Stop the memory rot Stop the memory rot TL;DR: You can keep your AI sharp by forcing it to summarize and prune what it remembers (a.k.a. compacting). TL;DR: You can keep your AI sharp by forcing it to summarize and prune what it remembers (a.k.a. compacting). Common Mistake ❌ You keep a single, long conversation open for hours. You feed the AI with every error log and every iteration of your code. Eventually, the AI starts to ignore your early instructions or hallucinate nonexistent functions. Problems Addressed 😔 Context Decay: The AI loses track of your original goals in the middle of a long chat.
Hallucinations: The model fills memory gaps with hallucinations or outdated logic.
Token Waste: You pay for the AI to re-read useless error logs from three hours ago.
Reduced Reasoning: A bloated context makes the AI less smart and more prone to simple mistakes. Context Decay: The AI loses track of your original goals in the middle of a long chat. Context Decay: Hallucinations: The model fills memory gaps with hallucinations or outdated logic. Hallucinations: Token Waste: You pay for the AI to re-read useless error logs from three hours ago. Token Waste: Reduced Reasoning: A bloated context makes the AI less smart and more prone to simple mistakes. Reduced Reasoning: How to Do It 🛠️ Restart often: You can start a new chat once you finish a sub-task.
Request a State Summary: Before you close a conversation, ask the AI to summarize the current decisions and plan.
Add Human Checkpoints: After the summary, confirm you are still on track.
Use Markdown Docs: Keep a small context.md file with your current stack and rules.
Prune the Logs: You should only paste the relevant 5 lines of a stack trace instead of the whole irrelevant 200-line output.
Divide and conquer: Break large tasks into smaller ones, invoking their own skills with local tokens and a fresh context.
Divide the responsibility: A General doesn't need to know what every soldier is doing on the battlefield.
Create and persist as Skill: After you have taught the AI, you should refactor the knowledge and business rules.
Keep an Eye on the Context Size: Most tools have visual indicators of the window consumption.
Use Local Persistence: Some tools allow sharing memory among agents and their sub-agents. Restart often: You can start a new chat once you finish a sub-task. Restart often: start a new chat Request a State Summary: Before you close a conversation, ask the AI to summarize the current decisions and plan. Request a State Summary: plan Add Human Checkpoints: After the summary, confirm you are still on track. Add Human Checkpoints: confirm Use Markdown Docs: Keep a small context.md file with your current stack and rules. Use Markdown Docs: context.md Prune the Logs: You should only paste the relevant 5 lines of a stack trace instead of the whole irrelevant 200-line output. Prune the Logs: Divide and conquer: Break large tasks into smaller ones, invoking their own skills with local tokens and a fresh context. Divide and conquer: fresh context Divide the responsibility: A General doesn't need to know what every soldier is doing on the battlefield. Divide the responsibility: Create and persist as Skill: After you have taught the AI, you should refactor the knowledge and business rules. Create and persist as Skill: Keep an Eye on the Context Size: Most tools have visual indicators of the window consumption. Keep an Eye on the Context Size: Use Local Persistence: Some tools allow sharing memory among agents and their sub-agents. Use Local Persistence: Benefits 🎯 You get more accurate code suggestions.
You avoid divergences
You follow the AI's train of thought.
You spend less time correcting the AI's hallucinations.
The AI follows your project constraints more strictly and keeps focused on your tasks You get more accurate code suggestions. You avoid divergences You follow the AI's train of thought. You spend less time correcting the AI's hallucinations. The AI follows your project constraints more strictly and keeps focused on your tasks Context 🧠 Large Language Models have limited attention. Long context windows are a trap. Many modern models offer a very large context window. In practice, they ignore a lot of them to your frustration. Even with large context windows, they prioritize the beginning and end of the prompt. Prompt Reference 📝 Bad Prompt Here is the 500-line log of my failed build. 

Also, remember that we changed the database schema 

Three hours ago in this chat.

Add the unit tests as I described above.

Now, refactor the whole component. Here is the 500-line log of my failed build. 

Also, remember that we changed the database schema 

Three hours ago in this chat.

Add the unit tests as I described above.

Now, refactor the whole component. Good Prompt I am starting a new session. Here is the current state: 

We use *PostgreSQL* with the 'Users' table schema [ID, Email]. 

The AuthService`interface is [login(), logout()]. 

Refactor the LoginComponent` to use these. I am starting a new session. Here is the current state: 

We use *PostgreSQL* with the 'Users' table schema [ID, Email]. 

The AuthService`interface is [login(), logout()]. 

Refactor the LoginComponent` to use these. Considerations ⚠️ You must ensure you don't purge essential context. essential essential If you prune too much, the AI might suggest libraries that conflict with your current setup. Review the compacted information. Type 📝 [X] Semi-Automatic Limitations ⚠️ You can use this tip manually in any chat interface. If you use advanced agents like Claude Code or Cursor, they might handle some of this automatically, but manual pruning is still more reliable. Tags 🏷️ Context Context Level 🔋 [X] Intermediate Related Tips 🔗 https://maximilianocontieri.com/ai-coding-tip-004-use-modular-skills https://maximilianocontieri.com/ai-coding-tip-004-use-modular-skills https://hackernoon.com/ai-coding-tip-005-how-to-keep-context-fresh https://hackernoon.com/ai-coding-tip-005-how-to-keep-context-fresh AI Coding Tip 010 - Create Skill from Conversation Conclusion 🏁 You are the curator of the AI's memory. If you let the context rot, the code will rot, too. Keep it clean and compact. 🧹 More Information ℹ️ https://arxiv.org/abs/2307.03172?embedable=true https://arxiv.org/abs/2307.03172?embedable=true https://llmlingua.com/?embedable=true https://llmlingua.com/?embedable=true https://www.ibm.com/think/topics/ai-hallucinations?embedable=true https://www.ibm.com/think/topics/ai-hallucinations?embedable=true https://www.promptingguide.ai/?embedable=true https://www.promptingguide.ai/?embedable=true Also Known As 🎭 Context Pruning
Token Management
Prompt Compression Context Pruning Token Management Prompt Compression Tools 🧰 Claude Code
Cursor
Windsurf Claude Code Cursor Windsurf Disclaimer 📢 The views expressed here are my own. I am a human who writes as best as possible for other humans. I use AI proofreading tools to improve some texts. I welcome constructive criticism and dialogue. I shape these insights through 30 years in the software industry, 25 years of teaching, and writing over 500 articles and a book. This article is part of the AI Coding Tip series. AI Coding Tip https://maximilianocontieri.com/ai-coding-tips https://maximilianocontieri.com/ai-coding-tips

Keep

AI Coding Tip 009 - Compact Your Context and Stop Memory Rot

Buy My Book

Share your ideas with me!

AI Coding Tip 009 - Why You Should Compact Your Context

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

10 Predictions About the Future of Finance

How Artificial Neural Networks Can Code Smarter Than GUI Programmer

How GPT Pilot Codes 95% of Your App

The Modern AI Stack to Increase Developer Productivity

When AI Joins the Dev Team

How I Made My Own Screenshot-To-Code Project

10 Predictions About the Future of Finance

How Artificial Neural Networks Can Code Smarter Than GUI Programmer

How GPT Pilot Codes 95% of Your App

The Modern AI Stack to Increase Developer Productivity

When AI Joins the Dev Team

How I Made My Own Screenshot-To-Code Project

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps