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

A paper titled ReTrace presents a method that converts natural-language standard operating procedures and a few logged LLM agent runs into reusable code-driven workflows in which code, not an LLM, selects each next step. On 79 procedures from the authors' production platform, ReTrace workflows reached 93.8% completion versus 96.8% for the agent at 24% lower cost per task, and on τ²-Bench they reached 75.8% task success (agent: 80.6%) while running in 1.3s instead of 170s. Three rounds of repair on training failures lifted k=2 workflows from 66.7% to 81.2%, and on eight SOP-Bench procedures repair from labeled training failures raised ReTrace from 80.6% to 87.8% at 0.40s per task, compared to 85.5% in 46.6s for the agent.

Papers https://huggingface.co/papers ReTrace: Distilling Tool-Using Agents into Planner-Free Workflows Abstract ReTrace converts natural-language SOPs and logged LLM agent runs into reusable code-driven workflows that cut cost and latency while maintaining competitive task success. deepseek-ai/DeepSeek-V4.1-Flash https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash Business process automation runs the same Standard Operating Procedure SOP again and again on new inputs. Large Language Model LLM </papers?q=Large%20Language%20Model%20 LLM agents can follow an SOP written in natural language, but they call an LLM to choose every step of every run, which adds latency and cost at scale. We present ReTrace, which turns an SOP and a few logged agent runs into a reusable workflow in which code, not an LLM, selects the next step. The agent's reasoning, the explanation it writes with each tool call, often states why it took an action; the tool calls https://huggingface.co/papers?q=tool%20calls alone do not. ReTrace extracts the condition behind each step, merges these conditions across runs into a compact rule document, and generates an executable workflow from it. Only this last stage depends on the platform, so rule documents can be compiled for different orchestration platforms. On 79 procedures from our production platform, ReTrace workflows reach 93.8% completion, against 96.8% for the agent, at 24% lower cost per task. On τ^2-Bench, they reach 75.8% task success agent: 80.6% and run in 1.3s instead of 170s; three rounds of repair on training failures lift k{=}2 workflows from 66.7% to 81.2%. On eight SOP-Bench procedures, ReTrace without repair trails few-shot generation https://huggingface.co/papers?q=few-shot%20generation 80.6% vs. 83.6% , but repair from labeled training failures lifts it to 87.8% at 0.40s per task, compared to 85.5% in 46.6s for the agent. Models citing this paper 0 No model linking this paper Datasets citing this paper 0 No dataset linking this paper Spaces citing this paper 0 No Space linking this paper Collections including this paper 0 No Collection including this paper collection https://huggingface.co/new-collection to link it from this page.