Notes · N1
Why your AI should ask permission
An AI that acts on its own is a liability. An AI that asks first is an employee you can trust with real work.
Every founder I talk to has the same quiet fear about AI: not that it won’t work, but that it will work too well, in the wrong direction, while nobody is watching. A wrong email sent to a client. A refund issued twice. A file deleted because a prompt was vague.
The usual answer is to keep AI in a chat window, where it can only suggest. That is safe, and it is also why most companies get almost nothing from it. Suggestions still need a person to copy, paste, check and send. The busywork stays.
There is a better design, and it is old: the approval gate. Accountants have used it for centuries. A junior prepares, a senior approves, and the ledger records who did what.
Apply that to AI. Let it research, draft and prepare. Anything that spends money, contacts someone outside the company or deletes something waits for a person to say yes.
Then keep a record of what it did and why. Not because anyone expects fraud, but because a system you can read is a system you can improve.
The result is counterintuitive. Adding a permission step makes AI more useful, not less, because people finally trust it with real work. You stop reviewing everything and start reviewing only what matters: the moments with consequences.
If you are evaluating AI for your business, ask one question before any other: what can it do without asking me? If the answer is “anything”, walk away. If the answer is “nothing that matters”, you are looking at something you can build on.
Douglas H.
Next step
If this sounds like your company, bring the problem to a briefing.
Request a systems briefing