Genre roulette
Each episode starts with a roll. The genre — tower defense, platformer, puzzle, whatever — is decided before anyone touches a model.
About PromptBash
We stress-test the future. One build at a time.
Regular people — a tinkerer, a rookie, a wildcard, and a newcomer — stress-testing whether AI can really build software for anyone. Real prompts. Real code. Real receipts.
Origin
Regular people. One promise from every AI lab on earth. Time to stress-test it.
[Draft — Scott to edit] Every model lab on earth is making the same pitch: their AI can build real software for anyone, no coding experience required. Some of them can, sometimes, on a good day, in the right hands. PromptBash exists to find out what happens when those hands aren't a software engineer's. The crew runs the full beginner-to-power-user spectrum: a tinkerer who wrote his first program on an Apple IIe and has been around computers for nearly forty years (Scott), a complete beginner who's never written a line of code (Mayson), a tech-literate non-coder who's pure unpredictability (Devin), and a true newcomer with even less computer experience than the rest (Ashley, Scott's daughter). If the AI promise is real, every one of us should be shipping working software. If it's not, you'll see exactly where it breaks.
[Draft — Scott to edit] The format is brutally simple. Same genre for everyone. Same prompt budget. Different models, different brains, different builds. Then we put every game side by side and let viewers play them, vote on them, and decide whose AI actually delivered. The results are sometimes a coronation, sometimes a disaster, and sometimes a surprise nobody saw coming. That's the show.
The Format
The rules are simple. The outcomes are not.
Every episode runs on the same set of constraints. The model and the person behind the prompts decide the rest.
Pick your model
Each crew member picks the AI they want to bet on for the episode. Claude Opus, Sonnet, Haiku, GPT-5, Gemini, Grok — anything currently shipping is fair game.
Five prompts. No edits.
Every crew member gets exactly five prompts to ship a complete game. No takebacks. No editing prompts mid-stream. Whatever the AI hands back is what we work with.
Side by side
Every game goes up at the same time. Viewers play them all, vote for their favorite, and the leaderboard tells the truth.
Bragging rights
Winner gets bragging rights until the next episode. Losers explain themselves on camera. Everyone learns which models can actually deliver under pressure.
Episode Formats
Not every episode plays the same.
The five universal rules above apply across the board — but how they get applied depends on the episode format. Here's what we run.
Showdown
Same genre. Same budget. Different brains.
The classic format. Crew members get the same genre and a fixed prompt budget, then each picks their own AI model and builds in parallel. At the end, all the games go up side by side and viewers vote on which one actually delivered. Participant count varies per episode — sometimes it's the whole crew, sometimes a subset.
- Same genre for every participant
- Each picks their own AI model
- Fixed prompt budget per person
- Games go up at the same time for side-by-side play
Round Robin
One codebase. Taking turns. Watch it evolve.
Crew members share ONE codebase and take turns evolving it across multiple rounds. Each round is one prompt applied to whatever the previous person left behind. The game gets weirder, smarter, or funnier depending on who got the last word. Participant count and round count vary per episode — sometimes a tight rotation, sometimes a free-for-all.
- Single shared codebase across the whole episode
- Participants take turns in a defined order
- One prompt per round on whatever the previous person left behind
- Each round is a wildcard model pick
Model Comparison
Same task. Different AIs. Whoever wins, wins.
One or more crew members run the same task through multiple AI models — Claude, GPT, Gemini, Grok, AiFr3d, anything we can get our hands on. The focus is on comparing the AIs themselves, not the people. The contestants are the models.
- Same task run through multiple models
- Models include any shipping AI: Claude, GPT, Gemini, Grok, AiFr3d
- Outputs scored side by side
- Participant count varies — solo run or multi-crew comparison
The People
The people behind the prompts
A tinkerer with nearly forty years around computers, a complete beginner, a tech-literate wildcard, and a true newcomer. Click a card to dig into each one.
Scott
The Tinkerer
Nearly forty years of bossing computers around — wrote his first program on an Apple IIe and never looked back. Not a coder, just stubborn enough to make machines do what he wants. Been playing with AI tools and bots since they hit the mainstream, and now he’s letting the AI take the keyboard.
Opus
Profile
Mayson
The Rookie
Has never written a line of code in his life. Walks into every episode with nothing but curiosity and a prompt budget. If AI can really build for anyone, he’s the proof.
Sonnet
Profile
Devin
The Wildcard
Tech-literate but not a coder. Heavy software user, quick to learn, completely unpredictable when the AI does something weird. Comes in hot with ideas no one asked for.
Haiku
Profile
Ashley
The Newcomer
Scott's daughter and the absolute beginner of the crew. Knows even less about computers than Mayson — which is the whole point. If AI works for her, it works for anyone.
No episodes yet
ProfileThe Roster
The AIs that have shown up
Every episode mixes the roster up. Here's who's been in the ring so far.
On the show
Claude Opus
Anthropic
Anthropic's heavyweight. Slowest to respond, sharpest reasoning, hands down the most expensive prompts on the show. Scott's pick when he wants the AI to actually think before answering.
On the show
Claude Sonnet
Anthropic
The middle child. Fast enough to iterate, smart enough to keep up. Mayson's bet that a balanced model can carry a complete beginner across the finish line.
On the show
Claude Haiku
Anthropic
The speedster. Cheapest by far, ships answers before you finish reading the prompt. Devin's wildcard pick when he wants quantity and surprise over precision.
Questions
The stuff people ask
And the answers, where we have them. Some are still drafts — we'll keep refining as the show evolves.
Behind the Scenes
The site is a meta-build.
Yes, we're aware. A channel about non-programmers stress-testing AI build tools, on a website built almost entirely by AI under Scott's direction. The recursion is the point. Take a peek under the hood.
The Stack
- Next.js 14App Router, server-rendered
- TypeScriptStrict mode
- PostgreSQLContent database via Drizzle ORM
- Tailwind CSSCustom design tokens
- Framer MotionSection reveals and motion
- Radix UIAccessible primitives
- Lucide IconsIcon set
- Linux VPSUbuntu + nginx + PM2
- CloudflareDNS + future tunnel
Built collaboratively by Scott and Claude Opus over the spring of 2026 — a tinkerer with nearly four decades of computer experience (he wrote his first program on an Apple IIe), working alongside an AI that did all the actual code. The clock behind the logo is the viewer's local time. The character art was generated and post-processed; backgrounds were algorithmically removed via corner flood-fill so even Mayson's solid black shirt survived intact. If a self-described non-coder can ship this site, that's the show's whole pitch in one website.
Stay in the Loop
Catch every showdown.
New episodes drop on YouTube. Subscribe and you'll never miss the moment an AI rage-quits halfway through prompt 4.
Subscribe on YouTubeContact (coming soon)