Most AI tools know only what you tell them. You paste in context, explain the situation, describe your workflow — and the model responds to that snapshot. ChatGPT’s new computer history feature flips that entirely. Instead of waiting for you to explain what you’re doing, it watches.
The feature is currently limited to the ChatGPT desktop app on Mac, and requires a Pro, Business, or Enterprise plan. It’s also unavailable in Europe, the UK, and Switzerland for now. That’s a narrow audience — but the underlying idea is broad enough to matter to anyone thinking about how AI fits into their actual day.
What Computer History Actually Does
When you enable it, ChatGPT passively monitors activity across whichever apps you choose — browser tabs, documents, file management — without recording your screen or audio as video. It’s reading what’s on your screen, tracking the sequence of what you do, and building a running log of your work sessions.
That changes the nature of what you can ask. Instead of pasting a bunch of context into a chat and saying “here’s what I’ve been doing, help me write this up,” you can just ask “what was I working on this morning?” and get a coherent summary back. More usefully, you can ask it to package something you just did into a reusable workflow.
Say you spend an hour every Monday pulling data from a report, reformatting it in a doc, and emailing a summary to your team. If ChatGPT has watched you do that a few times, it can describe the workflow precisely — not because you explained it, but because it observed it. Then it can turn that into a repeatable skill you can trigger later.
The Automation Audit Nobody Asked For (But Needed)
The most underrated use here isn’t memory or workflow capture. It’s the ability to ask a simple question: which parts of what I do could be automated?
This is a question most people can’t answer well on their own. You’re too close to your own habits. The stuff that feels routine is invisible. But an outside observer watching you work for a week would spot patterns immediately — the repetitive copy-paste, the three steps you always take in the same order, the formatting you redo every time.
That’s what this feature can surface. It’s not scanning for dramatic inefficiencies. It’s noticing that you always open the same two tabs before writing a brief, or that you rename every downloaded file before moving it to the same folder. Those small loops are exactly what agent-based automation handles well — once someone identifies them.
Having AI identify them for you, based on what it actually observed rather than what you remembered to mention, is genuinely new.
The Privacy Tradeoff Is Real
None of this comes free. Giving any application persistent visibility into your screen activity is a significant decision, and it’s worth thinking through carefully rather than just clicking through the setup.
A few things worth considering before you turn it on:
- Scope it tightly. The feature lets you whitelist specific apps rather than opening it up to everything. Start with two or three low-sensitivity tools — a writing app, a project folder — before you add your email client or anything with personal or financial data.
- Exclude communication apps. Anything involving other people — messaging apps, video calls, email threads — introduces consent issues that go beyond your own privacy. The setup flow mentions this, but it’s easy to dismiss.
- Separate work from personal. If your browser is where you do both, think hard about what you’re handing over. A browser tab history is a surprisingly complete picture of a person’s day.
Microsoft got torched a couple of years ago for a similar concept called Recall. The backlash was loud enough that they pulled back significantly. The underlying technology wasn’t wrong; the rollout was tin-eared about what people actually feel when they picture a company’s server knowing everything they looked at all day. OpenAI is threading the same needle, and whether they’ve done it better remains to be seen.
Why This Matters Beyond the Feature Itself
The harder problem with agentic AI has never been capability. The models are good. The problem has been relevance — figuring out which parts of your specific work are worth handing off to an agent, and then giving that agent enough context to do the job without a lot of hand-holding.
Computer history is a direct attack on that problem. It generates context automatically, from real behavior, without requiring you to be a prompt engineer or know what a vector database is. That matters because the people who stand to gain the most from AI automation are often the least equipped to describe their workflows in abstract terms.
If this pattern catches on — and it will, because the competitive pressure among AI companies right now is intense — expect every major AI platform to ship a version of this within the year. The question each one will have to answer is how much trust they’ve earned with users before asking for that level of access.
How to Try It Without Overcommitting
If you’re on a qualifying plan and want to test it, here’s a low-risk way to start:
- Enable computer history for only a text editor and your file system — nothing else.
- Do a contained piece of work: draft a document, organize some files, export something.
- Ask ChatGPT what it saw and whether any part of that is worth automating.
- Evaluate the output. Is it accurate? Is it useful? Does anything feel like it captured more than you expected?
That feedback loop will tell you more than any review can about whether the tradeoff is worth it for you specifically. The feature is powerful. Whether it’s the right tool for your situation depends on what you’re willing to share — and what you’d actually do with the automations it finds.