Lesson 1453 of 1596
Giving Agents a Scratchpad They Re-Read
Use a working file the agent updates and consults each step.
Creators · Agentic AI · ~7 min read
The premise
Long-running agents lose track of their own plan. A persistent scratchpad they update and re-read keeps them coherent across many steps.
What AI does well here
- Write and update a plan file between steps.
- Use the file to recover after a tool error.
What AI cannot do
- Maintain reliable working memory inside the model alone.
- Trust that 'I will remember this' actually persists.
Key terms in this lesson
Practice this safely
Use a small project example from your own work. The useful move is to compare the AI's draft against your goal, sources, and constraints before you trust it.
- 1Ask AI to explain scratchpad in plain language, then underline anything that sounds uncertain or too broad.
- 2Give it one detail from "Giving Agents a Scratchpad They Re-Read" and ask for two possible next steps plus one reason each step might be wrong.
- 3Check working-memory against a trusted source, teacher, adult, expert, or original document before you use it.
End-of-lesson quiz
Check what stuck
10 questions · Score saves to your progress.
Tutor
Curious about “Giving Agents a Scratchpad They Re-Read”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Creators · 9 min
Agentic AI: Choose Short-Term vs Long-Term Memory Without Building Both
Most agents do not need a vector database — pick the simplest memory that solves the actual recall problem in front of you.
Builders · 7 min
Short-Term vs Long-Term Memory for Agents
Know which facts the agent should remember within a run vs across runs.
Creators · 48 min
Computer Use API: Letting AI Click Through GUIs
Computer Use lets Claude see your screen and use it — mouse, keyboard, apps. The capability is real, the gotchas are real. A hands-on look at what works in 2026.