One sentence made me think this week. Dr.-Ing. Maurice Preidel, an AI expert and friend, rightly reminded me: describe the desired outcome to the AI, not the path to get there.
And in fact, I often do exactly that. For my small autonomous robot engineering project, I even made it a guiding principle. Still, I continue to define certain guardrails and workflows myself. And there are good reasons for that, more layered than they seem at first glance.
One of them: trust doesn’t arise on its own. In my article “AI in engineering needs rules – and trust”, I put it like this:
Responsibility requires understanding: which data, rules, contexts, checks and approvals were involved in a result?
In other words: I don’t have to prescribe every technical step to the AI. But I need something like “minimum viable governance” – enough structure for the process to remain traceable, verifiable and accountable.
I will soon discuss this with Maurice in more depth. I’m curious where our perspectives overlap – and where they don’t.
The path to an AI task doesn’t always have to be prescribed – but it doesn’t work entirely without guardrails either. A minimum of structure is needed so that results remain traceable and accountable.
