Meta context engineering beats hand-tuned context
Meta context engineering (ICML 2026) learns the context-engineering process itself, beating ACE-style curation by 18 points while training 13.6x faster.
Further reading
Articles about Context Engineering
63 articles from the Wire blog, sorted newest first. Return to the Context Engineering definition for context.
Claude Fable 5 put a price on context portability
Claude Fable 5 returns refusals as HTTP 200s and retries them on Opus 4.8. The fallback API reveals exactly what agent context survives a mid-task model swap.
Five criteria of good context for AI agents
A 2026 paper formalizes five criteria for good AI agent context: relevance, sufficiency, isolation, economy, and provenance. Here's how to design for each.
Sub-agent context isolation: the fix for context rot
Sub-agent context isolation gives each agent its own scoped window, stopping the context rot that kills multi-agent runs. Here's the pattern and its limits.
When to process AI context: ingestion, dreaming, query
There are three moments to process AI context: ingestion, a background pass some call dreaming, and query time. Match each kind of work to the right one.
Claude Opus 4.8 hallucinates less by answering less
Claude Opus 4.8 tops a hallucination benchmark without getting more accurate. It learned to abstain. Why retrieval honesty is a context engineering win.
RAG, long context, or fine-tuning: how to choose
RAG, long context, or fine-tuning? A 2026 decision guide on cost, accuracy, and freshness, with a use-case table for choosing the right one in production.
File-native agents: when reading files backfires
A 9,649-experiment study found file-native retrieval lifts frontier-model accuracy 2.7% and drops open-source accuracy 7.7%. Match architecture to the model.
AI notetakers ship the wrong artifact
AI notetakers ship transcripts, but downstream work needs decisions, drafts, or handoffs. The artifact gap is a context engineering problem, not transcription.
Constraint decay: structural rules break AI coding agents
Constraint decay: AI coding agents lose 30 points of accuracy under architecture and database rules. New EURECOM study explains why and where it hurts most.
Context offloading: 3 patterns for AI agents
Context offloading keeps an AI agent's working context window small by moving state to a destination outside it. Three patterns, and what each one costs.
Your MCP server is installed. Your agent isn't using it.
Connecting an MCP server is easy. Getting an agent to call its tools on the first relevant turn is where teams lose, and the cause is context.
Why 14 MCP CVEs trace back to one design choice
OX Security's April 2026 advisory traces 14 MCP CVEs and 200,000 exposed servers to a single design choice: STDIO as the default local transport.
What Anthropic's context engineering guides leave out
Anthropic's 2026 trilogy on context engineering, tools, and code execution with MCP each assume the same missing layer: the substrate where context lives.
What 466 AGENTS.md files teach about context engineering
An MSR 2026 study of 466 open source projects maps the five modes developers use to write AGENTS.md context, and what 50% file staleness reveals about practice.
When agent memory needs sleep, and when it doesn't
Memory consolidation fixes one specific failure: agents writing the same claim dozens of times into a flat scratchpad. When it helps and where it breaks.
MCP vs Skills vs CLI: which one wastes the least context?
GitHub's MCP costs tens of thousands of tokens before any work begins. We compare MCP, Claude Skills, and CLI by context cost, not by user preference.
Tool calling is retrieval: what Needle proves at 26M
A 26M-parameter model just matched Gemini at function calling. Here is what Needle's distillation result means for MCP and agent context engineering.
MCP resources are the underused half of MCP
Every MCP discussion is about tools. The protocol's resources primitive is how you load context without paying for it every turn. Here's how to use it.
Agentic context engineering: how ACE evolves contexts
ACE (ICLR 2026) beats tuned prompts by 10.6% with self-evolving contexts that avoid brevity bias and context collapse, two real failures of prompt tuning.
Context engineering lives in substrates, not harnesses
Codex shipped codex-plugin-cc and AGENTS.md joined the Linux Foundation. The signal is consistent: context engineering is substrate work, not harness work.
Long context tripled hallucinations in 35 open models
A 172-billion-token study across 35 open models found hallucination rates triple from 32K to 128K context, and exceed 10% at 200K for every model tested.
Progressive tool loading is the new MCP context pattern
Preloading every MCP tool into an agent's context is the bottleneck of 2026. Progressive tool loading defers definitions until needed and saves tokens.
Anthropic's Managed Agents memory: what it changes
Anthropic launched Memory for Managed Agents on April 23, 2026 in public beta. What the design means for agent scope, freshness, and context engineering.
What MCP's 2026 roadmap means for context delivery
The MCP 2026 roadmap reframes Model Context Protocol as enterprise context infrastructure: stateless transport, MCP Apps SEP-1865, audit logs, SSO auth.
Why every agent handoff corrupts your context
Every multi-agent handoff is a lossy compression event. Learn which five types of context degrade at agent handoff boundaries and how to preserve them.
Tool poisoning: how MCP tool descriptions hijack agents
Tool poisoning hides instructions inside MCP tool descriptions the agent reads as trusted context. The MCPTox benchmark recorded a 72.8% attack success rate.
Tool-based agent memory: why 2026 benchmarks favor it
Tool-based agent memory exposes store, retrieve, and navigate as callable MCP tools. 2026 benchmarks from Mem0, Memanto, and Wire show why the pattern wins.
Why AI customer support replies sound generic
AI support replies sound generic because teams treat brand voice as a prompt problem. Context engineering fixes it by selecting the right exemplars.
TOON vs JSON: why smaller doesn't mean cheaper for LLMs
TOON looks more compact than JSON, but a 9,649-test study found it cost LLMs 38% more tokens. The reason: model training distribution beats format size.
GPT-5.5 didn't cut hallucinations 60%. Here's what it did.
