The year is almost over and we can take stock: it just didn't happen!

In Trouw this week, I read a column by Ilyaz Nasrullah about the podcast Shell Game. In it, creator Evan Ratliff tries to set up a business that runs entirely on AI agents. The goal: one million dollars in revenue. The result: chaos.

From coming up with a business name to building a website – nothing goes smoothly. An AI agent calls an applicant on a Sunday evening, even though the interview was scheduled for Monday morning. When the (human) applicant asks if something went wrong, the agent apologizes profusely... only to then just carry on with the conversation.

Hilarious. But also telling.

98% don't make it to the workplace

This story is not an isolated incident. A few days earlier, Volkskrant headlined: "98% of AI applications don't make it to the workplace."

Ninety-eight percent. That's not a bug, that's a pattern.

The reason? Not the technology. The approach.

Organizations start with AI because of the hype. Because of "we need to do something with AI". Because of the technology. But they forget the most important question: what problem is this actually a solution for?

And that is exactly where things go wrong.

We made the same mistake

I cannot pretend we are above this. We also flew off the track with plenty of enthusiasm.

Since GPT-2 came out in 2019, we have been experimenting with AI. Thousands of tests. Built our own agents. And at a certain point we thought: let's give one agent a lot of responsibility over a large dataset.

The result? Hallucinations. Inconsistency. Misclassifications. Wrong decisions.

We hoped that OpenAI or Anthropic would fix this. That didn't happen. These limitations still exist.

The wild horse!

We treat AI like a wild horse. Incredibly powerful. But without reins, dangerous and unpredictable.

The mistake Ratliff makes in Shell Game – and that we made too – is treating AI like a human. Giving an agent a name, a role, a function. Expecting it to perform complex tasks independently.

Want to know what lessons we learned from this?
And how we are actually successfully using AI now?

Sign up for the webinar

On January 20 and 22, we are hosting a 30-minute webinar:
AI: From nice-to-have to value.

No hype. Just: what works, what doesn't, and how you can be part of that 2% that does succeed.

Tuesday, January 20 | 10:00 AM Sign up
Thursday, January 22 | 4:00 PM Sign up