Recently I decided to rebuild the mobile version of my game "Growing City" (https://rastushiy-gorod.ru). I wanted more than a responsive version of the desktop interface — a full mobile game screen with convenient controls and a proper layout of the main elements. I asked Claude to help with the implementation, but got a fairly confident answer: a task like this can't really be done in Cursor, you need a developer.
"This can't be done in Cursor"
The problem was quite real. The game was originally designed for a large screen: the card market, player actions, and information about cities and participants all fit fine on a computer monitor, but on a phone this whole structure started falling apart.
The simplest path is to shrink everything down to the mobile screen. But that usually ends badly: tiny elements, awkward buttons, and the constant need to scroll.
A different approach was needed: a separate mobile layout with its own logic for arranging elements. A header with the key information, the card market, a player panel, an action block, working with swipes and fixed interface areas. Claude assessed this as full-blown mobile client development and concluded that it couldn't be done without an experienced developer.
Someone else's opinion turned out to be more useful than mine
Instead of wasting time arguing with the model, I opened ChatGPT and asked it to break the task down from an architectural angle: how best to organize the screen, which blocks to make independent, where to use fixed panels, how to handle mobile interaction. The result was a detailed implementation plan.
I simply copied it and sent it to Claude. Without extra explanations or attempts to convince it that the task was actually doable. Claude analyzed the proposed structure, suggested its own corrections, and started implementing. In the end I got exactly that architecture: separate interface blocks, a bottom action panel, gesture controls, and a screen designed specifically for a smartphone rather than just a shrunken computer version. In other words, a task that had literally just looked "too big for Cursor" turned into an ordinary development iteration.
Don't argue with AI — verify its answers
The most interesting takeaway here isn't that Claude was wrong. That's normal: the model doesn't see the whole project the way a developer does, and it evaluates a task based only on the available context. The problem starts when you treat such an answer as a final verdict. AI can misjudge complexity, miss another approach, or assume a task requires far more resources than it actually does.
In situations like these, it's better not to prove the model wrong but simply to get a second opinion and test the hypothesis in practice.
The mobile version of "Growing City" is already live at https://rastushiy-gorod.ru. I'd be glad to get feedback on the interface, the game mechanics, and ideas for attracting the first users.
And I'm curious: have you run into situations where AI first confidently said "this is impossible," and then, after you changed the approach, calmly solved the task? How often does that happen to you?
Top comments (0)