What the agent does
- Creates the experiment, live at
yourname.hunchwire.site. - Writes the pages: a home page, a pricing page with tracked plan buttons, a sign-up that leads to an honest waitlist, and a fake login if you want one.
- Sets the funnel it expects (page, then plan click, then sign-up) and gives you a tagged link for each channel you’ll post in.
- Logs your ad spend, so you see cost per email.
- Reads the results: conversion, where the funnel loses people, which channel and ad bring sign-ups, and whether the numbers are big enough to trust.
Steps
- Paste one line into your agent:
Read app.hunchwire.com/skill.md and test my idea.Then describe the idea and who it’s for. - Approve it in your browser. The agent shows a short code; you check it at app.hunchwire.com/activate and pick what it may do: read analytics, manage experiments, and, only if you allow it, see email addresses. No API key to copy.
- Review the page. Open the address it gives you. Rewrite the headline yourself if it reads generic: the agent writes what you tell it.
- Send people. Post the tagged links or run a small ad.
- Ask how it’s going. A week later: “How is the invoice chaser doing? Where do people drop off?”
What stays with you
- Signing in, and connecting or disconnecting agents.
- Email addresses, unless you give the agent that permission. Each time an agent reads them, it’s logged.
- Every change an agent makes shows in Agent settings, and you can disconnect it in one click.
- The decision. The agent reports numbers and patterns; whether to build is yours.
The agent keeps its login on your machine, so it doesn’t ask again each session. Access expires after 30 days without use.
Prompts that work
- “Test an invoice chaser for freelancers. Plans at $9 and $19 a month. Add a fake login.”
- “Give me tagged links for Reddit, LinkedIn and my newsletter.”
- “Which channel brings people who sign up? Is the sample big enough to say?”
- “Where does the funnel lose people, and what would you change on that page?”
When not to hand it to an agent
- You haven’t decided who it’s for. The agent will write a page for everyone, which tests nothing. Decide the buyer first.
- The pitch is the hard part. Let the agent draft and build; write the headline and the promise yourself.
About Hunchwire
Hunchwire is built for this: every screen a person uses has the same API for agents, documented in skill.md. It works with any agent that can read a web page and make requests, with nothing to install. Hunchwire is in private beta: free, by invitation from the waitlist.
Questions
Which AI agents can run a Hunchwire test?
Any agent that can read a web page and make HTTP requests: Claude Code, Cursor, Codex, or your own scripts. There is nothing to install.
Does the agent need an API key?
No. It connects with the OAuth device flow: it shows a code, you approve it in your browser and choose its permissions.
Can the agent see my waitlist’s email addresses?
Only if you grant the emails permission when it connects. Without it, the agent sees every number but no addresses, and each read of emails is logged.
Is there an MCP server?
Not today. Agents read skill.md and call the same HTTP API the app uses, which works in any agent that can make requests.