PMs design products for AI agents, not just humans

Pascal Akunne, writing on the LogRocket Blog, argues that product managers now design for two users: the human who sets goals and the AI agent that carries them out. His case is that agents fail where humans compensate; ambiguous labels like "health score," inconsistent API states such as "active" versus "live," and generic error messages that a person would shrug off and retry. Akunne's fix is structural rather than cosmetic: expose capabilities through documented APIs rather than burying them in the UI, keep behavior and schemas stable over time, return machine-readable errors that specify whether retrying will help, and log agent actions for audit. He also wants PRDs rewritten to include agent-specific user stories and metrics like retry rates and escalation frequency, treating reliability itself as the new usability.