AI agents just got their own wallets. Here is the one approval I will never hand over.

AI agents just got their own wallets. Cloudflare, one of the companies that runs big parts of the internet, spent a whole week this month launching tools built for AI agents, and the headline for me was wallets: a way for an AI agent to pay for things on its own, with no human touching the payment.

I run an organization of AI agents. They write my content, answer my customers, and build my products. If anyone should be first in line for this, it is me. And my answer is still no. Not because the tech is bad, but because of a rule I use every single day, and I think it is the most useful thing I can hand you this week.

What did Cloudflare actually launch?

In plain words: until now, an AI agent that needed to buy something had to stop and wait for a person. Cloudflare built the plumbing that removes that stop. An agent can hold money and spend it by itself, the same way it can already write an email by itself.

That is a real milestone. The companies building the roads of the internet are now assuming AI agents will be customers, not just tools. Whatever you think of it, that direction is set.

Why would anyone want an AI that spends money?

Because asking for permission is slow. If your agent hits a paywall, needs a stock photo, or has to pay a small fee to finish a job, every “wait for the human” moment is an hour lost. Remove the human and the work just flows.

That argument is completely true. And it is exactly the argument that should make you careful, because the pitch for removing a safety step is always speed.

The two-pile rule: can I undo it?

Everything my agents do gets sorted into two piles.

Pile one: things I can undo. A bad draft gets rewritten. A wrong page gets fixed. A clumsy reply gets improved tomorrow. My agents do all of this without asking me, all day, and that freedom is why the setup works at all.

Pile two: things I cannot undo. Money sent. An email sent. A post published. Something deleted for good. For anything in this pile, the agent prepares everything down to the last detail and then waits for my yes.

Think of it like a new employee with a company card. You would not skip the spending talk on day one. It is not about distrust, it is about the fact that some mistakes cannot be walked back.

Why does the money gate stay human?

Money is the clearest pile-two case there is. So my agents can research a purchase, compare options, fill in every field, and put the finished payment in front of me. The actual pressing of pay stays mine.

Here is the honest part. An agent spending on its own does not scare me on day one, when I would be watching every move. It scares me later, when things have gone well for months and I have stopped looking. That is when a quiet wrong payment becomes an expensive one. The gate is not there for the week I am paying attention. It is there for the week I am not.

What should you decide this week?

If you use AI in your business, or you are about to, decide this before the tools decide it for you:

1. Which actions in your business can be undone, and which cannot?
2. Does anything in the cannot pile currently happen without a human saying yes?
3. If a tool offers to remove that yes in exchange for speed, is the time saved worth the one mistake you cannot take back?

AI agents with wallets are coming either way. Mine will keep preparing the payments. I will keep pressing pay.

I build AI agents that run all parts of my online business. Check them out at pelleforsman.com.