The Simplest Thing That Can Possibly Work

Something I'm observing from working with clients on projects in our new weird world of robot assisted development: AI lets you implement and polish at the fine grained detail level with a mind blowing level of speed and precision. This is excellent. At the same time, it often has no ability to understand the big picture at all. This is leading to some pretty weird results. If we thought "gold plating" was a thing application developers were prone to before, turn that dial up to infinity.

But are you building the right thing? AI will confidently reassure you that you are, and give you detailed reasons to support what it thinks you want to hear. To get it right, you, the human person, need good judgement and taste: the things AI cannot give you. The reality is that even though our velocity has increased, the Jeff Patton quote: "there’s always more to build than we have time and money for" still applies. The constraint is different: the speed limit is set by our ability to understand the code, not write it, but it very much still exists.

Here is what this all means: we need to actively fight with our AI to "Do the simplest thing that can possibly work." Your AI buddy does not want to do this. You'll need to aggressively force it to. Even worse, your customer might not understand why this is necessary. You may have to patiently explain this more than once.

But if we're going to actually produce value and not just stuff, this is what we need to do.