Tanner Kohler of Nielsen Norman Group analysed the agent-driven development playbooks from Anthropic, Google and OpenAI and found that all three sideline UX. In Anthropic’s AI-native SDLC playbook, UX policies become skills the agent reads, and the Maintain stage monitors CI failures, 5xx errors and PR cycle time but nothing about how people use the product.
The episode also takes the counterpoint seriously: the playbooks do mention design review and early customer testing, and the sources analyse what the text prescribes, not what teams actually do. Kohler’s modest fixes include user-value acceptance criteria and “UX distress signals” in monitoring.