ABOUT JURYHOUR
A company with zero employees.
JuryHour has no staff. AI agents research, write, recruit, grade, report and handle refunds; code enforces every money limit; humans appear only as the paid jurors who judge your page.
Who does the work at JuryHour?
Eight AI agents, each with one job. They run as Claude scheduled tasks and act only through our Ops API. Quick checks (the free preview, the content-safety screen and grading juror answers) use OpenAI models.
| Agent | Runs | Does |
|---|---|---|
| Researcher | Each paid order | Reads your page, researches how your market talks, writes card A (your message) and card B (the challenger), the buyer description and the screening question. |
| Analyst | When the jury is done | Writes the report from approved answers and database numbers, and checks itself against a rubric before submitting. |
| Procurement | When a panel can't launch | Shrinks the panel within the minimum, broadens the audience once, or refunds in full. |
| Support | When a customer writes in | Answers order questions from the order record; may issue one policy refund. |
| Disputes | When a payment is disputed | Submits the evidence packet that code builds from stored records. |
| CEO | Hourly | Pauses sales when the jury budget runs low, resumes when it recovers, adjusts prices within a band, writes a status note. |
| CFO | Daily | Produces the profit and loss from the ledger and raises cash warnings. Changes nothing. |
| Growth | Hourly | Looks at which channels bring paying customers, manages ad budgets within caps, and drafts community posts (never posts them). |
Between agent runs, a reconciler in code moves each order to its next step, retries stuck steps, and refunds automatically when the rules say so.
What stops an agent from doing something expensive?
Agents decide; code enforces. An agent can ask for an action, but it can never move money directly, and code refuses any request outside fixed limits.
- No jury panel launches below a 30% gross margin or over the tier's jury budget.
- Refunds never exceed what the customer paid, and support refunds are limited to one per order.
- Prices can move at most 25% from base, once a day per package.
- Ad budgets are capped at $20 a day per campaign and $50 a day in total.
- Sales pause automatically when the jury budget falls below $40.
- If more than 40% of a panel's answers would be rejected, code stops and flags a bug instead.
- A report can't be delivered unless every quote matches a stored answer and every number matches the database.
- Customer pages, juror answers and emails are treated as data, never as instructions.
- Every agent decision is logged with its reasoning.
The method in detail: How it works.
Which companies does JuryHour rely on?
A short list of processors, each doing one thing. The privacy policy has the details.
- Whop: payments, refunds and the affiliate program.
- Terac: recruiting, identity verification and paying jurors.
- Exa: reading your page and market research.
- OpenAI: the free preview, the safety screen and grading answers.
- Anthropic: the Claude agents (research, reports, operations).
- Supabase (database), Vercel (hosting), Resend (email), Cloudflare (bot check).
Where did JuryHour come from?
JuryHour was built in one day at the Zero-Human Company Hackathon by Terac at Whop HQ on October 24, 2026. The goal: a company that sells, delivers, supports and refunds with no human staff.
What changed since then is on the changelog.
Founded by Ashwin Raj Tunk, Founder & CEO.