OpenAI's GPT-5.5 system card reports 23% better claim-level accuracy, not the 60% hallucination reduction making press rounds. Here's what actually changed.
Agent drift: why long-running AI agents lose the plot
Agent drift is how AI agents silently deviate from goals over long-running tasks. Six mechanisms cause it, and most have nothing to do with the model.
Provenance is a context engineering primitive, not a trust score
Retrieval provenance for AI agents isn't an audit log or a trust verdict. It's structural metadata (source, position, time, edges) agents use to plan.
Why token cost doesn't scale with knowledge base size
AI token usage scales with knowledge base size only when the full corpus loads per query. The real variable is selective context delivery, not KB size.
One job per tool: why adding wire_navigate cut agent calls 24%
We restructured Wire's MCP surface from 2 overloaded tools to 3 single-purpose ones. The counterintuitive result: adding a tool cut total calls 24%.
GPT-5.4-pro hallucinates more than GPT-5.4-nano
Vectara's 2026 benchmark shows OpenAI's flagship GPT-5.4-pro hallucinates at 8.3% while its nano variant stays at 3.1%. The reasoning-model tradeoff, explained.
How to build a private AI second brain
Native Notion and Obsidian MCP give every connected agent the same coarse scope. Build a private AI second brain with per-agent, revocable access across tools.
RAG vs fine-tuning: when to use each
RAG vs fine-tuning: RAG wins for knowledge injection and freshness, fine-tuning wins for style and format. The right choice is a context engineering call.
Context budgets: how to allocate tokens for AI agents
A practical guide to context budgets for AI agents. How to allocate tokens across system prompts, tools, retrieval, history, and a buffer in production.
Context Poisoning: When Bad Data Becomes AI Ground Truth
Context poisoning plants false data into an AI agent's memory or RAG index. The model treats it as truth. It's a context engineering problem, not a model bug.
Why your AI costs are a context problem
Token prices fell 280x but enterprise AI spend rose 320%. Poor context architecture drives 60-70% of total AI costs. Here is where the money actually goes.
RAG vs long context: what the 2026 data shows
RAG vs long context in 2026: which wins on cost, speed, and accuracy, and when each one beats the other in production. What the benchmarks actually show.
How context engineering reduces AI hallucinations
Most AI inaccuracies in production are context quality failures, not model fabrications. Here's the research on what context engineering actually changes.
How to connect AI to private data safely
77% of employees share sensitive data with AI tools. Five context engineering patterns give AI what it needs without exposing what it shouldn't see.
Context compression: why less context means better AI
Context compression reduces AI agent memory usage by 26-54% while preserving task performance. Here's how it works and why bigger context windows aren't the answer.
How prompt caching cuts AI agent costs by 90%
Prompt caching reduces AI agent API costs by up to 90% and latency by 31%. Here's how it works, where it breaks, and how to implement it right.
Why Customer Support AI Gives Wrong Answers
AI customer service fails at 4x the rate of other AI tasks. Support bots need five types of context most teams never provide. The model isn't the problem.
Why AI agents forget mid-task (and how to fix it)
65% of agent failures come from context drift, not token limits. Here's how context compression keeps long-running AI agents on track.
Why AI Agent Memory Keeps Failing
AI agent memory fails because it's a context engineering problem, not a storage problem. Research reveals three failure modes and what actually works.
Why AI Coding Assistants Can't See Your Codebase
84% of developers use AI coding tools, but only 29% trust the output. The problem has less to do with models and more to do with codebase context.
How to measure context quality for AI agents
Five dimensions of context quality that determine AI agent performance, with metrics, benchmarks, and practical measurement approaches for production systems.
Why Semantic Search Beats Keywords for AI
Hybrid search improves AI retrieval accuracy by up to 41% in technical domains. Here's how semantic search works, where keywords fail, and when you need both.
Why Product Management AI Gives Wrong Priorities
84% of product teams doubt their products will succeed despite AI adoption. The problem: PM tools see feature requests but not the context behind what to build.
Why Sales AI Fails Without Deal Context
87% of enterprises missed revenue targets despite AI investment. Sales AI needs five types of deal context most teams never provide. The model isn't the issue.
Why multi-agent AI systems fail at context
Up to 86.7% of multi-agent AI runs fail. Most failures trace back to how agents share context, not the agents themselves. Here's why and how to fix it.
7 context engineering techniques for production
Seven context engineering techniques used in production AI systems, with implementation patterns, research backing, and guidance on when each one works.
Structured Context vs Raw Text for AI
ETH Zurich found AI-generated context files hurt agent performance by 3%. Format choice alone swings LLM accuracy by 40%. Here's what the research says.
Why MCP failures are a context engineering problem
New research analyzed 3,282 MCP bug reports across GitHub. The patterns reveal a context delivery problem, not a protocol problem. Here's what it means.
What Is a Context Window?
A context window is the total text an AI model can process at once. Learn how they work, why size isn't everything, and what actually affects performance.
AI agents have too much access
88% of organizations report AI agent security incidents. The root cause is a context engineering failure: agents get all-or-nothing access, not scoped context.
Why AI Hallucinations Are a Context Problem
GPT-5.2 hallucinates at 10.8%, o3-pro at 23.3%. The fix has less to do with better models and more to do with context engineering. Here's the research.
Context engineering: the end of prompt engineering
Prompt engineering is a dead end. Context engineering — designing what information AI models receive — is replacing it. Here's how to start applying it.
Context engineering: what replaces prompt engineering
Prompt engineering has a new successor: context engineering. Learn why Karpathy and Tobi Lütke made the switch, and what it means for production AI systems.
Put context into practice
Create your first context container and connect it to your AI tools in minutes.
Create Your First Container