Wanted to test what Pro actually does with the kind of prompt a founder types at 9pm, so I gave it one: build a SaaS that tracks my AI subscriptions and their usage limits, warn me before I blow past them, include tests and seed data. Then two curveballs, then a cold-install check. Every build and test ran outside Cursor so it couldn’t grade its own homework.
What one run produced: a working Express app with a JSON store, 18 passing tests, seeded with real data. It also named the product itself, generated five candidates with reasoning and picked “Ceiling,” which I kept because it’s genuinely better than what I had.
Curveball one: CSV import that skips bad rows and reports which ones failed and why, instead of crashing. It added the importer plus a test feeding it good and bad rows. 20 green.
Curveball two, the one I cared about: add freemium without dark patterns, demo checkout that charges nothing. It printed the free limit on the form up front, enforced the gate server-side on the 4th add, and the checkout says DEMO in three places with no fake card fields. 23 tests, all passing.
Cold install: copied to an empty folder, npm install, 68 packages, 0 vulnerabilities, 23/23. Runs from nothing.
Two questions for the room, because they’re the only part of the test that came up short and maybe I’m missing settings:
Is there anywhere to see a remaining-quota number on Pro? Two full sessions cost me nothing extra, which is generous, but the usage page never showed me a ceiling, and finding it by hitting it is a rough way to budget.
Is there any way to see, after the fact, which model an Auto run actually used? The ledger just says “auto” permanently. I’d take even a per-session log. Related: in an earlier session Composer accepted a calendar date that doesn’t exist and Opus caught it, which is exactly why I want to know who did the work.
Fair disclosure: I run a small reviews site (Okaneland) and this test is written up there with screenshots and the full meter breakdown. I’ll put the link in a reply to keep this post about the build. Also for scope: I drove most of this through cursor-agent CLI and the cloud agent, so Tab and Cmd+K went unexercised. Not claiming anything about them either way.
Happy to share the prompt verbatim if anyone wants to reproduce the run.
