Human judgment,
callable by your AI.
When an agent hits uncertainty, Pukot turns the problem into small, independent human judgments, combines the signal, and returns structured output to the workflow.
Agents do not need another dashboard. They need somewhere to send what they cannot resolve.
Pukot is the layer between cheap, abundant machine cognition and scarce, expensive expert attention.
Agent hits uncertainty
A workflow reaches ambiguity, verification risk, disagreement or a decision it should not manufacture.
Pukot structures it
The kernel minimizes context and converts the problem into bounded microjudgments.
Nodes answer independently
Multiple human nodes respond without seeing or influencing one another. Dissent is preserved.
The workflow continues
Pukot aggregates the signal, recurses when useful, and returns a structured result to the caller.
A human cognition layer for production AI.
Use Pukot for the edge cases where another model call is not enough and loading a full-context expert is wasteful.
const result = await pukot.resolve({
type: "compare",
question: "Which output is clearer?",
options: ["A", "B"],
responses: 5
});
// Structured signal returns
{
answer: "B",
agreement: 0.80,
dissent: ["A explains the risk faster"],
human_objection: "Neither states the next step"
}
Minimum context by design.
The public worker network should never become a back door into customer systems.
Your judgment has value.
Answer short, independent tasks on your phone and get paid for completed work. General pilot tasks do not require a CV or full business context. Pukot gives you only what you need to answer.
Initial pilot: 18+ only. Task availability and earnings are not guaranteed. Pilot payouts may use Wise, Revolut, bank transfer or cash where practical.
Put real unresolved agent work through Pukot.
We are looking for bounded production workflows where human judgment is currently expensive, manual or hiding behind repeated model retries.