# What is agent memory?

> Agent memory is the notes an agent writes down while it works, such as corrections, decisions and facts about your business, and reads at the start of its next run. A correction is written once and applies to every later run.

**Agent memory:** The notes an AI agent writes down while it works, such as corrections, decisions and facts about the business, and reads at the start of its next run.

## Why it compounds

Each correction becomes a note the agent reads next time, so the same mistake does not return. The memory grows with the work.

## How we use it ourselves

In our own workspace, finished work ends with a few lines written to a dated log, and durable facts are promoted to a wiki page. Each later session reads those first.

## What goes in, and what never does

- In: corrections, decisions, preferences and facts about the business.
- Never in: passwords or keys. Credentials are referred to by name only.

## Who owns it

You do. The code, the prompts and the accounts are in your company's name, and the memory lives with them.

## Related

- [What is a Personal CEO Agent?](https://agenticengineeringconsulting.com/wiki/personal-ceo-agent)
- [How we run agents after they go live](https://agenticengineeringconsulting.com/wiki/how-we-run-agents)
- [What does human in the loop mean for AI agents?](https://agenticengineeringconsulting.com/wiki/human-in-the-loop)

## Next step

- Apply: https://agenticengineeringconsulting.com/apply
- Free AI audit: https://agenticengineeringconsulting.com/audit
- Book a brief: https://calendar.app.google/sdYkXF3ny2kvbvp39
- Email: henryvantieghem@agenticengineeringconsultingllc.com

---
Canonical page: https://agenticengineeringconsulting.com/wiki/agent-memory