Open Brain
arXivPaperNeeds Review

ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows

ReTrace turns logged agent runs into executable, planner-free workflows for repeated procedures. It trades a small amount of completion rate for lower cost and much shorter runtimes in the reported evaluations.

arXiv · Nov 10, 2026
Open Source Open MarkdownOpen JSON
Source Summary

ReTrace extracts decision rules from an SOP and logged agent runs, then compiles them into workflows where code selects each next step. Across **79 procedures**, completion was **93.8%** versus 96.8% for the agent, at 24% lower cost. On τ²-Bench, runtime was **1.3 seconds** versus 170 seconds for the agent.

Practical Implication

For repeated, well-defined tasks, consider turning reliable agent traces into rules and workflows instead of asking a model to plan every run. The rule document can be compiled for different orchestration platforms.

Agent-Ready Context
ReTrace extracts decision rules from an SOP and logged agent runs, then compiles them into workflows where code selects each next step. Across **79 procedures**, completion was **93.8%** versus 96.8% for the agent, at 24% lower cost. On τ²-Bench, runtime was **1.3 seconds** versus 170 seconds for the agent.

For repeated, well-defined tasks, consider turning reliable agent traces into rules and workflows instead of asking a model to plan every run. The rule document can be compiled for different orchestration platforms.

The workflow did not match the agent’s completion rate on the production procedures, and results varied by benchmark. On SOP-Bench, repair using labeled training failures raised performance to 87.8%; the paper’s results do not establish how well this approach handles changing procedures.
Connected Context · Feed7 Judgment

ReTrace turns SOPs and logged runs into executable decision rules for repeated procedures, reducing runtime and cost while retaining slightly lower completion on production procedures than the agent. Its results support compiling stable workflows, with procedure changes and benchmark variation remaining important limits.

Can an Agent Debug a Web Corpus? Agentic Auditing of Slovak FineWeb2Both put deterministic, reusable checks after agent-led discovery: ReTrace compiles observed procedures into workflows, while the corpus audit found static sampling more effective for validated selectors after agent exploration.Tribal Dungeons of Global Shipping: AI Agents at Global Scale — Dmitry Buykin, MaerskBoth emphasize executable procedures and replayable traces as production reliability infrastructure; ReTrace adds evidence that logged runs and SOPs can be compiled into lower-cost workflows, while the Maersk account describes the operational correction system around such procedures.Grow the Harness, Not the Context: From Strategy-Free Scaffolds to Reusable Specialist AgentsBoth move recurring agent control into reusable code to reduce model work and cost. ReTrace compiles procedures from SOPs and runs, while Grow the Harness learns reusable control code and reports stable web-agent performance in its tested settings.
Context Map
agentcodingdata#harness-engineering#agent-reliability#tool-use
Uncertainty
The workflow did not match the agent’s completion rate on the production procedures, and results varied by benchmark. On SOP-Bench, repair using labeled training failures raised performance to 87.8%; the paper’s results do not establish how well this approach handles changing procedures.