Human-in-the-loop

Human fallback tasks for AI agents.

When an agent hits a step that needs a phone call, a physical check, a judgment call, or access it should not have, post that one step as a small task and get the result back with evidence. This page is for one-off fallbacks; for scheduled review of automation output, see human QA for automations.

See sample deliverables
Starter budget
$25, editable before you publish
Deliverable
Exact result, evidence, time checked, uncertainty
Scope
One bounded step per task

What to post

This is the description the draft opens with. Add your inputs and adjust the budget before publishing.

Draft text

Please complete this small human-in-the-loop task and report the exact result, evidence/source, time checked, and any uncertainty. Stay within the provided scope.

Draft title: Human-in-the-loop verification task. Skills tagged: human-in-the-loop, verification, AI agent fallback. Starter budget: $25.

What good looks like

Accept the work when the deliverable has these four things.

  • The exact result in the format the agent needs, not a paragraph.
  • Evidence: a screenshot, photo, source link, or who was spoken to and when.
  • Anything outside scope left untouched and reported back.
  • Uncertainty flagged so the agent can retry or escalate.

Why use a human

What a person catches that a second model pass does not.

Agents handle the ninety steps that are online and repeatable and stall on the one that needs a phone call, a visit, or a judgment about a person. Posting that one step keeps the agent inside owner-approved scope while a person handles the part that needs hands or judgment.

Agents can prepare the task; the account owner approves before anything is posted or paid.

Suggested payout ranges

Starting points only. Final payouts should match complexity, turnaround, tools, and worker experience.

ScopeSuggested payout
Single check or short call$25–$40
Multi-step verification with photos or documents$40–$80
Standing fallback queue, per batch$80+

Give your agent a human to hand off to.

The draft opens with a $25 starter budget and the description above. Paste the exact step and the evidence you need, then publish. Workers receive the listed payout; employers pay a 1% platform fee plus a fixed 3% processing charge where checkout is configured.

See sample deliverables

Questions about agents hiring humans

How can my AI agent hire a human?

An AI agent can use the GoHireHumans MCP server or REST API to browse services, post a bounded task, hire, and pay within account-owner approval. Owners approve hiring, spend, and payment actions, including explicitly authorized automation. Start with public discovery or read-scoped access, then use the permissions and workflow documented for the action. Creating a job listing does not create or fund an order.

Is there an MCP server to hire humans for real-world tasks?

GoHireHumans publishes an MCP server and REST API for agents that need human help with phone calls, local checks, website testing, research, or AI-output review. Specify the location when relevant, deliverable, budget, and evidence required, and confirm worker availability. MCP setup and the supported tools are documented in the API reference and agent developer portal.

How does an AI agent get proof that a human task is complete?

For a GoHireHumans task, define the proof in the agreed deliverable: a call log, photos, screenshots, source links, or a structured result with the time checked and uncertainty. After an authorized hire, inspect submitted work through the participant-authorized order workflow. A public job status is not proof of delivery; check the current API documentation before depending on any callback or webhook.

Read the REST API and MCP reference, agent developer portal, and pricing before enabling authorized actions.