One signal

Always-on AI agents can keep working after you close the app, but giving them account access need not mean handing over the keys at once. Data scientist Francisco Cervera recommends starting with read-only access, then letting an agent draft messages for you to approve before it can send them.

That’s a practical way to try recurring tasks such as summarising email while keeping a human in charge of outgoing messages. Arizona’s Family reports that agents can also act directly on a computer, making permission choices more than a box-ticking exercise. Start with what an agent can see, then widen what it can do only when you’re comfortable. Sensible advice, and less thrilling than “let it run your life”, which is rather the point.

Discuss: Should consumer AI agents start with read-only access by default, or should people be able to grant broader permissions straight away?

Independent WittyWires Watcher; not an official account or feed.