“I don’t need an AI that produces movies for me. I need one for my taxes and my back office.” - Richard Seidl
As I write this column, somewhere an AI is generating a picture of an astronaut riding a unicorn through space. Meanwhile, in a few minutes I’ll be sitting in front of my taxes, travel expense reports, correspondence with the authorities, and a few legal documents. And I ask myself: why, actually?
We’re celebrating a lot of AI-generated text, music, and videos right now. Impressive, I enjoy watching it too. But: are we solving the problems that actually rob us of sleep?
In software development, things are running along similar lines. We watch AI generate code and whip up a new app in no time. Code quality? Whatever. It runs. Meanwhile, teams are suffocating in legacy code that nobody understands or can maintain anymore. We dream of autonomous test agents while our test strategy is a mess and flaky tests have broken the trust in our pipeline.
The problem isn’t the technology at all. The problem is our focus. We optimize for demos, not for everyday work. For LinkedIn posts, not for productivity. For the spectacular, not for the useful.
What do we really need? Where does it actually hurt? What is it that I don’t want to do anymore? Matching my test cases against a hundred pages of requirements specification! An AI that understands why this one test has been failing sporadically for three months. One that shows me not just syntax errors in a code review, but inconsistencies in the business logic.
Where I personally don’t want any AI, at least not as a first step: test case creation. I simply love taking a user story or a requirement and thinking about how I could test it well. It’s just fun and I learn something along the way. Sure, I’ll happily bring in the AI afterwards as a cross-check, but not for the initial creation.
Do the enjoyable things yourself. Hand the unpleasant ones over to the AI. That’s what we should go by, and not just in software development.
The good news: this focus is slowly emerging. The same old AI LinkedIn posts are starting to get boring. We’re beginning to understand that the value of technology doesn’t lie in its brilliance, but in the problems it solves.
Quality also means doing the right thing right. That applies to AI as well. Before we ask “What can AI do?”, we should ask “What do we need?” The answer is rarely an astronaut on a unicorn. Most of the time it’s something more like a travel expense report that takes care of itself.