Your computer should run your business, not wait for you to drive it
Two bets that point in the same direction
Mark Gurman reported this week that Apple may build AI that can fully operate your iPhone, iPad, and Mac on your behalf. The detail that matters is where it would run: on the device. Local. Private. Your own machine doing the work instead of a server farm somewhere.
At the same time, the AI engineering community has a new fixation it calls loopcraft. The idea is to stop prompting agents one request at a time and start building loops that run on their own, continuing as long as the work takes with nobody in the middle.
Put on-device autonomy next to loops that run themselves and you get something genuinely new: a computer that does real work for you, locally and privately, without being driven.
A loop needs a trigger
There is a question buried in all of this that almost nobody asks. A loop has to start somewhere. Something has to set it off. What is the trigger?
For the loopcraft crowd, the trigger is a goal they set: ship this feature, run this experiment, refactor this module. That works because they are engineers architecting their own loops. Most people are not, and never will be.
So look at how most knowledge workers actually spend a day. They don’t choose their own adventure at the office. They don’t stroll in and decide to do some marketing or write some code. Their work lands in their inbox, sent by customers, colleagues, and their boss. Most days are spent tackling that incoming work, and often falling behind, because the inbox fills up faster than it can be cleared.
If most work arrives rather than originates, then the most valuable loop is the one that starts with communication. Something comes in, and it gets handled. The trigger has been sitting in plain sight the whole time.
We built that loop
This is the loop we have been building at this+that. It reads what arrives across every channel you use, email, Slack, Teams, Telegram, and more, turns each thread into clean tasks, and acts on them, grounded in enough context about you and your work to do it safely. It runs on its own, surfacing the decisions that need you and handling the rest. It is the same reason we think fit-for-purpose AI beats a chat box you have to visit, an argument we made in the second wave of AI.
Today that loop runs in the cloud. The horizon Apple is pointing at is running it on your own silicon, keeping routine work on the device and reaching for a larger model only when the reasoning calls for it. That is the direction we are heading.
A machine that runs your business
Stack it up. On-device intelligence, loops that run without supervision, and a trigger that fires on the work that actually arrives. Get there, and the computer stops being a thing you drive, a surface you click into, instruct, and type at all day. It becomes something closer to a machine that quietly runs your business in the background, bringing you the decisions worth your attention and handling the rest.
That is the one we are building toward.