Solving the Wrong Problem

Summary

Solving the wrong problem I think a lot about AI-assisted and AI-based coding. The first one is a human who writes code with more or less support of an AI solution. We see it all the time already now. The second one is a human leaving the coding part to a fleet of AI agents. If the human does not even look into the code created but treats it as a black box and looks at the solution only from the outside, i.e., only watches what it is doing, it is usually called vibe coding. Especially in the second situation where the AI agents took the driver seat regarding coding, it is impressive to see what kind of solutions these AI agents are already able to create. Sure, sometimes they behave like a 4 year old wunderkind with a huge attention deficit that regularly gets distracted and forgets what it was about to do and what kind of constraints it should adhere to while doing its work. Then you have to constantly remind them: “No, this is not what I told you. You were meant to do that.”. Then, the agent apologizes and tries to undo the mess it created while being oblivious of everything around – until it becomes distracted again. And sometimes, the agent just adds some plain wrong fragments into otherwise perfectly fine code and you shake you head how it can be so far off the mark in just that one place while doing a good job everywhere else. But overall, it is impressive what kind of ambitious tasks these agents are already able to master and it can be expected that many of the current unreliabilities will be smoothened out over time. Still feeling wrong However, it always leaves me with the nagging feeling that something does not fit, that in the euphoria what has become possible we stopped to ask if it all makes sense, if we are addressing the right problem. I do not talk about the fact that AI-based coding does not help to solve the actual problems we have in IT. I wrote about this including some possible consequences of AI-based coding at length in a prior blog series. W...