News · September 8, 2026 · 1 min read
Ask First: How Much Should Your AI Do Without You?
An agent that can post, pay and send email on your behalf needs one setting more than any other: when to check with you first.
By Hamilton Cheong
The moment an AI can act, not just answer, a new question appears. Should it send that email? Publish that post? Delete that file?
Most products pick an answer for you. Some ask about everything, until you stop reading the prompts. Others ask about nothing, until something goes wrong.
We think the answer belongs to you.
Three settings
When you set up your computer, you choose how much your agent may do on its own.
Ask every time. Nothing happens without your yes.
Ask before anything that changes things. Reading and researching are free; sending, posting and deleting wait for you.
Act freely. For people who have seen the work and trust it.
You can change it any time, and you can go further: allow some capabilities, block others, and require a check for the ones that matter most.
Approvals that fit in a second
Asking first only works if saying yes is easy. So an approval is a small card that shows exactly what your agent wants to do, with Approve and Decline beside it. It appears in the chat on your computer, and soon in the phone app and on Cloudy.
If nobody answers in time, the answer is no. An agent that goes ahead because you were busy is not asking. It is informing.
Trust is a record
Every run is kept as a session you can read afterwards: each step, each tool it used, each result. What your agent did is a log, not a claim.
“Start with ask first. Loosen it when the work has earned it.”
That is how you would treat a new hire. It is how we think you should treat an AI one.
