Cabinet wants to remove barriers to AI for small businesses. What now?

Friday afternoon, quarter past three. Between two client emails, a newsletter lands with a headline saying the cabinet wants to boost AI. The third one this week. You click it away and carry on with the VAT documents of a client who is once again handing them in at the last minute.

Still, this news is worth reading, if only to know what has and has not been decided.

What happened

On 6 November 2026, the Council of Ministers approved the Cabinet Vision on AI by State Secretary Aerdts. In it, the cabinet says it is choosing AI that moves the Netherlands forward, and it wants, among other things, to remove barriers for small and medium-sized businesses. The vision rests on three pillars. In early 2027, a plan for AI infrastructure and computing power will follow. You can read about it in the government's announcement (https://www.rijksoverheid.nl/actueel/nieuws/2026/11/06/kabinet-kiest-voor-ai-die-nederland-vooruithelpt) and in Computable's coverage (https://www.computable.nl/2026/11/06/kabinet-wil-ai-ontwikkeling-meer-sturen/).

Note what it is: a vision. Not a law, not a regulation, not a subsidy you can apply for tomorrow. What will concretely come out of it is not yet known. That plan from early 2027 is the first moment when more will become clear, and even about that, all we know now is that it is coming.

What it means for you

If the barriers get lower, more businesses will start. That is fine. The risk usually lies not in starting, but in letting things through.

Take a bookkeeping firm with forty client emails a week. Someone has connected an AI assistant that drafts replies and sends payment reminders. The first few weeks go well, so the checking gets looser. Then one Tuesday a reminder goes out with the wrong amount, and it is already gone before anyone sees it. The damage is not the reminder itself. It is the phone call from the client, and the question of whether the other thirty-nine emails that week were correct.

These are not exotic scenarios. It is what happens when a task is handed over without it being clear in advance what will come out.

Three questions before you hand over a task

Regardless of what the cabinet ultimately decides, these three questions help you choose.

1. What can go wrong, and who notices first? Is it you, or the client?
2. Do I see beforehand what is going to happen? Not what the tool intends to do, but what the outcome would be: which amount, which recipient, which text.
3. Who decides about money and about everything that goes outside the company? If the answer is that the software does that itself, you are in the situation of the firm above.

If you can give a clear answer to all three, the task is ready to be handed over. If you can't, that is not a reason to wait for policy. It is a reason to make the task smaller.

How that works at Stilops

Stilops provides AI staff per task, for those without an IT department. The three questions above are also how it is built.

With us, you don't approve an intention, you approve an outcome. Before a task actually runs, the agent does a test run. In it you see what would happen: the amounts, the recipients, the texts. Only when that is right do you give your approval. A human decides about money and about anything that goes outside. That moment does not get lighter the longer the agent has been running. After a good week you might think you can skip it, and that is exactly what we don't want to happen.

You see per task what it costs, so you can weigh whether it is worth it.

The honest downside: this is slower than an agent that handles everything itself. There is always a moment of you in between. For work where a mistake costs money or trust, we think that is the right price, but it is a choice and not the best one for every task.

We don't have customer cases to show yet, and we are not going to invent any. If you want to see what the test run looks like, you can go through the demo without an account.

What you can do this week

You don't have to wait for the plan of early 2027 to do something with this. Pick one recurring task, for example following up on open quotes or sending payment reminders. Write down for a week how many minutes it costs you. Then decide which part of that task a human must decide: the amount, the recipient, or the tone of the text.

Then you know two things. How many hours there are to gain, and where the approval moment belongs. Policy can lower the threshold, but what your customers receive from you is still your decision.