TeleTune: Evolving Agent Skills From Offline Telemetry
Justin Chih-Yao Chen and Mohit Bansal (UNC Chapel Hill) with Elias Stengel-Eskin, Benjamin Van Durme, Gaurav Verma and colleagues at Microsoft introduce TeleTune, which learns a textual skill library for computer-use agents from offline user telemetry.
Ask this paper
Problem. Telemetry logs record no goals, cannot be replayed, and may interleave several tasks. TeleTune infers a goal per trajectory and uses the logged actions themselves as supervision.
Skill-guided progress. Library edits are proposed from action-prediction errors and kept only if they raise held-out action-prediction accuracy. This offline score tracks the live success rate during optimization.
Results. Average success reaches 77.1% on WorkArena and 80.6% on Online-Mind2Web, 6.7 and 7.7 points above the strongest baseline on each. Under the heaviest task interleaving it stays highest at 68.5%.
Cost. Scoring edits on fixed logs uses 5 to 75 times fewer tokens than validating the same edits with live episodes, and skill optimization and workflow-based demonstration retrieval add to each other.
Abstract
Computer-use agents need to capture procedural knowledge of how people use software. User telemetry offers a scalable source of this knowledge. However, learning reusable skills from these logs requires addressing three challenges: (1) Goal Underspecification, since logs do not record the goal behind each action; (2) Non-Replayability, since past activity cannot be replayed to evaluate skill updates; and (3) Interleaved Trajectories, since logs may mix several tasks without marking their boundaries. To address these, we introduce TeleTune, a framework for learning a textual skill library from offline logs without recorded goals, cannot be replayed during optimization, and may interleave tasks. TeleTune uses action-prediction errors on logged trajectories to propose library edits and keep only those that improve held-out action-prediction accuracy, which we call skill-guided progress. The learned workflows also enable retrieval of demonstrations that cover the subgoals of a new task. At test time, the agent is provided with the learned library and the workflow-based retrieved demonstrations. Experiments on WorkArena and Online-Mind2Web show that TeleTune outperforms random retrieval, Agent Workflow Memory (AWM), and their combination. We find that the best baseline varies by setting, whereas TeleTune achieves average success rates of 77.1% and 80.6%, respectively, improving over the strongest baseline on each benchmark by 6.7% and 7.7%. Under the heaviest perturbation of the WorkArena training data,TeleTune keeps the highest average success rate at 68.5%, 6.3% above the strongest baseline. Our analyses show (1) skill optimization and workflow-based retrieval are complementary, (2) optimizing on fixed logs costs 5 to 75 times fewer tokens than validating the same edits with live episodes, (3) skill-guided progress tracks the live success rate.