Copilot Studio adds stronger evaluation and review tools
Diesen Beitrag auf Deutsch lesen
Copilot Studio’s roundup confirms GA evaluation and review features, while self-learning and reusable Skills catalog capabilities remain roadmap items.
TL;DR
Copilot Studio’s September roundup confirms Foundry IQ integration and the Review panel as generally available. Evaluations now cover task completion, tool accuracy, safety and latency, with reusable custom graders. Self-learning and the Skills catalog remain in development with October general availability targets.
Original by Josh Cook, on Flow Alt Delete. Read the original
This is our own summary, not a republication or full translation.
Why it matters
- Makers: Use the Copilot Studio Review panel to find agents that were never evaluated or changed models, then rerun checks for task completion, tool accuracy, safety and latency.
- Makers: Give reviewers the Evaluation Viewer role where appropriate so evaluation results can be checked separately from agent building.
- Leadership/Business: Treat self-learning and the Skills catalog as in development, not shipped capabilities, and plan pilots around maker review, testing and publishing.
Frequently asked questions
What does the Copilot Studio Review panel flag?
The Review panel flags agents that have never been evaluated or whose model changed after evaluation. It also catches missing channel or trigger setup.
Which measures can Copilot Studio evaluations report?
Copilot Studio evaluations report task completion, tool accuracy, safety and latency results. Reusable custom graders can support repeated checks.
What is the status of Copilot Studio self-learning and Skills catalog?
Both roadmap items remain In development with October GA targets. Self-learning recommends improvements from completed runs, while the Skills catalog supports reuse of approved skills.