{"slug": "what-context-does-a-coding-agent-actually-need-to-act", "title": "What Context Does a Coding Agent Actually Need to Act?", "summary": "A new study on arXiv finds that coding agents need minimal context to edit code: the signal is in the code being edited itself, not in natural-language summaries or surrounding files. Compressed context matches whole files at a third of the tokens, and temperature-0 API inference introduces a ~9% noise floor on SWE-bench Verified.", "body_md": "arXiv:2607.09691v1 Announce Type: new\nAbstract: A modern coding agent can hold an entire repository in its context window. Most of its reading is wasted -- and the interesting question is not how much context an agent can use, but what it actually \\emph{needs}. We study that question at the moment it matters most: when the agent must \\emph{edit} code. Separating \\emph{finding} the work site from \\emph{acting} on it, we hold localization fixed with an oracle, vary only how the code is represented, and score context against real issue resolution on SWE-bench Verified. The answer is starkly minimal. The signal lives in the code being edited itself: natural-language summaries of it answer almost none of the behavioral questions that the source answers ($4/45$ vs.\\ $27/45$, held-out repositories, independent judge), and the gap belongs to the representation, not the summarizer -- a frontier model's summaries score exactly as poorly as a 3B model's. The surrounding context hardly matters either: across every multi-file instance in Verified, under a protocol frozen before any data, rendering a file's remainder as UML skeletons and signatures resolves no more issues than deleting that remainder outright ($N{=}70$, exact McNemar $p{=}0.75$). That was our registered hypothesis, and it failed. Compressed context, meanwhile, matches whole files at a third of the tokens: a resolved issue costs $19$K context tokens, not $94$K. The instrument also yielded a finding the field should keep: temperature-0 API inference flips ${\\sim}9\\%$ of per-instance outcomes between byte-identical runs. That is a noise floor under every small effect reported on this benchmark, including ours. We release the instrument -- gold-validated environments, per-instance proof that every reference edit is expressible from every arm's context, deterministic patch construction, and pre-registered hypotheses whose nulls we publish.", "url": "https://wpnews.pro/news/what-context-does-a-coding-agent-actually-need-to-act", "canonical_source": "https://arxiv.org/abs/2607.09691", "published_at": "2026-07-14 04:00:00+00:00", "updated_at": "2026-07-14 04:22:19.766800+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-agents", "developer-tools"], "entities": ["arXiv", "SWE-bench Verified"], "alternates": {"html": "https://wpnews.pro/news/what-context-does-a-coding-agent-actually-need-to-act", "markdown": "https://wpnews.pro/news/what-context-does-a-coding-agent-actually-need-to-act.md", "text": "https://wpnews.pro/news/what-context-does-a-coding-agent-actually-need-to-act.txt", "jsonld": "https://wpnews.pro/news/what-context-does-a-coding-agent-actually-need-to-act.jsonld"}}