OAT uses 100 successful trajectories to debug failing AI agents without failure labels
TheTuringPost · x · 2026-07-22
OAT flags failing agent steps without failure training
A new paper proposes OAT, a method for debugging failed AI agents without training on failures. It learns the pattern of successful trajectories and then identifies the steps in a failed run that deviate from the learned path.
Key points:
- Trained on only 100 successful trajectories
- Uses no failure examples and no step-level labels
- Reported to be 200–5,000× faster than prompting-based attribution
- Achieves +20% F1 in-domain and +7% OOD
The main idea is a shift from asking another LLM to explain what went wrong, toward cheap anomaly detection over the execution trace. The paper also notes an important caveat: unusual steps are evidence, but they do not prove causation.
More from coding & agent
- Grok Build turns one prompt into a full ARPG with AI-generated game assets — tetsuoai · 2026-07-22
- A dad built a controller-ready game in two hours with Grok 4.5 — minchoi · 2026-07-22
- Atomic-Chat pitches a fully offline open-source ChatGPT alternative — rohanpaul_ai · 2026-07-22
- ComfyUI Wan dance test renders a 30-second clip in 70 minutes — tostane · 2026-07-22
- Claude Code adds iOS Simulator control for side-by-side mobile testing — xiaohu · 2026-07-22
- Salesforce says enterprise agents are moving from loops to graph routing — msrivastav13 · 2026-07-22