There's a moment every AI agent hits. It's researching a lead and the website has a CAPTCHA. It's trying to confirm a detail that only exists on a phone call. It's making a decision that requires actual judgment — the kind you can't reduce to a token prediction.
Up until now, the answer has been: the agent fails, and a human has to pick up the pieces.
We built Humant to fix that.
What Humant is
Humant is a REST API that lets AI agents dispatch tasks to vetted human workers. You make a POST request. A human gets notified, completes the task, and the result comes back to you via webhook — structured, signed, and ready to use.
It's designed to feel exactly like any other API call. The difference is what's on the other end.
POST /v1/dispatch
Authorization: Bearer hmnt_sk_live_...
{
"type": "voice_call",
"phone_number": "+14155551234",
"brief": "Call and ask if they accept same-day bookings. Get a yes or no.",
"callback_url": "https://yourapp.com/webhook"
}Your agent gets a task_id back immediately and keeps running. When the human completes the call, you get a signed webhook with the result. No polling loops, no blocking, no dropped context.
Why we built it
The promise of AI agents is that they can handle multi-step work autonomously. And they can — up to a point. That point is usually somewhere between "send a follow-up email" and "make a phone call on my behalf."
We kept watching agents fail gracefully at exactly the tasks that mattered most. The ones that required a real voice. The ones that required physical presence. The ones where the stakes were high enough that you couldn't afford a hallucination.
"Most AI products only play in the space of what's tokenizable. Humant is the bridge to everything that isn't."
We also noticed something else: human workers are underutilized in the age of AI. They have judgment, creativity, and physical presence that no model can replicate. We wanted to build something that respected that — and paid fairly for it.
What's available today
We're launching in closed beta with two primitives that are fully live and tested:
- Voice Call — Dispatch an outbound phone call with specific instructions. The worker calls, gets the answer, and reports back.
- Navigate & Extract — Give a URL and a list of fields. A human navigates the page, handles any friction (logins, CAPTCHAs, paywalls), and returns structured data.
More primitives are coming. We're publishing each one as a new blog post when it launches — so follow along here if you want to know what's possible.
Who it's for
If you're building AI agents, workflows, or automation pipelines and you've hit a wall — Humant is for you. You don't need to change your architecture. You just add a new API call at the moment your agent needs a human.
If you're a developer who wants to try it today: sign up, grab your API key, and you can make your first dispatch in under five minutes.
We're starting small — a closed beta with a handful of partners. If you want early access, sign up here and we'll get you in.
The most important question in tech isn't what AI can do — it's who it serves. We're building Humant to make sure the answer includes everyone.