# CORVUS Attacks the Coding Agent's Real Bottleneck: an Append-Only Trajectory That Bloats Context

The method restructures how agents accumulate reasoning, tool calls, and results, targeting the context growth that makes long agent runs slow and expensive.

- Published: 2026-07-28T05:31:39.988Z
- Canonical: https://polylog.news/ai/2026-07-28/corvus-attacks-the-coding-agent-s-real-bottleneck-an-append
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [arXiv (cs.LG)](https://arxiv.org/abs/2607.22711)

A paper called CORVUS targets a structural weakness in how large language model coding agents work. Today's agents build a trajectory that appends every reasoning step, tool call, and result in order. That append-only design ties context gr…

This story is for subscribers. Read it in full at https://polylog.news/ai/2026-07-28/corvus-attacks-the-coding-agent-s-real-bottleneck-an-append (subscription information: https://polylog.news/pricing).