Building trust in AI-powered support at Microsoft: Our Employee Self-Service agent story

Microsoft's IT support agent won employees over only after its team wrote the agent's behaviour for each trust breakdown, including a clear handoff to a person.

Microsoft launched its Employee Self-Service agent, which answers IT and HR questions, to more than 300,000 staff, but early use was weak. Research with Designit, the design arm of Wipro, found that people felt trapped in loops and could not see how to reach a person. The team defined how the agent should behave in each of those moments, and use of the agent rose.

• Employees were offered fixes they had already tried, lost context when they had to repeat themselves and could not see the next step.
• The team paired each documented trust breakdown with a specific behaviour, such as acknowledging earlier steps and stating when the agent is unsure.
• The agent tells people it will pass the case to a human support agent if it cannot fix the problem.
• The share of employees who start with the agent rose from 27% to 50%, and support tickets fell by 30%.

Design teams can treat the handoff to a person and the admission of uncertainty as designed states of an AI agent, written before launch. This is an inference and is not the author's claim. The piece is Microsoft's account of its own product.

Derived by Working Surface from the article; more in the Playbook. Source line: It’s not about speed or performance; it’s more about continuity and relevance.

9 October 2026: when AI agents take on routine work, teams write down in advance what a person still decides, put those rules where the agent reads them, and keep the people who approve able to understand the work.