Howdy wizards,
Here’s what’s brewing in AI.
The big thing
ChatGPT will now log every click and keystroke on your Mac if you let it.
It’s a new feature called Computer History that records your clicks, typing, keyboard shortcuts and apps.
You can ask about anything you were working on in the past, and it can answer you, on a timeline. E.g. what was I working on before lunch on Monday.
When it sees you doing the same thing over and over, it offers to turn that into a skill or automation. It uses Codex to write it for you.
The way it captures your work is through the same accessibility layer as screen readers do on Mac — although it doesn’t take screenshots.
Privacy wise, they had the common sense to make this feature turned off by default.
You can exclude specific apps or sites, or only allow it access to the ones you pick. It also backs off if you browse in incognito mode.
The raw record of what you typed doesn’t get used for training, but there’s a but: it goes to OpenAI’s servers for a little bit. Enough that it can generate a summary of it. Those summaries are stored as memories. And what’s their policy on memories?
“When ChatGPT or Codex uses a memory in a future chat, relevant memory contents and interaction events may be included as context. This chat content may be used to improve OpenAI models if allowed by your ChatGPT data controls.”
They can’t train models on your clicks and strokes. They can train on ChatGPT’s stored memory of them.
And they don’t say whether the processes you end up turning into skills or automations are trained on. Though I assume millions of proven, executable processes are a tempting data source.
It’s currently available for paid ChatGPT users, Mac only and, what do you know, not available in EU yet (as per usual).
Why it matters
Satya Nadella recently said enterprises using AI models are paying twice. Once for the tokens. Once with the proprietary knowledge that makes the models useful.
OpenAI, and all other AI labs, want more real data on how humans do work.
Enterprise is their most lucrative customer segment. They are out of internet data to train on.
It’s the name of the game now: Google just bought the internal business data of bankrupt Spirit Airlines for $10m. It consists of 100m emails and 500m Teams messages. Millions of examples of humans doing real work.
And while OpenAI doesn’t train on business accounts, it does on personal Pro accounts. And more than half of enterprise AI conversations run on personal plans.
I don’t think it’s a stretch to say Computer History is the same harvesting logic, disguised as a useful feature.
Is it useful, though? I think yes, in a similar way your browser history is. Convenient, for stuff you’ve forgotten to save. But also not really how you get work done.
Frequency ≠ importance.
The place where I actually want AI to record everything I do is meetings. That’s a goldmine of context for anyone’s work. Granola is the only note taker I trust, and also this week’s sponsor:
IN PARTNERSHIP WITH GRANOLA
Your meetings are your richest source of context. Use Granola for a while and you’re sitting on a goldmine: months of decisions, discussions, and commitments, all captured.
But how often do you actually go back and search through them? Basically never. It’s a hassle, and your notes have no way of talking back to you.
Granola Chat fixes that: an AI chat that works across all your notes, personal and team. Ask it a question and it searches everything: your notes, your Team Space, and privately shared notes.
Quick question? Fast answer. Deeper analysis, like spotting trends across fifty sales calls? It handles that too, and every answer links back to the source with inline citations.
All that context you’ve been building compounds now. Chat turns your archive into an always-on analyst that already knows your business.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
All the small things
New tools & product features
Cursor’s GitHub competitor is released, months early. Origin does repos, pull requests and code browsing, with an agent attached to each. It syncs both ways with GitHub, which stays the source of truth, which makes the threshold really low to try it for those already using Cursor. I might give it a try soon.
Vercel is offering up to $1m in bounties to anyone who can break out of its sandbox. The target is the Firecracker-based sandbox its AI agents run inside. The challenge lasts for two weeks, open to anyone. Inviting users to break your product isn’t new, and is possibly more relevant than ever.
Industry moves
Nvidia is co-signing $105B so OpenAI can build on a Cold War uranium site in Ohio. The factory previously made fuel for nuclear bombs, which took as much electricity as a city, and those power lines are still there. Power like that is now harder to get than chips. OpenAI doesn’t have that much cash, so Nvidia co-signs the loans that build the site. Then it gets filled with 1.5 million Nvidia chips, a separate and even bigger bill, paid to Nvidia. Something smells circular.
Cold brew
❦
You are a delight.
What's your verdict on today's email?
See which AI use cases are paying off with Context Windows Pro
Most companies pick AI use cases by brainstorming internally. 90% of those initiatives fail.
I’ve created Context Windows just so you can pick the winners.
🟦 Find high-performing use cases from 2,000+ companies at contextwindows.ai, or book a demo with me
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.