Let Them Eat Bugs: Practical Showcase of Agentic Issue Resolution
summary
Hud builds a runtime intelligence layer that detects production errors, queries ClickHouse for function-level context, runs opinionated investigation skills and triggers automations (GitHub Actions / Azure pipelines) to open PRs against affected code. Systems integrating Otel/Grafana/Datadog alone fail because agents need sparse high-signal data and code-aligned metrics. Pilot deployments show engineers approve ~70% of suggested fixes and aim for a ~90% threshold before full automation, reducing MTTR and surfacing concrete root causes.
AI-written summary · about
chapters (15)
- 0:00Agentic issue resolution intro
- 2:27Building at Hut with Had
- 3:31Agents and observability tools
- 5:39Three barriers to agentic resolution
- 7:04Language barrier: alerts vs code
- 9:15Runtime intelligence stack
- 11:45Agents iterate with runtime context
- 13:09Human oversight and trust
- 15:36Pull requests with explanations
- 17:03Key lessons from testing
- 18:46Iterating with human feedback
- 20:10Future of observability
- 23:22Risk assessment before production
- 24:45Runtime intelligence next wave
- 25:07Closing and Q&A
related talks
23mFrom Blind Spots to Merged PRs: Continuous Agentic Performance Optimization
30mAI Agents for Incident Management
17mFighting AI with AI
30mArchitecting Integrity: Responsible Software Evolution in the AI Era
35mAgentic Incident Response
31mBeyond the Copilot: A Blueprint for Autonomous Mobile Reliability
20mPremiere
16mAI Agents for Incident Investigation: The Good, The Bad, and The Ugly