Most of it comes back to one question: what can these models actually make when you hand them a real brief? Benchmarks, games, small experiments and a couple of shows, all live.
Benchmark
I give the same prompt to different AI models and coding harnesses, then publish what each one actually built, side by side, so you can open the results and judge for yourself.
Static site, GitHub Actions, Cloudflare Pages
model-review-3yp.pages.dev ↗
Games
A shelf of small browser games, each one made by an AI model from a single prompt. 109 of them so far, and you can play every one.
HTML5 games, JavaScript, Cloudflare Pages
ai-game-studio-amz.pages.dev ↗
Experiments
Small, measured experiments around one question: when does a typed model answer actually beat plain code? Each comes with a playground and its results.
Recorded model runs, browser playgrounds, Cloudflare Pages
lab.jev-labs.pages.dev ↗Every one of these was made by an AI model from a single prompt. A few worth starting with:
A small bright 3D platformer: three floating levels of hops, lifts and sliding platforms, 17 gems and a flag a
Run a tiny pixel office: drag client requests onto the kanban board, keep six worker NPCs caffeinated and unst
A cozy patience journey: light sixty lamps along the Kindleway by winning solver-proven deals of Klondike, Fre
Top-down arcade circuit racing: drift for mini-turbos, beat seven rivals, and watch the shift gates reroute th
Build truss bridges from timber, steel, cables and hydraulics on a budget, then watch convoys, tall ships, mou
A word garden: build crossword words from a letter rack to bloom every seed, but your tiles wilt unless an acr
Line up the impossible.
Weave the night's songs.
A drama series set in 1997, written and produced with AI tools. Season one is coming out one episode at a time.
Model showdowns, coding timelapses and plain-spoken explainers.
Shorts starring an AI character with strong opinions about smart fridges, captchas and robot vacuums.
I like building with new AI models the week they come out, and putting the results somewhere anyone can try them. Everything here is made in public and keeps changing. The quickest way to reach me is on X.
