Starting point: two hours—let’s give it a try

It’s Tuesday, 4:00 p.m. Production is running—not very smoothly, but it’s running. You’re called into a meeting room. On the table: an open laptop, access to Copilot, an in-house agent (IT) cobbled together by the IT department, and sometimes a large language model (LLM) hosted on-premises for sensitive data. The manager sets the stage: “We have two hours. We need something we can show people.” No more, no less.

And that’s where it all comes down to: not the technical aspects, but the decision-making. In two hours, the risk isn’t failing to code, but coding too quickly… without having defined what really matters. Because when the tool runs very fast, your ambiguities become immediately apparent.

The team in front of you is competent. Automation specialists, methods engineers, and a data engineer who has already completed three proofs of...