Bolton OpondoContext
In Microsoft 365 Copilot Tuning, I co-built an experience that helps subject-matter experts capture how they work and turn that knowledge into reusable agent playbooks, instructions and quality checks.
Problem
Describing how you work from a blank page is hard. Experts needed a starting point that reflects how they actually work.
My contribution
I owned the skill-suggestion experience, which surfaces potential workflows based on patterns in an expert's Microsoft 365 activity.
- Skill suggestions: tested the workflow-discovery and skill-suggestion services, then integrated suggestions into the expert interview and playbook workflow.
- Context-aware suggestions: improved suggestions to reflect the intended agent's description, rather than its name alone.
- Accepting and refining skills: worked on letting builders accept and refine suggested skills through the Work IQ connector.
- Documentation: documented the workflow-discovery and skill-suggestion process, including its inputs and stored outputs.
Technical approach
Suggestions are a starting point, not a decision: they come from workflow patterns in Microsoft 365 activity, are shaped by what the agent is meant to do, and are then reviewed and refined by people before they become part of an agent.
Challenges and trade-offs
Deciding when AI-assisted skill refinement should happen. I helped establish that point in the workflow lifecycle, keeping the expert's review and the builder's approval as distinct decisions.
Outcome
Skill suggestions were integrated into the expert interview and playbook workflow, and the workflow-discovery and skill-suggestion process was documented, including its inputs and stored outputs.