Should AI products let users control how much freedom the AI has?

While working on an AI product recently, I ran into a product-design question that turned out to be much more important than I expected: how much freedom should the AI have when the user gives it constraints?

I initially assumed that better AI meant giving the model more freedom to find the best answer. But in practical workflows, that isn’t always what users want, and sometimes the constraint itself is the point.

I’ve found that making this explicit can be more useful than trying to guess the user’s preferred level of autonomy every time.

And I think the same problem appears in many AI products: coding assistants, travel planning, writing tools, research, design, and more.

Do you think AI products should expose this kind of control directly, or should the model infer the right level of autonomy from context?

I’d be especially interested to hear how other makers are handling this in their products.