As developers, we're constantly hunting for ways to speed up our workflow and rapidly bring ideas to life. I decided to put Cursor—one of the many AI coding assistants—to the test by building a small game with TypeScript and Phaser in under an hour.
The result? Fast, occasionally maddening, but ultimately pretty remarkable.
What I Built in Just 60 Minutes
Armed with nothing but a few prompts, I managed to create a fully interactive game prototype including:
A dynamic world map with varied terrain types that actually affected gameplay A smoothly controllable player character A follower character that mirrored player movements Custom sprite designs rather than default placeholder graphics Character-swapping mechanics with directional logic (sprites faced their movement direction)
This wasn't just a basic sprite bouncing around—it was a legitimate game foundation built in under an hour. For quickly testing concepts and visualizing game mechanics, that kind of speed is genuinely impressive.
The AI Colleague That Means Well
Working with Cursor sometimes felt like pairing with an enthusiastic but overconfident inexperienced dev. It would confidently tell me it had handled a task, but the implementation was often bare-bones—or flat-out wrong.
Several pain points emerged consistently:
Missing imports regularly derailed the app at launch Uncaught exceptions were everywhere—Cursor seemingly had no concept of defensive programming When bugs appeared, Cursor would either miss them entirely or fix one issue while introducing another Even after manually fixing problems (especially null checks), Cursor would sometimes revert my fixes in subsequent edits
Interestingly, Cursor sometimes performed better with looser prompts than detailed ones. When I got too specific, it seemed to lose track of the bigger picture. But when given room to "think," it occasionally delivered impressive results.
Why This Happens
To understand these behaviors, consider how these tools are trained. Cursor, like many coding AI tools, is built on language models trained on millions of public code snippets, documentation, and developer forums. While this enables it to generate remarkably relevant code, it doesn't "understand" architecture, edge cases, or logical flows the way human developers do.
A Cornell study found that AI-generated code can appear competent even when containing subtle errors. Developers often spend as much time verifying AI output as they would writing code themselves—but the initial output appears lightning-fast, which is where the real value lies.
The Real Value: Acceleration, Not Autopilot
Despite these rough edges, my biggest takeaway is clear: AI dramatically accelerates development. I wasn't writing production-ready code—but I was creating and iterating on functional game mechanics at a pace I couldn't match solo.
Cursor isn't replacing senior developers anytime soon—but as a creative accelerant, it's legitimately useful. It gets you 60-70% of the way there in record time. That remaining 30% still needs your experience, polish, and judgment—but you're starting miles ahead of where you'd begin on your own.
Final Result
The final output is a really simple game with basic mechanics. You have two characters, one that leads and one that follows. You can control them with the keyboard arrow keys. Hitting the spacebar switches which character you are controlling. The trees, rocks, and fence in the scene collide with the character but the lakes and rivers you can swim through at a slower speed. It seems to only work with a keyboard so it’s difficult to use on mobile unless you have an external keyboard set up. Perhaps I’ll ask cursor next about swipe controls.
Give it a try and leave your thoughts in the comments!
https://jacoblandry.com/princessanddonut.html
Looking Ahead
As we continue exploring what AI can do for development, tools like Cursor offer a glimpse of the future: not just souped-up autocomplete, but collaborative assistants that handle the repetitive parts while we focus on the challenging, interesting aspects of development.
There's plenty of room for improvement. But with realistic expectations and a healthy dose of skepticism, AI coding assistants can be a powerful addition to any developer's toolkit—especially when you need to build something cool, fast.
