Does the AI remember me tomorrow?
A short learning video. No voice, only titles and screens, with music.
What this video shows
Today you tell a chat your shop: “My shop is called Blue Door.” It answers “Nice name.” Then the sun sets, the chat window folds away, and it is gone. It keeps nothing.
The next day a new chat opens, empty. “Write a sign for my shop.” The answer: “Which shop?” Tomorrow it starts empty.
Then the same day again, with one change: a small notebook marked Memory sits beside the chat. A gold pen writes one line in it: “Shop: Blue Door.” At night the chat folds away as before, but the notebook stays, open, with its line. Next morning a new empty chat opens, and before anyone types, that gold line slides out of the notebook into the chat. It reads the note first. “Write a sign for my shop.” The answer is a small sign: BLUE DOOR. Now it knows.
The last part is what you can do. The notebook grows so you can read it, and one faded line is crossed out and fades away. Then a tag marked Private is tapped: the notebook closes, and nothing moves from the notebook to the chat. The AI keeps nothing. The app keeps a note. You hold the pen.
That is the honest shape of AI memory in 2026. The model keeps nothing between chats. The app keeps notes, reads them back, and lets you edit them. In the small AI apps I build for daily work, the notes an agent keeps are files you can open and read, for the same reason.
What you will learn
- Why the AI “forgets” you between chats: the model starts each chat empty, by design.
- What memory really is: a note the app writes at the end of a chat and puts back at the start of the next.
- What you control: see the notes, edit them, pause or wipe them, or open a private chat that writes none.
- Why that control matters: a note can be wrong, or planted, and the AI will trust it next time unless you read it.
The numbers behind the picture
Nothing with a digit is on screen. The dates and numbers belong here.
- 2025: ChatGPT began to use all past chats in April (reported). Gemini turned memory of past chats on by default in August. Claude got memory for Team plans in September and for Pro and Max in October. The Mem0 research paper reported a 26% relative gain on a long-conversation test with far fewer tokens. (Press on OpenAI, Apr 2025; Google help, Aug 2025; Anthropic, Sep and Oct 2025; Mem0, Apr 2025)
- 2026: Claude memory reached free users in March, was rebuilt as “individual, categorized entries” in July, and is on by default since August. Gemini renamed its feature “memories” in March. OpenAI announced a rebuilt memory system in June with a readable summary and a one-step “delete and turn off” control (reported). On the LongMemEval-V2 benchmark, the best system scored 72.5% in May: far from perfect. Security studies showed planted memories staying active after a hundred sessions. (Anthropic release notes, 2026; Google help, Mar 2026; press on OpenAI, Jun 2026; LongMemEval-V2, May 2026; memory-poisoning studies, 2026)
- 2027, my guess: Memory becomes the normal setting everywhere, and the useful question changes from “does it remember?” to “what did it write down, and who can see it?” I expect memory to be a place you open and read, like a shared notebook, not a hidden store.
Questions people ask
Does ChatGPT, Claude or Gemini remember previous conversations?
The model does not. Each chat starts empty. The app can remember for it: ChatGPT, Claude and Gemini each keep notes about you and put them back into new chats. In 2026 that is on by default in all three for most users, and each one lets you look at the notes and switch the feature off.
Why did the AI forget something I told it in the same chat?
That is a different thing. Inside one chat the AI reads from a window of limited size; in a very long chat, old lines fall out of it or get summarised. Memory across chats is the notebook in this film. Forgetting inside a chat is the window, and it has its own fix: say the important thing again, near the end.
Can I see and change what it remembers?
Yes. Claude shows its memory as entries you can read, edit, pause or reset. Gemini shows saved memories with the same controls and keeps temporary chats out of them. ChatGPT shows a memory summary and a one-step delete-and-turn-off control (reported, June 2026). A private or temporary chat writes nothing to memory in any of the three.
Is AI memory a risk for a business?
