Hello {{first_name | Reader}},
Two questions came up this month that sound unrelated. One was asked out loud in a room full of professionals: should I actually be worried about AI agents? The other is the one you ask yourself at six o'clock, when the thing you planned to do is still untouched. This newsletter has been quiet since June. It comes back with both.
✍️New this week: should you be worried about AI agents?
This month Anthropic published a report on criminals and state actors pointing AI agents at real targets, and Dario Amodei asked the industry to pace the frontier rather than halt it. Neither is machines going rogue. Humans chose the targets and read the results, and that distinction changes what you should actually do about it on Monday. The post is the honest answer, including the one security habit worth fixing this week.
⌛Also worth your time: the sentence everyone leaves out
The average professional is interrupted every two minutes. That is 275 times in a working day. Most people who try to hand their inbox to an AI write one line and hope. The instruction that works has five parts, and the one almost everyone drops is the boundary: the sentence telling the agent where to stop.
📰 Story worth watching
On 8 September, Meta launched Muse, a personal AI agent that works across your email, browser and apps. It asks for approval before it sends an email or makes a purchase, and it keeps a record of everything it has done and plans to do.
That is the boundary sentence from post 7, built into the product. The assistant you use at work may not come with it. Then you write it yourself, and it takes one sentence.
That's it for this week.
If you found this useful, forward ⏩⏩ it to a colleague trying to figure out how AI agents work or reply to this email — I read everything.
See you soon, Linda