Harry Kimpel ☁
@harrykimpel
Software craftsman, DevRel @NewRelic; fmr @snyksec, @Microsoft; passion for technology, data nerd; H-D fan; profile linkedin.com/in/harrykimpel/
4.8K Following 2.5K Followers 51K Posts
Latest posts 5
This session grew from my work on Microsoft’s What The Hack — tackling explainability, trust, and unsafe behavior in real deployments.
If you’re heading to PTH, come join me!
What’s the biggest challenge you’re facing when taking AI agents into production?
4/5
If you’re heading to PTH, come join me!
What’s the biggest challenge you’re facing when taking AI agents into production?
4/5
I’ll show how to:
• Trace agent handoffs, model calls, and tool failures
• Evaluate if an answer meets the user’s request
• Detect prompt injection attempts and enforce blocking
• Turn quality checks into CI/CD release gates
3/5
• Trace agent handoffs, model calls, and tool failures
• Evaluate if an answer meets the user’s request
• Detect prompt injection attempts and enforce blocking
• Turn quality checks into CI/CD release gates
3/5
Take AI agents from cool prototype to production with OpenTelemetry
We’ll use a multi-agent travel planner built with Microsoft Agent Framework, OpenTelemetry, and an observability backend to make behavior visible and testable.
2/5
We’ll use a multi-agent travel planner built with Microsoft Agent Framework, OpenTelemetry, and an observability backend to make behavior visible and testable.
2/5
An AI agent can return HTTP 200 and still fail the user.
The tool timed out. The itinerary ignored a constraint. The weather forecast was invented — and the answer still looked convincing.
I’ll dig into this at Porto Tech Hub 🎤
1/5
The tool timed out. The itinerary ignored a constraint. The weather forecast was invented — and the answer still looked convincing.
I’ll dig into this at Porto Tech Hub 🎤
1/5