▶ YouTube Summaries

← All summaries

Casey Muratori judges vibe-coded digging games

2026-09-27 Sun ⏱ 54 min theprimeagen

A TheStandup segment where Prime, Teej and Trash show Casey Muratori the digging games they built with AI over two weeks, and he judges them. The comedy is that almost nothing works on stage, but the interesting part is what Casey's verdict reveals about the gap between a demo and a game.

Prime rigs the contest before it starts

Prime opens by asking Casey to judge not on completeness but on how fun the games could be - powerups added, effort applied. Teej immediately calls it leading the witness; Prime declares himself judge, jury and executioner and overrules the objection. This matters, because Casey's eventual ruling hinges on potential, and everyone accuses Prime of having poisoned the well from the first minute.

Trash: two weeks of pivots, no digging game

Trash spent two weeks building a mini-games site rather than the digging game he had asked everyone else to make. He walks through the evolution: a bucket-of-water sandbox (FPS counter attached to a single bucket; bucket capacity rendered as 241.2 ml, leftover debug code), a chopstick kitchen-prep game, an excavator, a dung-beetle multiplayer race where the beetles walk backwards even without poop, a tow-truck game, a fly-catching Karate Kid homage, a goose game.

The pattern Casey and Teej push back on: Trash abandons each concept before finding out whether a game is in there. His own criterion was how long he could imagine playing it, which is not something you can know from a prototype. The multiplayer never shipped because PartyKit was "lag city". The site eventually crashes Casey's machine under the weight of Three.js - which Casey declines to blame on AI, since 3D web pages have been spinning up fans for a decade and the training data is just full of bad sites.

Panicking the night before, Trash pivoted to an actual digging game: dig.qwerop.com, a first-person digging simulator with QWOP-style controls (Q/W raise and lower, E/R angle, O push in, P pull out), depth milestones saved to local storage, fossils and worms further down. Then a baseball pitching game on the same control scheme - arm, torso, legs - with a live leaderboard, max score 602, built the previous day. Prime is instantly addicted and Casey notes aloud that Trash just cost him a week of Prime's productivity. Prime's only feedback: reset takes a second, make it instant on space.

Prime: a DDR digging game that does not work

Prime built his in under two hours that morning on Opus 5.5, explicitly to test it, off an idea he had the moment Trash said "game": Trash digging himself into a huge hole. Arrows scroll, accuracy of your steps determines dig speed, fall off screen and you die.

On stage: the accuracy meter does nothing, holding the down arrow just works, nobody dies, the arrows are not on the beat, and hitting left/right moves you sideways instead of digging - which, as Casey points out, does not even make sense as a concept. The art is unparseable visual noise; Prime claims it cost over $500. Casey presses him on why his art is always incomprehensible while Teej's and Trash's look fine, and gets no answer. The one thing that does work is the networking: real time-differential sync, everyone hearing the same song at the same instant.

Teej: Gravel Empire, RollerCoaster Tycoon for digging

Teej started at 10am, three hours before the show. Hire workers, build footpaths, mark dig sites, add cranes and dump trucks that automate the pit as it deepens. On stage Trash cannot get anyone to dig - no paths to the site, no shovels issued, workers possibly off-shift - and the UI font is unreadable at stream resolution. A loaded late-game scenario finally shows the intended shape, with trucks driving through each other. Casey likes the art style unreservedly.

Teej also made a trash-parody narrative game that wins the room outright - Casey wants to show his wife, compares it to Pony Island, and gives it the narrative award.

Casey's verdict

Nothing presented is a game. The only thing that resembles one is Trash's QWOP dig, and only because the genre's whole point is that it does not work - QWOP is what you ship when you hook springs to the legs and stop.

On potential, Casey gives it to Prime, while saying it looked absolutely horrible. If the DDR steps actually registered and you moved the arrows under the character instead of off to the left - game design 101 - that is the one he would play. He recuses himself from the baseball game since he dislikes QWOP games, and hands that judgement to Prime after seventeen hours of forced play. Gravel Empire he cannot rate: sim games live or die on whether organising the thing feels good, and there is not enough there to tell.

Prime's defence is the best line of the segment: judging vibe-coded games on potential is entirely in the spirit of the AI hype cycle, where nobody talks about what exists, only about what we might have. He then argues the real lesson - off-the-shelf assets, a close-up on the digger, cuts on the beat, and his broken game would have made an irresistible demo video. Which, he notes, is exactly what demo videos are right now.