It can be. Security researchers have planted false memories through a web page or a document and watched the AI act on them in later sessions. The rule I use in the apps I build is simple: an agent’s memory is a file a person can open, and anything that touches money or customers is checked against the real record, not the note.
Further reading
- Anthropic developer docs, “Memory tool”. Memory as files that persist between sessions, separate from the context window. platform.claude.com
- Anthropic, “Bringing memory to teams at work”, September 2025, with the October 2025 update. claude.com
- Claude Help Center, release notes (memory for free users, March 2026; categorised entries, July 2026; on by default, August 2026). support.claude.com
- Google Gemini Apps Help, memory and temporary chats (2025 and 2026 pages). support.google.com
- Press on ChatGPT memory: Neowin and Decrypt, 11 April 2025 (all past chats); coverage of the June 2026 rebuild. OpenAI’s own pages were not reachable when this page was written.
- Packer and others, “MemGPT: Towards LLMs as Operating Systems”, 2023. The idea of tiered memory. arxiv.org
- Mem0, “Building Production-Ready AI Agents with Scalable Long-Term Memory”, April 2025. arxiv.org
- Anthropic, “Effective context engineering for AI agents”, September 2025. Structured note-taking as the way agents keep memory across long tasks. www.anthropic.com
Every agent I build keeps its memory as plain files its owner can open, and nothing it remembers can move money on its own (how I can help). If your team wants AI that remembers the right things, tell me about the work.
Transcript
Every line of text shown on screen, in order. Nothing is spoken. Each label stays at the bottom of the screen for at least four seconds.
Opening
- Logo: vai SoftLab (small, no title line)
Scene one: the question
Alone on a dark screen, the biggest text in the film:
- Does the AI remember me tomorrow?
Scene two: today
A chat window with a small grey tag, a small gold sun in the top corner, a small person on the left:
- Chat
Typed, letter by letter, then sent:
- My shop is called Blue Door.
A grey answer card; only the first words can be read:
- Nice name.
Label: Today it knows your shop.
The sun sinks and a grey moon rises. The words fade, the chat window folds to a thin line and disappears. The screen is dark except the moon.
Label: The chat ends. It keeps nothing.
Scene three: tomorrow, without memory
The sun rises. A new chat window opens in the same place, empty. Typed, letter by letter, then sent:
- Write a sign for my shop.
A grey answer card:
- Which shop?
Label: Tomorrow it starts empty.
The person on the left dims a little. The picture fades.
Scene four: with memory
The first day again: the chat window with the line already in it:
- My shop is called Blue Door.
Beside it, on the right, a small closed notebook with a gold tag:
- Memory
The notebook opens and a gold pen writes one line:
- Shop: Blue Door
Label: Memory is a note the app keeps.
The sun sinks and the moon rises. The chat window folds away. The notebook stays, open, with its one gold line.
The sun rises. A new empty chat window opens. Before anyone types, the gold line slides out of the notebook into the top of the chat:
- Shop: Blue Door
Label: It reads the note first.
Typed, letter by letter, then sent:
- Write a sign for my shop.
A gold answer card, a small shop sign with two words:
- BLUE DOOR
Label: Now it knows.
Scene five: you hold the pen
The hand cursor moves to the notebook. The notebook grows so its lines can be read: “Shop: Blue Door” and one faded grey line. The hand draws a line through the faded line; it fades away.
Label: You can read, change or close it.
A small grey tag appears over the chat window and the hand taps it:
- Private
The chat clears, the notebook closes its cover and dims. Nothing moves from the notebook to the chat.
Scene six: the answer
Dark screen. Big, two lines:
- The AI keeps nothing.
- The app keeps a note. You hold the pen.
Under it, small: the open notebook with its one gold line and a small gold pen.
Closing
- Logo: vai SoftLab
- vaisoftlab.com
Have an app idea, a stuck app, or a daily job you want an AI agent to do?
Write three lines. I reply within one working day with a plain answer.
