Two hundred and fifty euros and nothing more: an agent's starting capital

My agents can now spend real money. One simple rule comes with it: every agent with a good idea gets 250 euros of starting capital, in total, and nothing more.
It is not a monthly budget and it is not topped up. When it is gone, the spending stops. If the idea has earned money or needs more, it comes to me.
What it can be spent on
As of this week, agents can buy three kinds of things:
- a domain name;
- a rented server;
- AI capacity, meaning tokens or a share of an AI subscription.
Everything stays within the ceilings that already exist, per day and per ticket. A purchase still needs my yes, for as long as the tooling asks for it.
What is not included
Hiring people, signing contracts, acting in someone else's name and anything legal or irreversible: that stays with me.
Why one hard limit
A monthly budget invites running on. A one-time amount forces a choice: this idea, this expense, now. The risk is small and measurable, and the agent learns to run a venture with real money instead of a worked example.
The limit is also honest about what is not there yet. Today each ceiling lives in its own tool: the server has its own, the AI capacity has its own. The total per idea will live in one ledger, where every expense is booked with a reason. That is being worked on.
What you take from this
- Give an agent an amount in total, not per period.
- Say up front what it may and may not be spent on.
- Keep the limit in the tool, not in the instruction.
- Let a human say yes while you are still building trust.
Which idea in your organisation would you dare to fund with a small, fixed amount to try something real? I am happy to spend a day on that with you, during an Expert Day.
See also
An agent orders a server. What I let it do, and what I didn't
Little by little I give my agents more room. The first real step was a server. Dex, my Operations agent, ordered one, installed NonStopWorker on it himself and showed that it ran. From order to login took under two…
A chat window answers, an agent does the work
Many people now use AI every day. They ask a question, get an answer, copy it, paste it somewhere else and do the rest themselves. That is useful. It is also something different from an agent.
An agent that reads an email also reads the instructions inside it
An agent that reads an email also reads the instructions inside it.


