Context window · from the video

Why your AI "forgets" what you told it

Salamat sa pag-comment ng "CONTEXT"! Here's the visual guide from the video, 3 habits for long AI chats, and the handoff prompt you can copy.

  • ✓Try the desk
  • ✓3 habits
  • ✓Handoff prompt

Part 1 · The desk

Fill the desk, watch what happens

The AI's desk holds everything it can see right now: your instructions, the chat, your files. Your first message is the login rule. Keep adding messages and watch where it goes.

Desk space0 / 8

A simplified picture, not a real measurement. Real tools count space in tokens (pieces of words), the limits are much bigger, and each tool handles a full desk its own way.

Part 2 · What's actually going on

The desk is the context window

Every AI model has a limit on how much it can see at once. When a chat gets longer than that, the tool has to do something with the older parts.

Context window The desk

Everything it can see at once

Your messages, its replies, files you pasted or attached, and the instructions it was given. It's measured in tokens, and every model has a maximum.

Summarized Some tools

Old stuff becomes a short note

Some tools squeeze older parts of the chat into a summary. Claude Code calls this compacting. The gist stays, pero small details, like the exact rule you gave, can get lost.

Dropped Other tools

Old stuff falls off the desk

Other tools stop including the oldest messages, or ask you to start a new chat. It depends on the tool, and tools change how they do this over time.

Tandaan

It's not the only reason

A full desk is one possible reason, not the only one. Even with space left, one instruction buried in a long, messy chat is easier to miss, and AI can simply make mistakes.

So whatever tool you use, check what it changed before you accept it.

Part 3 · 3 habits

Keep the desk clean

You don't need a longer prompt. You just need a cleaner desk.

  1. One task, one chat

    New task? New chat. Huwag pagsabayin lahat sa isang thread. A short, focused chat leaves more room on the desk, and your instructions stay in view.

  2. Save your rules in a file

    Rules you don't want it to forget go where your tool loads them every session, instead of somewhere in the middle of a long chat:

    • Claude CodeCLAUDE.md in your project folder
    • CodexAGENTS.md in your project folder
    • ChatGPTCreate a Project and add project instructions, or use Custom instructions for every chat
    • Claude appCreate a Project and add project instructions

    Need a starter file? Get the BASIC starter template from our first video.

  3. Summarize, then start fresh

    When the chat gets too long, ask the AI for a handoff note of the key decisions, check it, then paste it into a new chat. The prompt below does exactly that.

Claude Code user?

You can see and clean the desk yourself

  • /context shows how full the context window is.
  • /compact summarizes the chat now. You can tell it what to keep, like /compact keep the login rule.
  • /clear starts fresh. Your CLAUDE.md still loads.

Using another tool? Check its docs or type / to see its commands.

Part 4 · The handoff prompt

Before you start a fresh chat

Paste this into your long chat. You'll get a short note you can carry into a new chat, with your rules written out in full.

handoff-prompt.txt
Before we continue in a new chat, write a handoff note of this conversation.

Include:
1. Goal: what we're building or doing
2. Decisions: what we decided, and why
3. Rules: everything I told you to do or NOT do (use my exact words)
4. Status: what's done and what's left
5. Open questions: anything still unclear

Use short bullet points. Don't add new suggestions.
  1. Paste it into the long chat you want to leave.

  2. Read the note. Check that every rule is there, word for word. If something's missing, add it yourself.

  3. Open a new chat, paste the note, then say what's next. A rule you'll always need? Move it into your rules file.