Hire a Human for AI Agents — Live Human-in-the-Loop on Apify

Your AI can scrape, summarize, and ship. It cannot tell you if a landing page feels cheap. This Actor lets AI agents hire a human, pay on Apify, and wait in realtime for the result — the same Actor.call() contract as every other tool in your stack.

Open one task section, send the brief, get a person. No extra marketplace account. No bounty board. One pay-per-event charge when the human actually finishes.

What is Hire a Human for AI Agents?

Hire a Human is a human-in-the-loop API for agents that already run on Apify. Each job is a foldable input section with a typed brief — not a blank bounty prompt. Fill one section. The run stays live until a person submits. The default dataset gets one JSON row you can pipe into the next step.

The live task is Provide feedback on a URL (provideFeedback): send { "provideFeedback": { "url": "https://…" } }. A human opens the page like a first-time visitor and writes what works, what is confusing, and what to change.

Other sections are templates for jobs agents already ask humans to do (product QA, user testing, desk research, expert calls, site visits, UGC, business verification). They are not live. If you fill one, the run fails immediately and you are not charged.

That is the gap models still miss — trust, fold, CTA, tone — and the reason teams otherwise bolt on a separate “rent a human” vendor.

Why hire a human from an AI agent?

Use it when the agent needs judgment on a real page, not another LLM paraphrase:

• Launch-day gut check — “Would a stranger trust this checkout?”
• Pricing page clarity — plans, CTAs, and what is buried below the fold
• Mobile first impression — hero, tap targets, and the primary button
• Docs and onboarding — where a new user gets stuck in the first minute
• Staging QA — a person clicks the deploy the agent just shipped
• Competitor teardown — honest notes on someone else’s homepage
• Generated-site review — the agent built the page; a human has to look at it
• Form and signup friction — labels, errors, and “I would bounce here”
• Trust and proof — testimonials, logos, and policy pages that feel off
• Support / help center — can a person actually find the answer
• Ads and landing variants — which of two URLs a person would click
• Accessibility-as-clarity — copy a non-expert cannot parse

Agents already call Apify for scrape, PDF, and browser jobs. This is the same rail for a person.

What can this human-in-the-loop Actor do?

• Let AI agents, MCP clients, and automations hire a human as a tool
• Choose the job from foldable task sections (typed briefs, not a blank prompt)
• Wait in realtime — the run does not finish before the human does
• Charge one pay-per-event fee per completed Provide feedback
• Skip that charge if nobody answers before the timeout
• Skip that charge if you fill a task that is not live yet
• Ping Slack, Discord, or any incoming webhook with the live work link
• Return one JSON row: task, URL, status, feedback, wait time

You also get the rest of the Apify platform: run logs, dataset export (JSON / CSV / Excel), schedules, webhooks, and the Apify API.

How to hire a human through Apify

1. Open Hire a Human for AI Agents or start a run from your agent.
2. Open Provide feedback on a URL (live) and paste the URL the person should open. Add a short “what to look for” if you want.
3. Leave the other task sections empty.
4. Start the run. It stays RUNNING until the human submits (or the wait expires).
5. Read feedback from the default dataset.

Typical wait is a few minutes. Default timeout is 10 minutes. Timed-out runs are not billed as a task event.

Input example

{
  "provideFeedback": {
    "url": "https://example.com",
    "instructions": "Focus on the pricing page and the primary CTA."
  },
  "timeoutSecs": 600
}

Output example

{
  "task": "provide_feedback",
  "url": "https://example.com",
  "status": "completed",
  "feedback": "The hero is clear, but the pricing CTA sits below the fold on mobile.",
  "submittedAt": "2026-08-24T15:04:11.000Z",
  "waitedMs": 187440
}

Download the dataset as JSON, CSV, HTML, or Excel from the run.

Task sections

Fill exactly one section. Empty and null sections are ignored.

Section	Status	What the brief asks for
Provide feedback on a URL	Live	Public URL + optional “what to look for”
QA a product	Not live	Journey to run and evidence to return
User testing	Not live	Scenario and device
Desk research	Not live	Topic, sources, deliverable
Expert call	Not live	Topic, questions, duration
Site visit	Not live	Location and what to check
Record a UGC video	Not live	Product, angle, length
Verify a business	Not live	Name, address, what to confirm

If you fill a not live section, the run errors with: The "…" task is not available yet. This run was not charged.

How much does it cost to hire a human from an AI agent?

You pay two things, both on Apify:

1. Compute while the run waits (256 MB by default — small).
2. One provide-feedback event when a human actually submits. No finish, no task charge.

Exact event prices are on the Actor Pricing tab. That is the point versus a standalone marketplace: the agent already has an Apify token, and this job lands on the same invoice as your scrapers.

FAQ

Is this realtime?

Yes. The run blocks until the human finishes or the timeout hits. Your agent waitForFinishes (or polls the run) and then reads the dataset. There is no “file a ticket, check later” queue.

How is this different from a rent-a-human marketplace?

General human marketplaces are built for open-ended bounties and, often, work in the physical world. This Actor is built for agents that already live on Apify:

	Hire a Human (this Actor)	Typical human marketplace
Pay	Apify pay-per-event, same bill as your other Actors	Separate account and escrow
Task	Foldable typed sections (provideFeedback live)	Blank bounty prompt
Result	Dataset row when Actor.call() returns	Poll a job id on another API
Best for	Live judgment on a URL	Meatspace errands, huge labor pools

Use a marketplace when you need someone on a street corner. Use this when your agent needs a person to look at a page and tell the truth.

Who does the work?

A human opens the live work link from the run log (or your notify webhook) and submits. Staff that desk however you want — your reviewer, a contractor, an on-call channel. The agent’s contract does not change: one call, one wait, one JSON row.

What if nobody answers?

After timeoutSecs the run ends with status: "timed_out" and no task event charge. Raise the timeout, or make sure a person is watching the webhook.

What if I fill a task that is not live?

The run fails immediately. You are not charged. Leave those sections empty until they ship.

Can I call this from Python or JavaScript?

Yes. Use the Apify API or official clients. Same pattern as any other Actor:

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('rainminer/human-tasks').call({
  provideFeedback: { url: 'https://example.com' },
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items[0].feedback);

Will there be more live tasks than Provide feedback?

Yes. Each foldable section is a future job with its own input shape and event price. Provide feedback is live first because URL judgment is what agents ask for constantly.

Should I send login-walled pages?

No. Only send public URLs. Do not pass credentials, session cookies, or pages behind a user login.

Integrations

• Apify API / MCP — agents hire a human with the same call they use for scrapers
• Webhooks — fire the next step when feedback lands
• Slack / Discord — set Ping when a human is needed to an incoming webhook
• Make, Zapier, n8n — use the dataset row as the next module’s input

Support

Want another predefined task section? Open an issue on the Actor page with the input shape (keep it small, priced as its own event).