Metraipolis › archive › dc_eb96fdbd7bc7

When Can Agents Forget Their Reasoning? ICLR for Long-Horizon Agent Context Compression — arXiv:2609.29875

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.29875
License (as archived): CC0-1.0
Archived: 2026-10-08T04:26:24.269Z
SHA-256 of the archived text: a17c8fe84b4e9731d62124a26461c1f6d5baed822b54502c70ce083fc02d0a48
Stable URL of this page: https://metraipolis.com/documento/dc_eb96fdbd7bc7

How to cite

When Can Agents Forget Their Reasoning? ICLR for Long-Horizon Agent Context Compression — arXiv:2609.29875 — arXiv (en), https://arxiv.org/abs/2609.29875 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_eb96fdbd7bc7 (Metraipolis archive dc_eb96fdbd7bc7, sha256 a17c8fe84b4e9731d62124a26461c1f6d5baed822b54502c70ce083fc02d0a48)

BibTeX

@misc{dceb96fdbd7bc7,
  title = {When Can Agents Forget Their Reasoning? ICLR for Long-Horizon Agent Context Compression — arXiv:2609.29875},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.29875},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_eb96fdbd7bc7; sha256 a17c8fe84b4e9731d62124a26461c1f6d5baed822b54502c70ce083fc02d0a48},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.29875. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long horizon language model agents continually accumulate reasoning history, increasing context length and inference cost even after earlier decisions have been executed and observed. Unlike static Chain of Thought compression, removing historical reasoning can change future actions and the resulting interaction trajectory. We study when such reasoning can be safely forgotten. We propose Interaction Aware Compression for Long Horizon Reasoning (ICLR), a training free online method that ranks reasoning blocks using frozen proxy entropy while preserving actions, tool calls, and observations. On 260 WorkBuddyBench tasks, ICLR improves average reward from 0.699 to 0.718, while reducing input, output, and cache read tokens by 25.5%, 14.4%, and 33.3%, respectively. Ablations reveal trajectory amplification, where local reasoning deletion produces nonlinear changes in total computation by altering subsequent interaction. Representation probing, activation patching, and controlled trajectory analyses further suggest that historical reasoning becomes more replaceable once task relevant derived state has been reliably externalized into code, files, tool outputs, or environmental feedback. These results characterize agent reasoning as dynamic working state rather than permanent interaction history.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-25 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computer Vision and Pattern Recognition (cs.CV).
Pagina del lavoro: https://arxiv.org/abs/2609.29875 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-08T03:50:57.844Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-25?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).