What this case covers
Fish used Codex to complete "Shiguang Portrait Studio": upload a life photo and generate ID photos, business portraits, artistic photos, or meme stickers. The product was built in 3 hours and deployed, with payments starting the next day; at an offline session, 11 participants, most without a programming background, also built runnable products within 3 hours.
Questions it answers
- Why verify whether someone will pay before building a product?
- How do you converge a vague idea into input, processing, and output?
- Which features should be cut from a 3-hour MVP?
- When there is no business license, how should the payment step be handled in stages?
- After AI completes the code, what real information still needs to be checked before launch?
Our observations
- "All 11 people built it" proves the production barrier has dropped, but it cannot be directly equated with every product being able to generate sustained revenue.
- The author has over ten years of development experience; his judgment on scope, testing, and deployment remains an important precondition for the 3-hour result.
- What is most worth transferring is validating demand first, cutting to the smallest closed loop, then using real user feedback to decide whether to continue investing.
From the author
"What exactly to build and how you want it done"—it can't make that decision for you; you need to make the judgment yourself.