# Preprint reports graph-based instructions improved long-running AI-agent reliability

_Published Tuesday, September 22, 2026 at 6:19 PM EDT · Science, AI · Latest · Tier 2 — Notable_

A preprint reports that a graph-based system for updating an AI agent's persistent instructions raised strict reliability on a held-out telecom task from 0.091 for Gemini 2.5 Flash without adaptation to 0.673, plus or minus 0.136. The proposed GRACE method stores instructions as a typed semantic graph, checks changes near the affected nodes and converts accepted updates back into deployment text.

Across five replications, the final result also exceeded a flat-text updating baseline of 0.191, plus or minus 0.051. The findings come from one controlled harness and have not undergone peer review.

## Sources

- [cs.CL updates on arXiv.org](https://arxiv.org/abs/2607.09175)

---
Canonical: https://techandbusiness.org/newswire/rv7C8_J7hdaDxubPQe2OFR
Published: 2026-09-22T22:19:55.644Z
Story chronology: 2026-09-22T04:00:00.000Z
Retrieved: 2026-09-23T00:37:06.350Z
Publisher: Tech & Business (techandbusiness.org)