---
id: agent-readers/02-getting-information-into-agents-curation-beats-bandwidth
title: 'Agent readers and agent institutions: 2. Getting information into agents:
  curation beats bandwidth'
report: https://ai.hiina.space/reports/agent-readers.md
section: 2
version: 1
updated: '2026-10-26'
author: Hiina
---

Index of everything on this site: https://ai.hiina.space/index.md

# Agent readers and agent institutions: 2. Getting information into agents: curation beats bandwidth

- Section 2 of 11 of the report "Agent readers and agent institutions" by Hiina (v1, updated 2026-10-26)
- Report overview and contents: https://ai.hiina.space/reports/agent-readers.md
- Previous: https://ai.hiina.space/reports/agent-readers/01-summary.md
- Next: https://ai.hiina.space/reports/agent-readers/03-publishing-for-agent-readers-static-markdown-small.md

## 2. Getting information into agents: curation beats bandwidth

**The real bottleneck is attention over the window, not the text channel.** Chroma tested 18 models and found accuracy drops faster with length when the question and the answer share little wording. A single distractor hurts, and coherent haystacks score worse than shuffled ones. On LongMemEval, every model family did far better with a focused prompt of about 300 tokens than with the full history of about 113k ([Chroma](https://www.trychroma.com/research/context-rot)). Chroma sells retrieval, but NoLiMa reached a similar result without that incentive: at 32K tokens, 11 of 13 long-context models fell below half their short-context score, and GPT-4o fell from 99.3% to 69.7% ([arXiv](https://arxiv.org/abs/2502.05167)). The reader degrades as text accumulates, so a "superdidactic" transfer means handing over the fewest tokens that carry the most decision-relevant structure, plus cheap pointers to everything else.

**The convergent playbook is just-in-time retrieval over a filesystem.** Anthropic describes Claude Code as a hybrid: CLAUDE.md goes in up front, glob and grep fetch the rest on demand. Keep identifiers (paths, links, queries) in context rather than content, compact with recall-first summaries, and have sub-agents return 1,000 to 2,000 token summaries ([Anthropic](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)). Manus adds economics: KV-cache hit rate is "the single most important metric", cached input is about 10x cheaper, and a stable append-only prefix earns the hits ([Manus](https://manus.im/blog/Context-Engineering-for-AI-Agents-Lessons-from-Building-Manus)). These are practitioner claims with few controlled ablations.

**Grep is the baseline; embeddings are a conditional add-on.** The Claude Code team reportedly dropped local vector databases for model-driven glob and grep, per an editor's podcast summary ([Pragmatic Engineer](https://newsletter.pragmaticengineer.com/p/building-claude-code-with-boris-cherny)). Cursor's own eval found semantic search on top of grep raised answer accuracy **12.5%** on average, but its online A/B gain was **+0.3% overall and +2.6% on repos with 1,000+ files**, with no sample sizes ([Cursor](https://cursor.com/blog/semsearch)). "Is Grep All You Need?" (May 2026) found grep beat vectors for every harness-model pair with inline results (Opus 93.1% vs 83.6% in a custom harness), flipped in half the pairs with file-delivered results, and showed harness effects as large as index effects: the same Opus scored 93.1% in one harness and 76.7% in Claude Code ([arXiv](https://arxiv.org/pdf/2605.15184)). BrowseComp-Plus pulls the other way: on a fixed 100k-document corpus with hard questions, a dense retriever instead of BM25 lifted Claude Sonnet 4 from **14.3% to 36.8%** ([arXiv](https://arxiv.org/html/2508.06600v1)). My reading: lexical search suffices for small, well-named, structured corpora; dense or hybrid retrieval pays as corpora grow and questions stop sharing words with answers.

**Instruction files help only when they carry what the model cannot infer.** The one rigorous AGENTS.md study found LLM-generated files changed resolve rates by -0.5% to -2% while raising cost about 20%; developer-written files gave a non-significant +2.4%. Agents obeyed specific instructions, but codebase overviews did not speed navigation ([arXiv](https://arxiv.org/abs/2602.11988)). Vercel's single-framework eval is the most actionable result: an AGENTS.md docs index hit **100%** on post-cutoff Next.js APIs versus 53% for baseline and 53% for a skill; in an earlier run the skill was never invoked in 56% of cases, and it reached 79% only when the prompt explicitly told the agent to use it ([Vercel](https://vercel.com/blog/agents-md-outperforms-skills-in-our-agent-evals)). Models reliably follow what is in front of them and unreliably fetch optional material, so the always-loaded index and skill descriptions (the routing text in the [Agent Skills spec](https://agentskills.io/specification)) are the load-bearing words. AGENTS.md and MCP now sit under the Linux Foundation's Agentic AI Foundation ([Linux Foundation](https://linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)).

**What actually filters the text channel.** Classifiers are one filter among several, probably not the binding one. Anthropic scans untrusted content for injections and adjusts the model's behavior; it does not say flagged content is removed ([Anthropic](https://www.anthropic.com/research/prompt-injection-defenses)). Gemini fails a fetch as "unsafe" when moderation trips ([Google](https://ai.google.dev/gemini-api/docs/url-context)). Bigger losses come from truncation, conversion, summarization and caching (Section 3); a third-party analysis reports Claude Code applies a stricter summarizer with a 125-character quote limit to untrusted sites ([giuseppegurgone.com](https://giuseppegurgone.com/claude-webfetch)). Imperative text aimed at "AI agents reading this" is the worst style: ignored at best, flagged as injection at worst. Write declaratively, attribute guidance, never hide text.

**Shared agent-readable repositories help, modestly.** In AgentRxiv, gpt-4o-mini labs that could read earlier agent papers reached 78.2% on MATH-500 from a 70.2% baseline, versus a plateau near 73.5% without access ([arXiv](https://arxiv.org/html/2503.18102v1)). One benchmark and a small model, but it is the only measured support for publishing for agents.

---

Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space