can produce useful work. It is who remains accountable when that work becomes a dependency inside production systems.
Serious adoption requires evidence: tests, diffs, logs, provenance, review trails, rollback paths, and scoped authority. A manager or engineer needs to know what was changed, what assumptions were made, and how failure can be contained. Bigger demos do not answer those questions.
For vertical AI builders, the same principle applies beyond code. Claims, care plans, dispatch decisions, quotes, and compliance documents need inspectable work products. Autonomy becomes valuable only when the organization can assign responsibility, verify output, and recover cleanly from mistakes.
Grey Haven read: The enterprise agent question is accountability. Autonomy becomes useful only when the organization can prove what changed, who approved it, what evidence was used, and how failure can be contained.
Reader guidance: Treat the item as a signal about operating readiness rather than headline momentum. The useful question is which control, workflow, or evidence requirement must be strengthened before the same pattern reaches a production system with customers, regulators, or internal teams depending on it.
Operator guidance: Require tests, diffs, provenance, review trails, and rollback before agents create production dependencies. Watch workflows where agent output becomes authoritative without a named accountable owner.
Source/provenance: Grey Haven Signal update linear-ASM-101-2, based on the existing backfill draft and source detail Hacker News discussion: https://news.ycombinator.com/item?id=48037128.
