We'll make it configurable. We sell configuration as flexibility, and flexibility is great! Customers love flexibility; it means developers make fewer future changes, and customers can customise the product to suit their needs.
But configurability can also mean that nobody made the decision about how the product should work.
Configuration can be great, if used deliberately. Consider an alert system. You want to inform customers when something that is important to them changes.
Immediately, questions appear. Who should know? What should they be told? When should they be told?
For the most serious alerts, the answer might be simple: everyone, immediately, with a clear message.
Move towards the informational end of the scale and things get harder. Alerts can arrive at different times, mean different things to different people, overlap with other alerts and quickly become noise. Too many create fatigue. Too many options make the system harder to understand and much harder to support.
If you solve that by making everything configurable, you've made the customer responsible for the design.
Good products have boundaries; they push back when you try to do certain things. These boundaries aren't indecision masked as a configuration panel. They're opinion.
If your opinion is wrong, 30 people will tell you, not 1. Your customers interacting with the product at the boundary means something; you may need to change your opinion, but it's your decision.
That boundary comes with less to build, less logging, less debugging, a more coherent workflow and an understandable product.
AI has changed the ROI of building software. Historically, building configuration came with engineering costs that forced discussion and decision.
Do we really need this?
Now the implementation might cost a few hundred pounds in tokens. The question of whether you needed something came from the friction it created. AI only removed that; the system's complexity is still there.
You haven't necessarily improved the product. You may simply have made the customer do more work before they can use it.
AI can reduce the cost of implementation while increasing the cost of adoption.
When everything becomes easier to build, judgement needs to become sharper.
Building a product means deciding how to solve a customer's problem.
It doesn't mean figuring out how many questions you can ask them.