Tool-Making and Self-Evolving LLM Agents in Low-Latency Systems
Quick Answer
This paper shows that A novel tool-making pipeline for LLM agents reduces latency by 42% and error rates by 53% in a Fulfillment Center alarm-triage system.
Quick Take
A novel tool-making pipeline for LLM agents reduces latency by 42% and error rates by 53% in a Fulfillment Center alarm-triage system. By compiling repeated procedural steps into validated tools, the system enhances reliability and operational simplicity, demonstrating the potential of self-evolving agents in industrial applications.
Key Points
- Tool-making pipeline compiles repeated SOP steps into validated tools before deployment.
- Production agent reduces p50 latency by 42% and end-to-end error rates by 53%.
- Direct tool calls simplify architecture, further reducing latency by 62% in controlled tests.
- Versioned tools improve auditability and expose specification gaps and data drift.
- Demonstrates the effectiveness of self-evolving agents in enhancing LLM system performance.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Production LLM agents often waste latency and reliability by regenerating code for the same procedural steps on every request. We replace this inference-time coding loop with an agentic tool-making pipeline that compiles repeated SOP steps into validated, versioned tools before deployment. The tool-maker grounds synthesis in the live environment as it collects execution traces, observes backend schemas and values, generates candidate tools, and repairs them against labeled cases. At runtime, the production agent calls these tools directly and falls back to code generation only when needed. We deploy the approach in a Fulfillment Center alarm-triage system, where an agent diagnoses alarms against a 44-node SOP over heterogeneous metric backends. In production, tool calls reduce p50 latency by 42%. On 1,500 historical alarms, they reduce end-to-end error rate by up to 53% by suppressing run-to-run variance in repeated steps. Because tools return compact structured verdicts, they also enable a simpler direct-call architecture, reducing p50 latency by a further 62% in a controlled ablation. Versioned tools also improve auditability and expose specification gaps and upstream data drift. Our results show that self-evolving agents can make industrial LLM systems faster, more reliable, and easier to operate.
| Comments: | Preprint |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG); Software Engineering (cs.SE) |
| Cite as: | arXiv:2607.08010 [cs.CL] |
| (or arXiv:2607.08010v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.08010 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kalle Kujanpää [view email]
[v1]
Thu, 9 Jul 2026 00:27:13 UTC (74 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Quantifying Prior Dominance in Systems
The study introduces the Normalized Context Utilization (NCU) metric to evaluate Retrieval-Augmented Generation (RAG) systems, revealing that Small Language Models (SLMs) outperform larger models in factual extraction. The findings indicate that traditional scaling laws yield diminishing returns, with a commercial API frequently failing against adversarial evidence due to systemic confidence collapse.