Saved in:
Bibliographic Details
Main Authors: Dennis, Simon, Patil, Rivaan, Shabahang, Kevin, Guo, Hao
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.22502
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910246177341440
author Dennis, Simon
Patil, Rivaan
Shabahang, Kevin
Guo, Hao
author_facet Dennis, Simon
Patil, Rivaan
Shabahang, Kevin
Guo, Hao
contents Agent orchestration frameworks have proliferated, collectively exceeding 290,000 GitHub stars across LangGraph, CrewAI, Google ADK, OpenAI Agents SDK, Semantic Kernel, Strands, and LlamaIndex. All follow the same pattern: an external orchestrator above the LLM, injecting instructions and routing decisions every turn. Recent work has shown this architecture is dominated for procedural tasks by simply providing the procedure in a frontier model's system prompt [Dennis et al., 2026a], at the cost of consuming the context window, requiring a frontier model for every conversation, and exposing proprietary procedures to third-party providers. Compiling the procedure into the weights of a small fine-tuned model -- creating a subterranean agent -- should resolve all of these concerns, and prior work (SimpleTOD, FireAct, SynTOD, WorkflowLLM, Agent Lumos) has shown the technique works. Yet developer adoption has overwhelmingly favored orchestration. We identify three perceived barriers and address each empirically across travel booking (14 nodes), Zoom support (14 nodes, product-specific knowledge), and insurance claims (55 nodes, 6 decision hubs).
format Preprint
id arxiv_https___arxiv_org_abs_2605_22502
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Compiling Agentic Workflows into LLM Weights: Near-Frontier Quality at Two Orders of Magnitude Less Cost
Dennis, Simon
Patil, Rivaan
Shabahang, Kevin
Guo, Hao
Artificial Intelligence
Machine Learning
Agent orchestration frameworks have proliferated, collectively exceeding 290,000 GitHub stars across LangGraph, CrewAI, Google ADK, OpenAI Agents SDK, Semantic Kernel, Strands, and LlamaIndex. All follow the same pattern: an external orchestrator above the LLM, injecting instructions and routing decisions every turn. Recent work has shown this architecture is dominated for procedural tasks by simply providing the procedure in a frontier model's system prompt [Dennis et al., 2026a], at the cost of consuming the context window, requiring a frontier model for every conversation, and exposing proprietary procedures to third-party providers. Compiling the procedure into the weights of a small fine-tuned model -- creating a subterranean agent -- should resolve all of these concerns, and prior work (SimpleTOD, FireAct, SynTOD, WorkflowLLM, Agent Lumos) has shown the technique works. Yet developer adoption has overwhelmingly favored orchestration. We identify three perceived barriers and address each empirically across travel booking (14 nodes), Zoom support (14 nodes, product-specific knowledge), and insurance claims (55 nodes, 6 decision hubs).
title Compiling Agentic Workflows into LLM Weights: Near-Frontier Quality at Two Orders of Magnitude Less Cost
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2605.22502