I spent five days running the free plan the way an actual user would, not the way a pricing page describes it, and wrote down what happened as I went. Screenshot-and-move-on isn't how you find the fine print; building something for a week and hitting whatever wall is actually there, is. Here's the log.
Monday, 9:40am — signed up, no card
First thing I checked before writing a line of code: does it ask for a card. It doesn't. I'd normally skip a detail like that, but a free tier that wants payment details before you've built anything is telling you what it thinks of you, so I logged it and moved on to actually building.
Picked a small-business site as the test project — nothing exotic, a landing page, a contact form, a blog section. By 10:15 I had a first pass live on the free subdomain. No paywall between "build" and "publish," which is where a lot of free tiers quietly turn into demos.
Monday, 2:00pm — the iteration loop, where I expected the wall
This is the part I was most suspicious of. First builds are cheap for any agent — clean brief, no existing code to reconcile — so a platform can make free look generous on build one and then bleed you on build two. I went after the loop: "make the header sticky," "fix the mobile nav," a content pass, two smaller tweaks after that. Each of those touched more files than the ask implied, the way real revisions always do. I didn't hit the ceiling that day, but I could feel where it was — the allowance is clearly sized around one full cycle like this, not five.
Also checked, mid-afternoon, whether the review pass gets skipped on free accounts to save compute. It doesn't — same linting, same security check, same functional verification before anything's marked done. I went looking for a shortcut here specifically because it's the first thing I'd cut under margin pressure if I ran this business, and it wasn't cut.
Tuesday, 11:00am — pushed a second project to see the backend limit
Spun up a second idea, this one needing a real backend — database, an API route, the works. This is where I expected the actual gate, and I found it: one live server-backed app per free account. Tried to bring up a second hosted backend alongside the first and it wouldn't run both live at once.
What surprised me was what doesn't count against that slot. I built a small downloadable tool later in the week — no server, just a packaged app — and it published fine, no competition with the live backend. Makes sense once you think about the cost shape: a hosted backend stays warm and costs money whether anyone's hitting it; a static or installable build is just files sitting there. Only the warm one is rationed.
Tuesday, 4:30pm — tried to break concurrency
Started a build, then mid-run asked the assistant to kick off something unrelated. It queued instead of running in parallel. Annoying for about ninety seconds, then irrelevant — I was watching the first build's preview anyway, so waiting for it "to finish before starting the next thing" was what I'd have done regardless. I could see this mattering more for a team running two agent pipelines at once — content generation on one page while a build agent restructures another — but that's a different usage pattern than mine.
Wednesday — the libraries, and why they're not a trial feature
Spent Wednesday testing whether the accumulation stuff — knowledge base, reference images, templates — actually works on free or is a paid-only teaser. It's not gated. By day three the platform was reusing my brand voice and reference screenshots without me re-explaining them, which is the entire point of a knowledge layer: locking it away would mean free users never get to see the product compound, and then why would they ever pay to keep using something they never got value from.
Also poked at whether tenant data isolation is somehow "lighter" on free — partial isolation, slower queue for isolation checks, anything. It isn't, and honestly it structurally can't be; isolation is binary. Either background jobs and queries are walled off per tenant or they aren't, there's no 80% version to sell cheaper. Free and paid get the same guarantee here, which I'd assumed but wanted to actually verify rather than take on faith.
Thursday — store publishing, the thing I almost didn't try
I nearly skipped this test because I assumed shipping to an app store would be paid-only — that's the instinct most people have, and it's wrong. Packaged the small tool from Tuesday as an installable and the platform-side publishing worked identically to what I'd expect on a paid account. The only friction was Google's own developer registration fee, which has nothing to do with the plan I'm on. Worth saying plainly because I've watched people self-impose limits that don't exist just because they assumed the worst.
Friday — what I'd skip next time
If I ran this test again I'd skip Tuesday afternoon's concurrency probe — it answered its question in under two minutes and I spent forty more trying variations that all confirmed the same thing. I'd also front-load the tenant isolation check to day one instead of day three, since it's the fact people should know before they put anything real on the platform, not after.
What I wouldn't skip: the iteration-loop test on Monday. That's the one place a free tier usually reveals itself as a funnel, and it's the one worth your own time re-running if you're deciding whether this is a real product or a trial with better marketing.
| What I tested | What actually happened |
|---|---|
| Sign-up | No card required |
| Iteration loop (5+ rounds of revision) | Allowance held for one full cycle; would tighten with heavier use |
| Review/verification pass | Identical to paid — not skipped |
| Second live backend | Blocked — one server-backed app at a time |
| Static/installable build | Doesn't compete for the backend slot |
| Parallel builds | Queues rather than running concurrently |
| Libraries (knowledge/refs/templates) | Fully available, compounds over the week |
| Tenant isolation | Same guarantee as paid, no partial version exists |
| Store publishing | Not gated; only friction is the store's own fee |
So: three legitimate reasons to upgrade off this, based on what actually bit me versus what I expected to bite me. You've run the iteration loop dry more than once, not just the one time everyone does in their first month — see how run-out actually behaves, nothing is lost, but a repeat pattern is signal. You need two hosted backends live at once, which is a structural need free wasn't built to carry, not a budgeting problem you can work around. Or the thing you built stopped being a test and started making you money, at which point running it near a ceiling is an anxiety cost, not just a technical one.
None of those hit me this week. I built a real site and a real downloadable tool, iterated on both, published both, and the only wall I found was exactly the one the fine print said would be there.
