When prompting is not enough for your agent @MicrosoftDeveloper
When prompting is not enough for your agent  @MicrosoftDeveloper
Uploaded October 2026 | Updated November 2026, 11 hours ago
https://aka.ms/foundry-portal
https://aka.ms/InsideMicrosoftFoundryPlaylist

You've rewritten Sparkles' instructions five times, and it still misses details or confirms orders too early. Another prompt tweak is just another guess. You use Agent Optimizer in Microsoft Foundry to turn that guesswork into a measured improvement loop. Start with a representative evaluation dataset and clear success criteria, score the current agent, then let the optimizer generate and evaluate candidate instructions. Compare the results against your baseline, review exactly what changed, and check the strongest candidate on held-out cases before shipping a new agent version. By the end, you'll know how to improve your agent's configuration, not retrain its model, with evidence that the changes actually help.

0:00 - Fix Inconsistent Agent Responses
0:36 - Introducing Agent Optimization
1:07 - Create a New Optimization
1:41 - Configure the Optimization Approach
2:14 - Select Candidate and Evaluator Models
2:47 - Use Production Traces as Data
3:20 - Review Estimated Optimization Costs
3:54 - Run the Agent Optimizer
4:29 - Compare Baseline and Candidates
5:02 - Analyze Quality and Latency
5:35 - Review Evaluation Scores
6:08 - Inspect Passed and Failed Tests
6:40 - Find Sensitive Information Disclosures
7:14 - Improve Your Agent from Results
7:47 - Agent Optimizer Recap and Next Steps
When prompting is not enough for your agentSecure AI Agents in Azure: AI Gateway, Tools, and TrustTrailer: Azure SQL Database FoundationsData Streaming for Modern Applications in SQL Server 2025, Azure SQL, and Fabric SQL | Data ExposedWhats your favorite new GitHub Copilot feature?Oct 8 Webinar on Copilot Studio!TypeScript 7 Is Here (And Its 10× Faster)🎙️Voices from the Skies: Build Your Own Personal Assistant with GitHub Copilot AppTurn Meeting Transcripts into Project Plans with GitHub Copilot Agents | MVP UnpluggedBuild an AI agent without leaving VS Code!MCP Live! | A half-day livestream about the latest in MCP🎖️ Captains Training: Master Copilot Studio | Use the New Harness, MCPs and More
Microsoft Developer |

When prompting is not enough for your agent

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER