I made Claude and Cursor build the same app. Then I put them on trial.

Cursor built my app, then told me I'd used up my free plan. Claude built the same app in four and a half minutes, then told me it had tested it on a phone. One of them was telling the truth.

This week is MLH Global Hack Week, and one of the challenges was "Build a Mini App with Cursor." So I built Excuse Court: you type your excuse, a cartoon judge bangs a gavel, and you get a verdict, a sentence and a believability score out of 100.

Then I asked Claude Code to build it from the exact same prompt. Head to head.

The race

Yes, Claude overestimated its own time. That is the most humble thing an AI has ever done to me.

Meet the judges

Cursor gave me The Honorable June Burden. "She has heard worse. Not often." She is scowling before you've typed a word, and she never stops.

Claude gave me Judge Dubious, in a powdered wig. His eyebrows go up when you're acquitted and crash down when you're guilty. He also gives you five excuses to tap if you can't think of one, which is the most useful legal advice I've ever received.

Same excuse, two judges

I typed the same four excuses into both courts. These are the real verdicts and scores.

Honestly? Cursor's lines are longer and weirder, and some are better. "Sentence: leave fifteen minutes earlier, which is not cruel, it is a clock" is a perfect sentence. Claude's are shorter and punchier, and Claude explains your score: say "I swear" and you lose 12 points, because "nobody innocent says that."

The raccoon incident

Here's the part I didn't expect.

I told both judges a raccoon stole my car keys. Cursor's judge: "You lost an argument to a raccoon and I have no follow-up questions. That is precedent."

Claude's judge issued a warrant for the animal and then said: "The court has met a goose and fears it too."

It was a raccoon. Claude wrote one line for every animal and named the goose in it, so the raccoon gets accused of being a goose.

It gets better. Claude's judge acquitted me with a believability score of 15 out of 100. Not guilty, but also, nobody believes you. And on a phone-sized screen, the big green ACQUITTED stamp runs straight off the edge of the card.

Remember how Claude said it had tested the app at phone width "with no errors"? Technically, a stamp falling off the page is not an error.

Also, Claude gave the dog excuse 1 out of 100 and docked me 10 points for blaming "3 different things at once." The three things: the dog, the homework, and the word "ate," which Claude files under snacks.

The scorecard

These counts come straight from the two files, so you can check them.

The verdict

When Claude finished, I looked at both and told it "you win." Mostly because it finished fast, the judge's face moves, and it still had usage left for round two.

But if this were a real court, Cursor would win on jokes. Claude would win on speed, and then get charged with impersonating a goose.

Since then

The pictures above are the two builds from race day. I kept working on the game after the trial, and it has been redrawn since. Here's Excuse Court today:

Play them yourself:

• Claude's race-day version: Excuse Court on Claude
• Cursor's version, all grown up: connabyss.github.io/excuse-court (code: github.com/connabyss/excuse-court)

Now I need a ruling from you: what's the worst excuse you've ever actually used? Put it in the comments and I'll run it through both courts.

I used Claude Code to help write this post and make the pictures. Every verdict and number above came from running the two real apps.