The Role of UX in Agentic Software Factories

Tanner Kohler of Nielsen Norman Group finds that Anthropic's playbook for AI-built software gives user experience staff no stage or approval, and urges them to claim the context agents read.

Tanner Kohler is a researcher at Nielsen Norman Group, a user experience research firm. Kohler examines a playbook Anthropic published in August 2026, in which AI agents do most of the work and product owners and engineers approve a chain of Markdown documents. User experience knowledge appears only as context for the agents, and the playbook never mentions user research.

• In the playbook the product owner approves each idea and iterates on the AI-generated mockup, and engineers approve the plan and each release. No stage or approval belongs to user experience staff.
• The test stage compares the built screen with the product owner's mockup, so a flaw introduced early passes unchallenged. Monitoring watches system health, so a confusing feature that people abandon shows no distress.
• Kohler advises practitioners to draft the organisation's user experience skills and attach their names, so that they are brought in when a spec raises a concern.
• Kohler also advises writing down which changes need a person's review, such as changes to navigation, and adding signals such as task abandonment and support ticket spikes to the monitoring.

A design team whose company is moving to AI-built software can decide now who owns the context the agents read and which changes must come to a person. The piece is written by a firm that sells training in this field, and it examines one playbook rather than a team's results.

Derived by Working Surface from the article; more in the Playbook. Source line: Context can encode only what we already know.

31 October 2026: as AI agents produce more of the work, teams move the checking to the people and records closest to the real output.