Getting Technical Limits Acknowledged

AI rarely says “can’t”

Something that’s surprisingly hard when working with AI: getting it to say “this is structurally impossible with this tool.”

AI leans, by default, toward finding a way to make it work. No matter what you ask, it proposes some method. Even if that method doesn’t really work, or works but becomes enormously complicated, or doesn’t get past a fundamental limit, AI will at least try.

So reaching the conclusion “what you want can’t be built this way” takes a few laps. It’s hard to get “this isn’t possible in principle” on the first try.

The laps I actually went through today

When collapsing and expanding a sort group in the list, I wanted a feel of “the whole list moving smoothly as one mass.” Not individual items moving separately, but the list sliding up as a single plane.

AI tried several approaches. Slow down the animation speed. Adjust the timing of each item’s appearance. Try a different component. Every time it said, “This should do it.” But when I looked at the result, every version fell short.

After four or five rounds, AI finally said:

Given the structure of the list component we’re using, each item is set up to animate independently. To make “the whole list move as one mass,” we’d have to swap the list component itself for something different. And other features may break in the process.

That was the answer I got after four attempts. Could I have gotten it from the start? I wanted to. But AI wasn’t leaning toward acknowledging limits first.

How to get limits acknowledged faster

So lately, when I talk with AI, I deliberately use phrasing that asks about limits.

• “If we push this to the end, where’s the wall?”
• “What’s a case where this approach breaks?”
• “What’s the fundamental constraint of this direction?”

When I throw these questions first, AI shifts out of “find a way to make it work” mode into “map where the limits are first” mode. That’s how you reduce attempts and quickly get real unworkable territory acknowledged.

And when you get “it’s fundamentally impossible,” it’s important to take that answer at face value. If you can’t, you fall back into the “want to try a workaround?” cycle. Some limits are just limits. Accept them, and a different road opens up.

Acknowledging limits unlocks decisions

When you receive “this can’t be done,” at first it feels like a dead end. But with time, that answer actually opens a path.

After getting “moving the list as one mass isn’t possible in this structure,” I made a decision. “Then let’s keep this structure. Give up the one-mass feel, and instead strip the animation entirely for a cleaner look.” This decision couldn’t surface until the limit was acknowledged. Before that, the fantasy of “just a little more polishing will do it” kept lingering.

Acknowledging a limit is an honest act of allocating resources. Only when what’s possible and what isn’t becomes clear do you actually concentrate on the possible side.

AI’s confidence and the builder’s realism

There’s a sense you have to cultivate when working with AI: the sense to not take AI’s confidence at face value.

AI defaults to saying everything is doable. That’s not quite a lie. In theory, most things are possible. But there’s a big distance between “theoretically possible” and “reasonably possible within this product’s structure.”

Reading that distance is ultimately a human job. When AI says “we can do it,” asking “how complicated will it get? Will something else break? Does it match this app’s principles?” is the product builder’s part.

This is also why working with AI keeps getting more fun. AI throws out possibilities, and the human holds the real constraints. As those two keep talking, what really needs to be built becomes clearer and clearer.

The path isn’t only made of what you can do. What you can’t do is also a path. Deciding what not to do is design.