# A proposal, not a verdict

> My agent can propose and justify what to delete. The decision is always mine.

Author: Aleš Nypl · https://engineloop.cz/ales-nypl

An agent goes through my email, my working inbox, knowledge base, files and other sources. It finds invoices that need paying, deadlines that are coming up, and a pile of things nobody needs. Then it sends me a list: keep this, file that, bin the rest. Only then do I tick the items that are right.

It isn't a matter of distrusting the technology. It is a matter of dividing the roles. AI reads quickly and never gets bored. With good rules and enough context, it forgets very little. But the consequences fall on whoever decides, and that's me.

## Three rules that follow from this

- **AI proposes, a person approves.** The agent sends nothing and deletes nothing. The originals stay in the archive.
- **Approving has to be cheap and fast.** I approve in batches, in a simple interface, in a few minutes. If it were a chore I'd stop, and the safeguard would guard nothing.
- **A rule mustn't depend on my discipline.** Routine is watched by scripts and AI, not by my memory. I save my memory and thinking for creative work.

My advice to a company comes down to one thing. Find the places where a person has to decide, and keep them short. Everywhere AI is only proposing, you can move quickly.

A proposal isn't a verdict. That's exactly why I can let it be ambitious.