Prototype it. Then prove it.
Hand a PRD to Lovable, v0, Bolt, Replit, Cursor or Claude as a build prompt you can edit. Keep the prototype's link on the PRD, let Claude try it against the acceptance criteria, then put it in front of customers and see what held up.
Prototypes
Prototypes built from this spec.
lapsed-reengage.lovable.app
Lovable · 4d ago · maya@creator-platform.com
figma.com/proto/reminders-v2
Other · 9d ago · design@creator-platform.com
Customer test · 3 sessions
Met in 2 sessions (T1, T3) · failed in 0
Met in 0 sessions · failed in 2 (T2, T3)
Met in 1 session (T1) · failed in 0
From PRD to tested prototype.
- 1
Open it in a build tool
Pick a tool. Euler turns the PRD into a short build prompt you can edit before it opens.
- 2
Save the prototype
Paste the link when it's built. It stays on the PRD, which tells you if it changes after.
- 3
Let Claude check it
Claude opens the prototype in a browser and tries each acceptance criterion, citing the steps it took.
- 4
Test it with customers
Euler turns the criteria into tasks for participants. Attach your notes, and each criterion comes back met, not met or unresolved, with the quotes behind it.
A prompt that looks like your product
Connect a GitHub repo and Euler reads your frontend's stack and theme: framework, component library, styling, icons and color tokens. The prompt asks the tool to match them, so the prototype looks like yours, not the tool's default. Already loaded your design system into Lovable, Bolt or v0? The prompt asks for it by name instead.
- Add your Storybook and the prompt names the components each flow uses
- Only names and theme values leave Euler, never your code
- The PRD's customer evidence stays out of the prompt
# Non-Goals
- No real delivery backend
# Look and feel
- Stack: Next.js (React) · Tailwind CSS v3 · shadcn/ui · lucide-react icons
- Theme tokens: --primary: 271 76% 53%; --radius: 0.75rem; …
- Dark mode: yes
Look and feel from acme/web-app (its stack and theme colors go to Lovable).
The link lives on the PRD
Euler records the handoff when you open the tool. Paste the link when the prototype is ready, or add one you built somewhere else.
- Kept with the prompt you actually sent
- Flagged when the PRD changes after it was built
- Unfinished handoffs are visible only to you
notifications-digest.bolt.new
Bolt · 3d ago
Opened in Lovable, no link yet
Lovable · 1h ago
Claude tries it before your customers do
One click and Claude opens the prototype in a fresh browser and works through the PRD's acceptance criteria. It clicks through, fills forms with made-up test data and takes screenshots. Each criterion comes back met, not met or couldn't check, with the steps behind it.
- A verdict must cite a step that shows it, or it reads couldn't check
- Kept to the prototype and the assets it loads
- Nothing it finds changes the PRD
Claude opened the prototype and tried each criterion. Each verdict cites a step it took; anything its steps don't show reads “Couldn't check”.
Settings lists each reminder with its own toggle. Turning off Live alerts left the others on.
The permission prompt opens on the first screen, before any explanation.
Nothing on screen shows when a notification is sent.
The agent said met; downgraded: no step the agent took shows it.
Tested against the PRD, not a hunch
A test turns the acceptance criteria into the checklist. Euler writes tasks for participants to attempt, never "would you use this?". A criterion is met only when two sessions show it working and none shows it failing.
- Tasks are written as goals, not as clicks
- Notes and recordings stay with the test
- Results land back on the PRD
Which acceptance criteria held across 3 processed sessions. Met needs two independent sessions and none failing.
Met in 2 sessions (T1, T3) · failed in 0
Met in 0 sessions · failed in 2 (T2, T3)
Met in 1 session (T1) · failed in 0
The details.
- Seven tools
- Lovable, Bolt and Cursor open with the prompt filled in. v0, Replit and Claude open with it copied. Claude Code gets an engineering brief instead.
- Edit before it opens
- Every prompt opens in a review step first. Change anything, then hand it off.
- Look and feel, kept current
- Euler re-reads your repo's theme each day the code has changed, and your Storybook every day.
- Criteria come from the PRD
- A test starts from the PRD's acceptance criteria, so write some before you test.
- You run the sessions
- Euler writes the task guide and reads your notes. The conversation is yours.
- Two sessions to call it
- One session is an anecdote. Met needs two, and none failing.
Where it stops.
It doesn't build it
The build tool does. Euler writes the brief and keeps the result.
Not discovery evidence
Reactions to your own prototype never reach themes, insights, search or chat. Your prototype can't vote for itself.
No opinion polls
Tasks asking whether someone would use or pay for it are dropped. What participants do is the evidence.
Only what's on screen
Claude's check sees what a browser sees. Backend behaviour, saved data and emails come back as couldn't check, never as met.
Public links, no sign-in
Claude checks a published or deployed prototype and never signs in. A localhost link has to go live first.
Questions, answered.
Only for the look and feel read from your code. A design system loaded in Lovable, Bolt or v0 works without it. With neither, the prompt asks for a standard React or Vue prototype, and the rest works the same.
Yes. Add its link to the PRD, then start a test from it.
No. It tells you whether the prototype does what the PRD says before you book sessions. Whether people can use it, and want to, is what the sessions are for.
No. They're kept as prototype-test notes, out of themes, insights, search and chat.
Stop drowning in feedback. Start shipping the right thing.
We're putting the agents through their paces. When early access opens, the waitlist hears first.