Stunt Double is the AI-first platform for product teams. Its actors step in for your users, go through your product, prototype or Figma design in a real browser, and show you where people get stuck, with the evidence.
A checklist run in the Parlour demo workspace: the First-time diner actor books a table on Ondine. Two of three checks passed. The failed check: Deposit amount and refund window are stated before payment. The deposit (102.50, with no currency shown) and the 48-hour refund terms first appear on the payment step, below the card fields and the pay button. The diner said: The deposit only became clear at the last step.
01 / 08
A stunt double takes the fall before the star does, and the fall is real. An actor does the same for your users. It is an AI user with a background, a goal and its own inbox, and it goes through your real product, so you find where people get stuck before a customer does.
Review a Figma design or prototype with actors cast as the people it is for, and see where they stop before anything is built.
Run a round of user interviews with a panel of actors and get the answers back in minutes, each at the step it was given.
Check your critical flows on every deploy or pull request, and see what broke with the screenshot and the steps to reproduce it.
02 / 08
A checklist run logs every page the actor opened and every action it took, with its reasoning. Fix what failed, rerun the same checklist, and see it pass.
03 / 08
One guide, a whole panel of actors at once. Follow each one as it works and read each answer at the step it gave it, so a round of research is back in minutes.
04 / 08
Open any finding for the step, the screenshot and the reasoning behind every pass and fail. This one points at the fix: show the refund window with the deposit, before guest details.






02 / 08
A checklist run logs every page the actor opened and every action it took, with its reasoning. Fix what failed, rerun the same checklist, and see it pass.


03 / 08
One guide, a whole panel of actors at once. Follow each one as it works and read each answer at the step it gave it, so a round of research is back in minutes.
04 / 08
Open any finding for the step, the screenshot and the reasoning behind every pass and fail. This one points at the fix: show the refund window with the deposit, before guest details.




05 / 08Run 03
Screenshots and recordings: every screen the actor saw, and the run on tape.
User journey: each step the actor took, with the reasoning behind it.
Topology: every screen the run passed through, on one map of the product.
Run history: what changed since the last run, check by check.
06 / 08
Describe what you need. Runner sets up the work and takes you straight to the result.
07 / 08
Start runs from the tools you already use. Findings land in the same place.
Run checklists and interviews from a conversation, without leaving the conversation.
Ask for a run in a channel and get the findings back where your team already talks.
Delegate an issue to Stunt Double and get the answer, or the run, back in the issue.
See what a pull request broke before it merges, right on the PR.
Have actors review a frame in character, with what they find pinned to the canvas.
Point actors at your latest preview or production deployment, even behind deployment protection.
08 / 08
Agents now browse, compare and buy for real people. Send an actor to see where they get stuck.
Run the same checklists and interviews for agents as for people, and fix what blocks them first.
See your site the way agents do.
Claude, ChatGPT, Gemini and more try to find you, understand you and act for a customer, scored in eight categories from discovery to checkout.
Know exactly what to fix.
Every score comes with the frictions behind it, worst first, and what to change for each.
Watch the score move.
Ship the fix, rescore, and see each category climb week on week.
#1 on the Stunt Double Index
Supabase
supabase.com
92.0
What to fix next
Delegated access: No discoverable MCP server. Agents and AI clients have no way to discover one automatically.
Delegated access: No OAuth/OIDC discovery metadata. Agents cannot bootstrap a scoped, revocable session; any delegation falls back to password sharing.
Delegated access: Rate limit response headers: failing (Is Agentic).
Supabase’s live scorecard, the top-ranked of 553 websites on the Index, as of October 19, 2026.
Try on your product, website, or a prototype for insights in minutes.