Intel / article

AutoFyn Technical Report: Non-Parametric Expert Iteration for Long-Horizon Agents

Source: arxiv.org
Author: Adib Hasan, Daniel Schaffield, Akashnil Dutta, Tarik Adnan Moon
Date: Published Sep 9, 2026

Context

AutoFyn describes an system that carries lessons between attempts without changing the underlying model's weights. Each round starts a fresh model session. Saved files, reports and repository state bring forward what the next attempt needs, while a verifier supplies task-specific feedback on the work. Technical-report abstract.

Within a round, an orchestrator coordinates specialized agents exploring alternative approaches. Verified results then update the persistent information used in the following round. The mechanism makes accumulated external state part of the agent's adaptation: a fresh conversation can still begin with evidence from earlier attempts, provided that evidence was checked and explicitly preserved.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.

Key quotes

“We introduce AutoFyn, an agent harness inspired by the Expert Iteration algorithm, adapting a frozen model across many rounds by updating persistent state from verified reward signals rather than model weights.”

“On the six fresh problems of the 2026 International Mathematical Olympiad, every model with room to improve scores higher under AutoFyn than in its provider's own coding agent.”

“AutoFyn also built the top-ranked agent on the Spider 2.0 dbt benchmark, and has produced 16 maintainer-confirmed vulnerability advisories in this http URL, MetaMask, pnpm, Warp, LiteLLM, Langflow, and Open WebUI.”

Read the source arxiv.org