What AI still does not decide for you
The question comes up in nearly every conversation: "with AI, is it still worth having something built?" The honest answer is neither "no" nor "obviously". AI has moved the cost of the work; it has not removed the work.
Writing was never the hard part
In a project, typing code is a small share of the time. The rest is understanding how people actually work, spotting the exceptions nobody documented, choosing what to keep and what to drop — then living with those choices for years.
A model produces a plausible answer to the question you asked. It does not tell you the question was wrong.
Three decisions that stay human
- Which problem to tackle — the most expensive one is rarely the first one stated
- What not to build — a good tool is mostly a tool that refused half the ideas
- Who answers when it breaks — someone has to pick up the phone when an invoice goes out twice on a Friday evening
The "it works" trap
A working demo proves very little. The real test comes six months later: the data has drifted, someone has left the team, an unforeseen case has shown up. That is when you find out whether anyone thought about the edge cases, or simply accepted the first suggestion that came along.
What genuinely changed
What AI changes is the entry cost: testing an idea is cheaper than it used to be. You can prototype a direction, throw it away and try another one in the same week. That is very good news — provided someone is still there to make the call.
In short
The tools changed, the job did not. Value moved from "knowing how to write" to "knowing what to decide": understanding a process, setting a scope, owning the maintenance. That is exactly the part nobody wants to hand over to a machine.