An OpenAI engineer wanted to share his album on X. The obvious path — upload the MP3, hope the audio player renders, hope the file size clears whatever cap Twitter enforces this month — was, in his estimation, too much friction. So Andrew Ambrosino (@ajambrosino, who works on Codex and Chat at OpenAI) skipped it entirely and prompted a full hosted music-player web app into existence on ChatGPT Sites instead — complete with a library view, album artwork, a 10-track playlist, and playback controls that mimic Spotify's own UI.
He named it, with the kind of self-awareness only an engineer at the company that built the tool could get away with, "Now That's What I Call Slop! Vol. 1."
Codex lead Tibo Sottiaux (@thsottiaux) surfaced the story on X on August 22, 2026, to 46.9K views: Ambrosino "thought it was too hard to upload music to twitter, so he just prompted a whole music app hosted on ChatGPT Sites into existence just to share his album." Ambrosino's own reply credited a collaborator: "credit @dkundel."
TL;DR
| Question | Answer |
|---|---|
| What happened? | Andrew Ambrosino prompted a hosted music-player app on ChatGPT Sites to share his album, instead of uploading audio to X. |
| What's the app called? | "Now That's What I Call Slop! Vol. 1" — a 10-track album, credited alongside collaborator @dkundel. |
| Who surfaced it? | Tibo Sottiaux (@thsottiaux), Codex lead at OpenAI, in an August 22, 2026 post (46.9K views). |
| Who can publish a Site? | Per Sottiaux: ChatGPT Work or the ChatGPT desktop app, Plus plan or above. |
| Why is this the real story? | Building the exact tool you need now costs less effort than working around an existing platform's limits. |
| Community read? | Genuine praise for the music, not just novelty interest — see james.'s reaction below. |
The actual story isn't the joke name — it's the choice
It's easy to read this as a one-off comedy bit: an OpenAI employee stands up a Spotify clone just to be silly about sharing a demo album. That undersells what happened.
X has native audio upload. It's clunky, but it exists. Ambrosino looked at that path — fight format constraints, hope playback renders cleanly in a feed card, accept whatever length or size limit applies — and decided it was more friction than opening ChatGPT and describing a music-player app. He didn't reach for a workaround inside an existing platform's constraints. He reached for a purpose-built tool, because building one no longer costs more than working around the alternative.
That's the shift explainx.ai has tracked through 2026's vibe-coding coverage: the bar for "just build a small app for this one specific need" has dropped low enough that it now routinely wins against "find a way to make an existing tool do something it wasn't designed for." A hosted, shareable web app — library page, album view, tracklist, playback widget — took roughly the effort of writing a prompt. Fighting Twitter's audio upload took... also roughly the effort of writing a prompt, except the output would have been worse.
explainx.ai covered ChatGPT Sites' team co-editing feature the day before this story broke — OpenAI's August 20, 2026 update letting workspace teammates jointly build and publish a Site. That post framed Sites as internal-tool infrastructure: ops dashboards, request trackers, launch calendars. Ambrosino's album is the same feature, aimed at a completely different, much more mundane problem — and that's exactly what makes it a better illustration of the pattern than a staged demo would be. Nobody asked him to build this. He built it because it was the path of least resistance for a Tuesday-afternoon "how do I share this" problem.
Who can actually do this
Sottiaux followed up with the concrete access detail, in his own words:
"You can make your own sites (and share them!) in ChatGPT Work or on the ChatGPT desktop app if you are on the Plus plan or above."
That lines up with what OpenAI's own Sites documentation says about the feature being in public beta across Plus, Pro, Business, Enterprise, and Edu plans — but Sottiaux's reply is the more actionable version for anyone reading this and wondering whether they could do the same thing this afternoon. The floor is Plus, and the creation surfaces are ChatGPT Work or the desktop app. No separate developer account, no CI pipeline, no hosting bill to set up before your first deploy.
The tracklist, and what people actually said about it
The album runs 10 tracks, roughly 36 minutes, hosted as a normal album-detail page with per-track durations and a currently-playing widget — the interface reads like a streaming platform because it was prompted to. One track, titled "All the Checks Are Green" — a wink at CI pipelines passing — got a specific shoutout from Sottiaux: "Favorite is [track link]." In line with how explainx.ai treats third-party creative work, this post describes the album's shape and title without transcribing or closely describing any of its actual lyrics.
The more interesting signal came from the replies. Developer james. (@jbk_io) wrote: "This has no business slapping this hard." That's not a comment about the app-building trick — it's praise for the music itself, independent of the fact that the hosting layer was vibe-coded in an afternoon. It matters because the easy dunk on any "AI built this" story is to assume the output is a novelty at best. Here, the reaction split cleanly: people were impressed by the workflow and the output held up on its own terms.
Kien Pham (@kienbuilds) added a lighter aside that ties this story to another one explainx.ai covered the same week: "I HAVE TO give this album a listen while my Luna swarm agents hack away at the banked reset!" — a reference to Codex crossing 20 million active users and OpenAI crediting every user a banked usage reset that Tibo Sottiaux announced days earlier. The two stories aren't mechanically connected, but the reply thread is a snapshot of the same community reacting to both in the same breath — Codex's growth milestone and someone's afternoon side project, discussed with equal enthusiasm.
Why this belongs next to explainx.ai's vibe-coding coverage
explainx.ai has tracked the "prompt a whole app into existence" pattern through several 2026 stories — from the JSON vibe-coding meme explainer to Meta's own internally vibe-coded Gizmos app. Those posts covered the pattern as a labor-and-tooling story: engineers building throwaway internal tools faster than they could requisition them through normal channels.
This story is the same pattern with the comparison point flipped. It's not "build a tool instead of asking IT." It's "build a tool instead of using the platform-native workaround everyone already accepts as good enough." X's audio upload is a known, tolerated inconvenience — millions of people deal with it every day by just not sharing music natively. Ambrosino's choice says something sharper than "AI makes building easier": it says the comparison itself has changed. The workaround used to be free (just upload the file, deal with the limits) and building was expensive (stand up hosting, write a player, deploy it). Now building is close enough to free that the workaround's remaining friction — however small — loses.
What this means for anyone shipping something small this week
- Stop reaching for the nearest platform hack first. If sharing something specific — an album, a dataset, a small tool — requires fighting a platform's constraints, check whether prompting a purpose-built Site is actually faster. Often it is.
- Plus is the floor, not Enterprise. You don't need a workspace or admin approval to try this solo — Sottiaux's reply confirms Plus-tier desktop app access, separate from the workspace-only team co-editing feature OpenAI shipped two days earlier.
- Output quality is a separate axis from build method. james.'s reaction is a useful gut check: judge the thing that got shipped, not just the fact that it was vibe-coded. "Fast to build" and "good" are no longer mutually exclusive defaults.
- Small, disposable apps are becoming the default unit of sharing, not the exception. A hosted Site for one album is the same instinct as a hosted Site for one internal dashboard — the size of the problem no longer has to justify standing up real infrastructure.
Related reading
- ChatGPT Sites team co-editing — build and publish together
- OpenAI Codex Sites launch — role-specific plugins
- Codex hits 20 million users — the banked reset story
- What is vibe coding, explained
- Vibecoding and the "who is JSON" meme
- Meta's internal Gizmos app, vibe-coded
- Spotify's AI artist labeling policy
- Official docs: ChatGPT Sites developer guide
Details reflect public X posts and OpenAI's Sites documentation as of August 22, 2026. Feature access and plan tiers may change; confirm current availability in your ChatGPT account before publishing your own Site.