Answering Developer Questions with Annotated Agent-Discovered Program Traces
summary
Trailblazer is a system that uses an agent-guided exploration combining static analysis and IDE operations like go-to-definition and find-all-references to iteratively trace program flow, then emits an answer digest with replayable annotated traces that step through the actual path and explain each jump. A lab study with 20 participants shows significantly faster task completion, lower reported mental demand, and quicker solutions on Parsons ordering tasks, demonstrating the traces improve answerability and flow familiarity.
AI-written summary · about
chapters (7)
- 0:00Introduction and motivation
- 1:10Limitations of current methods
- 2:55Introducing Trailblazer tool
- 4:19How annotated traces work
- 5:45Study design and findings
- 8:16Summary and future directions
- 8:57Q&A session
related talks
15mAdaptable Traces for Program Explanations
01h 20mMind the Gap (In your Agent Observability)
45mMap Your App for AI
43mEval-Driven Development: Mastering Agentic Tracing and Expert-Aligned Judges With MLf...
33mWhat Did My Agent Do? Observability and Accountability for AI Agents
16mGraph Traversal Agent for Code Debugging
21mHow agent o11y differs from traditional o11y
34mVisualising program execution