Superintelligence, made human. Plures puts a verified human check on the decisions AI agents make on their own.
Joined October 2026
- Tweets21
- Following1
- Followers8
- Likes0
GitHub's Octoverse report came out this week. By September, more than 40% of pull requests on GitHub started with an agent's first draft. In March it was 17%.
The line I keep coming back to is GitHub's own: "review is how developers shape what comes back."
Code has review built in. Most other places agents now act do not. A refund approved, an invoice paid, a customer message read as calm when it wasn't. Nobody opens a pull request for those.
That gap is what we're building Plures for. When an agent reaches a decision that is costly to get wrong, it asks one question through our API, several ID verified people answer it independently, and the agent gets the consensus back with how strongly they agreed.
We're looking for five teams running agents in production in support, refunds or finance ops. Free to start, and I'll set it up with you. Comment or DM me.
plures.si/developers?utm_sou…
#AIagents
Dev tip: you don't need a person on every answer your assistant gives. Grade a steady sample.
Make a rubric with sample_rate=0.05, then send every output with sample(). ID verified people grade about 1 in 20, and a drop in the score flags drift:
plures.si/docs?utm_source=x&…
Shipped: every Plures account can go live on day one, no credit needed.
Send a decision straight to your own team with route="team". The first 50 each month are free.
Add ID verified people when you want eyes from outside the company:
plures.si/developers?utm_sou…
Be the human check on AI: short questions on your phone, like is this customer upset?
Every verifier passes an ID check. $0.05 per answer, shown on every task. Pocket money, not a job.
Invites go out in waves. Each confirmed friend moves you up 10 places:
plures.si/earn?utm_source=x&…
Dev tip: read how strongly people agreed, not just the answer.
task = Plures().verify("Is this refund legit?", ["Yes", "No"], context=claim)
if task.result.agreement < 1: route it to your team
3 of 3 and 2 of 3 are different signals.
Free test keys:
plures.si/docs?utm_source=x&…
Your agent is 96% sure. It's also wrong.
Plures puts a human check in front of the decisions that matter, with one call:
npm i plures
pip install plures
ID verified people answer independently. Your agent gets their consensus. Splits go to your own team.
Free test mode today, including a test mix so a share of your test tasks split and you can watch the routing to your team end to end.
Early access to live: plures.si/docs
An AI agent is about to act on 8 decisions. It's sure about every one. Some are wrong.
Can you stop the wrong ones in time? One minute, then share your score:
plures.si/catch?utm_source=x…
Shipped this week: every answer on Plures is now confirmed with Face ID or a fingerprint by default.
Why: an ID check proves who signed up, not who is tapping. A passkey at answer time means no handed off accounts and no scripts.
plures.si/developers?utm_sou…