Mario Nawfal's crypto and AI account shares a prompt for having Claude Opus 5.5 send five test users through an app before launch
Before you launch anything, have Opus 5.5 send 5 subagents in as your first users. Paste this into Claude Code: "Act as my launch QA lead. Spin up 5 subagents, each a different first-time user: 1. The impatient one: skims, clicks the first button, quits at the first error 2. The skeptic: reads pricing, terms and privacy before signing up 3. The mobile user: does everything on a 375px screen on slow 3G 4. The non-native English speaker: only follows what the UI literally says 5. The breaker: pastes junk into forms, hits back mid-flow, double-clicks everything Each one opens the app in a real browser, goes from landing page to first success, and screenshots every moment they get confused, stuck or annoyed. Merge everything into one list ranked by how many users hit it. Fix it all in one pass, then rerun the 5 users to prove it's fixed." Your first 100 real users won't tell you why they left. these 5 will. (Works best with a browser MCP connected, Playwright or Chrome DevTools)
Context
Using agents as stand in users is a cheap way to catch problems early, though their reactions are not the same as real customers'.
This is a tips post from an account that describes itself as Mario Nawfal's crypto and AI account. The prompt has not been tested here, and the account gives no results from using it.
Deep dive
On October 18, the account RoundtableSpace posted a prompt for Claude Code, saying that before you launch anything you should have Opus 5.5 send five subagents in as your first users.
The prompt asks Claude to act as a launch QA lead and create five subagents, each playing a different first time user. One is impatient and quits at the first error. One is a skeptic who reads pricing, terms and privacy pages before signing up. One uses a 375 pixel screen on a slow 3G connection. One is a non native English speaker who follows only what the interface literally says. The last tries to break the app by pasting junk into forms, going back mid flow and double clicking everything.
Each subagent is told to open the app in a real browser, go from the landing page to a first success and take screenshots whenever it gets confused or stuck. The prompt then asks Claude to merge the problems into one list ranked by how many users hit them, fix them all in one pass and rerun the five users to show the fixes worked.
The post says it works best with a browser tool connected, naming Playwright or Chrome DevTools. It argues that your first 100 real users won't tell you why they left, but these five will. AI testers can miss problems that real people hit, and fixing everything in one pass can introduce new bugs, so results still need checking.
What to expect next
Watch whether developers report that agent testers like these catch real problems before launch, compared with standard automated tests and human testing.
Also watch for coding tools that build this kind of persona testing in directly. Agents that drive real browsers are becoming common, which makes simulated first users easier to set up, though their feedback still needs checking against real people.