Spent this week actually pushing a build live through all three paths instead of just describing them, so here's the log — what I ran, what broke, what I'd skip if I did it again.
Day 1, morning — the one-click subdomain
Started with the fastest option: hit publish, got yourname.buildmidas.com with no DNS, no account creation, no cost. Took maybe four seconds. This is the move when you just want to know if anyone cares about the idea — share the link, watch what happens, iterate. I half expected to hit a wall once the app needed real backend stuff — accounts, a database, a WebSocket multiplayer layer — but no, that's hosted and managed too. Nothing to configure on my end. Good first day.
Day 1, afternoon — trying to break the SFTP deploy
This is the part I was most nervous about. Once a product outgrows the subdomain and needs its own domain, you deploy over SFTP straight from the build page. I pointed it at a server that already had a webroot with old files sitting in it, half expecting a clobber. Instead the deploy agent inspected the server, picked a strategy, and — this is the detail that mattered — captured the pre-existing webroot before touching anything. Every version deployed after that gets kept too, and revert is one click. So when I deliberately shipped a broken build twenty minutes later just to test it, rolling back took about as long as it took to notice the build was broken. I wrote more on why that matters in Iterating without fear — the short version is that version history turns deploys from a held-breath moment into a non-event.
What I'd skip next time: I spent twenty minutes trying to trick it with a weird nested subdirectory structure before remembering the whole point is that it's your server, your domain, your files — the tool is just a careful mover, not a gatekeeper. Wasted effort testing something that was never the risk.
Day 2 — the store path, which I didn't finish
Didn't actually complete this one, and that's honestly the more useful note. Store shipping — Android to Google Play, extensions to Chrome Web Store and Firefox Add-ons — runs through your own developer accounts, with listings and privacy declarations prepared by agents. That's real, but it's also its own multi-day process with review queues outside anyone's control, so I stopped after the listing prep step. If you're chasing store distribution, budget for that separately; I wrote the full path in From prompt to app store.
What decided which path, in the moment
By day two the choice mostly made itself:
| What I was doing | Path I reached for |
|---|---|
| Testing if the idea had a pulse | Subdomain, same day |
| Needed a real brand domain | Subdomain first, then SFTP to my own server |
| Utility people should find in a store | Subdomain landing page + store ship for the app |
| Client project on their infra | SFTP to their server, versioned |
End of week
By Friday I had a marketing site on my own domain, the product itself live on a subdomain, and a store listing half-prepped — all three paths running at once for the same project, built and iterated from the same chat the whole time. That's apparently the normal end state for anything serious, not an edge case. Wish I'd started expecting that on day one instead of treating the three paths as separate decisions.
