What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

How Long Contexts Fail — mattwood.fyi

Connections

Challenges

Context Language Models Context Language Models are premised on leveraging context effectively - the failure modes described (poisoning, distraction, conflicting info) directly challenge whether context-centric model architectures deliver on their promise
UNREAL: Unifying Retrieval and Long-Context with a Single Model UNREAL model aims to unify retrieval and long-context processing, which directly addresses the failure modes described - good retrieval could mitigate context poisoning and distraction by being selective about what enters the context

Supports

Thinking Before Thinking: Scaling Agentic Inference Through Meta-Reasoning Meta-reasoning before acting (thinking before thinking) is a direct mitigation for context overload failures - agents that reason about what context is relevant can avoid distraction and confusion from irrelevant information
Unreal Agent: Cost-Efficient AI Agent Harness Unreal Agent explicitly focuses on cost-efficient agent harness design, suggesting architectural solutions to context management that align with avoiding the overloaded context failure modes described
Bag of Decisions Reranker Bag of Decisions Reranker addresses selective information prioritization, which is a mitigation strategy for context confusion and distraction - reranking helps surface relevant decisions rather than overwhelming agents with all context

Related to

Coding Agent Continual Learning: Analysis of ARC-AGI-3 Problem Solving ARC-AGI-3 coding agent analysis would encounter exactly these long-context failure modes in multi-step problem solving, making empirical observations from that benchmark relevant evidence for the claims made

Challenged by

Muse Spark Open: 816B Open-Weight AI Model The 1M-token context window directly confronts known failure modes of long contexts described in that article - Meta's optimization claims would need to address the documented degradation patterns
Via AI Launches Enterprise Coding Assistant at Half the Cost Via AI's Codegraph compression algorithm is specifically designed to address the failure modes of long contexts in coding tasks, positioning it as a technical solution to documented long-context limitations

Supported by

Sygal - Document to Markdown Converter Sygal addresses a root cause of long-context failures: poorly structured input. By converting documents to clean Markdown/JSON, it reduces the noise and formatting issues that cause LLMs to lose track of information in long contexts.
Dessa: Conversation Model Trained on Speech Dessa's speech-based training implicitly supports concerns about how long contexts fail — spoken language is inherently more concise and turn-based, suggesting speech data may produce more robust short-context conversational models
ThinkingCap: Gemma-4-26B-A4B Token Efficiency ThinkingCap's 39% reduction in thinking tokens directly addresses the token waste and inefficiency problems with extended context/reasoning that 'How Long Contexts Fail' describes
ThinkingCap: Gemma-4-26B-A4B Token Efficiency ThinkingCap addresses a specific failure mode of long contexts — excessive thinking token generation — directly relevant to how long contexts fail through token budget overruns and inefficiency

Type:	link
First published:	Tuesday, 13 October 2026 at 18:01 UTC
Location:	https://www.dbreunig.com/2025/06/22/how-contexts-fail-and-how-to-fix-them.html
URL:	https://mattwood.fyi/i/f2e3a42a

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections