Human in the loop — as infrastructure

# Human in the Loop for Reliable AI:  
One API Call Away

When your agent hits something only a person can do — review a doc, make a call, judge ambiguous data — hitl.ph routes it to a real human in minutes, and the result flows straight back into your workflow.

 [How it works ↓](#how)

Balanga, PH — real ops floor 130+ specialists Powered by Yoonet

routing to Balanga, PH…

![Chaotic multicoloured marbled ink streams converging through one point and leaving as a single clean brushstroke](/media/hero-spectrum.webp)

01 — The gap hitl.ph

## Agents are brilliant. Until they hit a thing only a human can do.

Every serious agent workflow runs into the same wall: a step that needs real world judgement, a voice, a pair of eyes, or someone willing to be accountable for the answer. Today that step just fails, or quietly hallucinates. hitl.ph gives it somewhere to go.

![](/media/failure-signal.webp)

## Why AI Projects Need Human Validation

02 — The failure rate MIT · State of AI in Business 2025

95%

of enterprise AI pilots fail to deliver measurable impact.

Not because the models are weak. Because almost nobody gives the workflow a reliable, structured way to reach a human when it matters. That is the missing layer, and it is the layer we build.

![Outer Edge](/outeredge-logo.jpg)

**AI without systems is just noise.** Outer Edge is our sister company: systemising and preparing your business for AI deployment and use.

[Get your business ready](https://outeredge.nz)

## What you can send to a human in the loop

03 — What you can send Five task types

![Review a document](/media/task-document-review.webp)

task.type = "document.review" 01

### Review a document

Something a model shouldn’t sign off on alone — a contract clause, a medical note, a flagged edge case. A trained person reads it and decides.

![Make a phone call](/media/task-voice-call.webp)

task.type = "voice.call" 02

### Make a phone call

Confirm a booking, chase a supplier, verify a detail. The jobs that still need a voice on the line.

![Validate ambiguous data](/media/task-data-validate.webp)

task.type = "data.validate" 03

### Validate ambiguous data

Two records that might be the same person. A field that doesn’t parse. The judgement calls models guess at and get wrong.

![Take a photo. Check the world.](/media/task-world-capture.webp)

task.type = "world.capture" 04

### Take a photo. Check the world.

Confirm something physically exists, capture proof, eyeball a real world state your agent can’t see from a server — then send back what it found.

![Make the call you can’t automate](/media/task-decision-own.webp)

task.type = "decision.own" 05

### Make the call you can’t automate

The decision that needs real accountability — a person who owns the outcome, signs their name to it, and stands behind it. Not a confidence score.

## How human in the loop works

04 — How it works One call out. A human back.

Step 01

### Call it.

One `POST /tasks` or a `hitl` CLI command. Send a brief, the task type, and how you want the answer back. API-key auth, installed in under a minute.

Step 02

### A human owns it.

Your task lands on our Balanga ops floor and a trained specialist picks it up — end to end, inside your working window. Not a crowdsourced lottery. A managed team with someone accountable.

Step 03

### It flows back.

Poll `GET /tasks/{id}` or take it on a return hook. A structured result your agent can read and keep moving — no human in your loop, just one on call.

[How an AI agent hands off to a human →](/use-cases/#handoff)

05 — The API POST /tasks

## A human in the loop API, not another dashboard.

Wire it in as a human validation API when model output needs a person’s sign-off before it ships, or as a human review API when a judgement call needs a name attached. Either way it is the same endpoint: your agent sends a task, a specialist owns it, and structured data comes back. The full integration shape is in [the docs](/docs/).

06 — The workforce Balanga, Bataan, PH

## Your AI’s  
Extended Workforce

hitl.ph runs on Yoonet — a real team in Balanga, Philippines, already doing operational work every day for businesses across the world. You’re not renting an API that happens to find strangers. You’re plugging into a trained, managed workforce with a timezone that covers yours and a name attached to every result.

130+

Specialists on the floor

PH

Balanga, Bataan

24/5

Your working window

1:1

A human owns each task

![A hand-inked grid of watercolour dots, one blooming into a full prismatic burst — one specialist lit up for the task](/media/workforce-constellation.webp)

07 — The bigger picture A Yoonet initiative

## AI will cost the Philippines millions of jobs. Everyone is missing the point.

The warnings are right about the risk: a country built on outsourcing is exposed like nowhere else on earth. What they keep missing is the industry rising in its place. Human in the loop work, real people exercising judgement inside AI workflows, is set to become one of the fastest growing industries there is.

We intend to champion that change: educate the industry on the shift, train the workforce it threatens, and go on to create far more opportunity than is ever lost.

![Small hand-inked marks ascending diagonally across paper, watercolour ripples blooming beneath each step, a faint pigment archipelago below](/media/mission-ascent.webp)

 [![Yoonet](/yoonet-mark.png) **Yoonet** Championing the shift — yoonet.io](https://yoonet.io)

08 — Access

## Give your agents a human to call.

We’re onboarding the first agencies now. Request access and we’ll send you a key.

 Early access · no spam · a real person reads these

***

Source: https://www.hitl.ph/
This is the markdown representation of that page, served to clients that send `Accept: text/markdown`.
Machine-readable index: https://www.hitl.ph/llms.txt · Knowledge bundle: https://www.hitl.ph/okf/index.md