Does My AI Agent Remember What I Told It Last Time?
The toggle that lets my AI agent learn from our chats, the one sentence that makes a correction stick, and the reminder I keep to read what it changed.
It can, when the edit own instructions toggle is on. My AI agent then learns from how I interact with it in the chat and automates updates to its own instructions. Gumloop's documentation says a correction holds once it's saved in the instructions or a skill, so ask for that. I keep a reminder on my calendar to read what it changed.
You corrected your AI agent, and the next time it did the job the old way. Where it keeps what you tell it decides whether a correction holds.
The builder in my session was Gumloop. On my relationship to Gumloop, I said this:
From the recording, 33:22. I don't work for Gumloop. I found this tool over the summer through Lenny's podcast, and I started using it and have since made myself an ambassador for them. I've created a relationship with Gumloop. I am not an employee.
More on that is in Do you work for Gumloop? (the answer from the session).
Where does my AI agent keep what I tell it?
The first place is its instructions. When I set up my chief of staff AI agent, I pointed to a toggle.
From the recording, 24:18. You'll notice there's something that says edit own instructions with a blue toggle button that is on.
From the recording, 24:26. What that does is it allows the agent to learn from your interaction inside of the chat feature.
I also said you can start more than one chat, and a new task opens another chat inside the same AI agent.
From the recording, 24:32. And you can start multiple chats. For example, if you click new task that opens another chat inside of this agent
The second place is a skill, which tells my AI agent what to do step by step (34:09). When a good result won't repeat, the fix I teach is a skill, and Why Does AI Give Me a Great Result Once and Then Stop Doing It? walks through it. How Do You Create Skills for an AI Agent? shows how I make one.
What does the documentation say about corrections?
Checked on 2026-11-05 against Gumloop's AI agent documentation. With Edit Own Instructions on, the page says a correction edits the instructions and persists across future tasks. It says the toggle is on by default and that there's no version history, so you revert by editing the prompt manually. It also says feedback in one conversation alone doesn't update every future task, so it tells you to ask the AI agent to save the correction in its instructions or a skill. It describes each task as its own conversation with its own history, and it lists an ability called Search Past Tasks, on by default, that searches and retrieves earlier tasks for context.
The two statements fit together. With the toggle on, the AI agent can write a correction into its instructions, and what is written there carries into future tasks. A remark that is never saved stays in that one chat, so ask the AI agent to save it.
I didn't cover on the recording whether a second AI agent can see what the first one learned, so I'm not claiming it here.
How do I make a correction stick?
- 01
Tell your AI agent what is off, in one sentence.
- 02
Tell it to keep the fix. The file on the session page puts it this way for my chief of staff AI agent:
From the file on the session page. If I tell you something is off, in one sentence, update the next answer to reflect it. Don't ask me to explain further unless my correction doesn't make sense. Keep the fix; don't quietly revert to the first version next time.
- 03
Ask your AI agent to save the correction in its instructions or a skill, as the documentation above describes.
- 04
Read the instructions on a schedule, so you know what is in them.
How often do I read what it changed?
I have a 15 minute meeting on my calendar that comes back every three weeks, and I use it to read my AI agent's instructions while they update on their own (24:51). For an AI agent you use daily, I said this:
From the recording, 25:10. If you're using your agent daily, you may want to check yours every two weeks.
How Often Should You Review an AI Agent's Instructions? has the cadence.
What goes wrong
When instructions update on their own, you want oversight into what the AI agent is updating.
From the recording, 25:31. You want to be able to manage and have oversight into what the agent is updating.
From the recording, 25:36. So please put a calendar reminder, at some frequency for you to come back and check that.
What the recording doesn't show: the text of my AI agent's instructions after weeks of updates.
Where this comes from
This article grows from the session I taught live with Women Applying AI on October 7, 2026. The quoted passages from the recording are my own words, taken verbatim. The passage marked as from the file on the session page comes from the files published with that session. The steps, tables and checklists around them were drafted by my team's AI agents from those words and my published pages, and checked against the recording before publishing.
Build Your First AI Agent in One Hour: the full session with the clips, the prompts and the transcript.
Related reading: What Does Edit Own Instructions Mean for an AI Agent?, How Do You Create Skills for an AI Agent?, and How Often Should You Review an AI Agent's Instructions?.
Frequently asked questions
Does my AI agent learn from our chats?
With the edit own instructions toggle on, it learns from how I interact with it in the chat and automates updates to its instructions (24:26).
Will a correction I make once hold?
Often, if it gets saved. The file on the session page tells my chief of staff AI agent to keep the fix and not quietly revert, and the documentation says to ask the AI agent to save the correction in its instructions or a skill.
How do I know what my AI agent changed?
I read its instructions on a calendar reminder. Mine comes around every three weeks.
Build something that actually runs your workflow.
A focused, free 60-minute live session with Jacqueline. You build alongside her, on your own real task, and leave with an AI agent that is already running.