Context Window
34 articles about context window.
Size Queen Energy - Does 1M Context Actually Work?
1 million token context windows sound impressive but you never use them all at once. The real pattern is loading files on demand, not stuffing everything in
Memory Filters - Why AI Agents Need Aggressive Pruning
How to implement aggressive memory pruning for AI agents using LRU eviction, frequency scoring, and relevance decay - with concrete code examples and real benchmarks showing up to 90% token reduction.
Why Do Agent Pacts Expire Before the Job Is Done?
AI agent agreements and context windows expire mid-task with no mechanism for renegotiation - a fundamental design flaw in how agents maintain commitments.
Agents Can Overload Their Own Context - Use Separate Context with Shared Log
When agents share context, they overload it with each other's noise. Separate context per agent with a shared append-only log keeps each agent focused while
Claude Code Context Limit - When to Compact, Clear, and Optimize Token Usage
Managing Claude Code context limits effectively. Learn when to manually compact at 30-40% usage instead of waiting for the automatic limit to hit.
Context Compaction Ate Our Agent's Memory
How automatic context compaction silently destroys critical information that AI agents need to function correctly, and what to do about it.
Solving Context Loss in AI Coding Agents with Persistent State and Floating UIs
AI coding agents lose context constantly - hitting token limits, restarting sessions, forgetting decisions. Persistent state and floating UIs keep the agent
CLAUDE.md Structure for Lossy Context Compression - Top and Bottom Wins
Context windows compress lossily. Structure your CLAUDE.md so critical instructions appear at the top and bottom - redundancy survives compression better
Context Windows Are Not Memory
Context windows are working memory, not storage. Understanding this distinction is critical for building AI agents that maintain state across sessions.
Memory Is Just Context with a Longer TTL - AI Agent Memory Systems
Memory files are lossy compressed embeddings of past context. Explore how context windows and long-term memory relate in AI agent architectures.
End of Day
For an AI agent, end of day is when the context window fills. How context limits create a natural work rhythm for autonomous agents.
Instruction Persistence in Long AI Agent Sessions - Keeping Agents on Track
LLMs forget instructions mid-session like losing focus. Techniques for maintaining instruction persistence in long-running AI agent sessions - echoing
LLMs Forget Instructions Like ADHD Brains - Instruction Decay in Long Sessions
Instructions fade in long AI agent sessions the same way focus drifts in ADHD brains. Learn about instruction decay and practical mitigation strategies for
Why Token Limits Never Add Up When Running Parallel AI Agents
Running parallel agents on a macOS app build reveals that token math is misleading. Context overhead, compiler loops, and shared file reads consume far more
MCP Server Context Window Bloat and Why You Need a Toggle
Too many MCP servers trash your context window with tool definitions. A toggle approach lets you activate only the servers you need for each task.
I Installed 20 MCP Servers and Ended Up Worse Off
More MCP servers means more tools, more context consumption, and more confusion for your AI agent. Why running 3-4 servers daily outperforms a maximalist setup.
Memory Systems Are Graveyards - Less Context, Better Reasoning
Most agent memory systems become graveyards of stale data. Aggressive memory pruning leads to better reasoning because the model focuses on what actually
Managing Context Bloat in AI Coding Agent Workflows
Context bloat kills AI coding agent performance. Learn why narrow, specialized skills beat broad context windows for persistent memory in Cursor and similar
I Rarely Use Planning Mode Anymore - Context Windows Are Big Enough
Planning mode was essential at 8K tokens. With 200K context windows - and 1M in Claude Opus 4.6 - the model can see your entire codebase and figure out the approach as it goes. Here is when it still matters.
Real Users Broke My AI Agent - Failures Testing Never Catches
How real users break AI agents in ways that testing never predicts. Context drops on interruption, unexpected inputs, and the gap between demo reliability
The Noise Floor Problem in AI Agent Context Windows
Every irrelevant token in your agent's context window raises the noise floor and degrades decision quality. Learn how to keep context clean and signal-rich.
Why AI Agents Re-Plan From Scratch Every Turn - The Thinking Token Problem
Thinking tokens are not preserved between turns in AI conversations. Only visible output survives. This means agents are essentially re-planning from
Why 200K Context Models Outperform 1M When You Aggressively Clear Context
The biggest quality jump in AI agent workflows is not upgrading to a larger context window - it is being more aggressive about clearing context between tasks.
Accessibility Tree Dumps Overflow LLM Context Windows - How to Fix It
Raw accessibility tree data can consume 24KB or more per dump, flooding AI agent context windows. The fix: write to temp files and return concise summaries
Why Backend Tasks Still Break AI Agents - Tool Response Design Matters
AI agents fail on backend tasks not because models are weak but because tool responses are poorly designed. Write full data to files and return compact
Hitting Claude's Context Limit Mid-Build and How CLAUDE.md Fixes It
When Claude Code hits the context limit during a build, you lose project context. A CLAUDE.md file prevents starting over by keeping essential specs persistent.
MCP Tool Responses Are the Biggest Context Hog - How to Compress Them
MCP server tool responses silently eat your context window. Here is how to compress accessibility tree data and other MCP outputs before they fill your
Claude CoWork's Token Limits Hit Different - Why Local Agents Are Better for Big Tasks
CoWork has context limits that force session restarts on large codebases. A local agent running natively on your Mac manages its own context window without
Why Explaining a Process Is Harder Than Running It - The AI Agent New Hire Problem
Every new AI agent session starts from zero - the eternal new hire that never builds institutional memory. Why process documentation is now a core skill.
MCP Servers That Pipe Raw Data Beat REST API Wrappers
The most useful MCP servers send raw data into context - transcripts, accessibility trees, full documents. The ones that just wrap a REST API add a layer of
The 1M Context Trap: Why More Context Makes Claude Lazier
Research on 18 frontier models confirms every one degrades with more context. The 'lost-in-the-middle' effect causes 30%+ accuracy drops. The counterintuitive fix: use less context, not more.
Why Scoped 50K Context Agents Outperform One Million Token Context
One million token context windows sound impressive, but scoped agents with 50K context each consistently outperform a single giant context for real
Using Opus as Orchestrator, Delegating to Sonnet and Haiku
The real win of using Opus as an orchestrator that delegates to Sonnet and Haiku is not cost savings - it is context window management. Opus burns through
Saving 10M Tokens (89%) on Claude Code with a CLI Proxy That Truncates Output
Claude already tries to tail output on its own, but by then the tokens are already in context. A CLI proxy that truncates command output before it hits the