AI agents are really, really good at helping formulate business requirements. Business speaks business. Engineers speak engineer. AI’s understand and speak both. Excellent.
And then it gets creepy. An AI agent’s deeply coded need to be helpful means it rarely stops at the core business need.
Ask for a feature and the agent offers, “You probably also need X.” Then, “If you have X, you really should have Y.” And inevitably, “Most successful mature platforms also support Z.”
No. No I would not.
To the business, this looks amazing. The AI is thinking ahead, finding gaps and making the requirements more complete. To engineering, it looks like the opening scene of a horror movie.
We call it scope creep.
Scope creep is one of the most insidious ways engineering projects lose focus, never quite finish, and blow through both time and budget.
Every extra requirement brings more code, more interactions, more edge cases, more tests, more failure modes and more things to implement. AI coding agents can certainly make coding faster, but they do not make scope free. Whether a human or an AI is building it, more scope still requires more work and more time.
This is where the human in the loop matters. Someone needs to look at the wonderfully comprehensive specification and ask the questions the AI rarely asks itself: Do we actually need this? Do we need it now? What is the smallest complete solution to the actual business problem?
Use AI to brainstorm. Use it to find missing cases. Use it to turn fuzzy business ideas into clear requirements. It is genuinely excellent at all of that.
But somewhere around the third, forth or fifth “you should also…”, get suspicious.
Cut it back to the core requirement. Define that properly. Build it. Ship it.
You can always add features X, Y and Z later.
If anyone even needs it.