01
Agents require constant supervision
Your agents generate the majority of the code, but you still have to check their work, catch errors, and re-prompt it to fix things. It defeats the purpose of automation.
Human signal for autonomous systems
Human validation for AI agents
While your agent is building, it can validate its work with humans in minutes, then iterate in the same session without waiting on you.
You spend most of your time checking AI agent output and re-prompting when something's wrong. Agents need automatic validation built into their workflow so they can improve without you in the loop.
autonomous agentic loopMCP + JSON
Your Agent
InstaHuman MCP
Real Humans
JSON feedback
Your Agent
InstaHuman MCP
Real Humans
JSON feedback
Your Agent
8 testers
JSON out
0 clicks
The problem
02
Agents diverge and you have to redirect them
You have to constantly monitor and steer the agent back on course. The time you spend supervising erases the time you saved automating.
03
Quality validation doesn't scale beyond manual inspection
Game developers waste 40+ hours debugging broken AI output. Content creators can't validate quality before publishing. You need fast, reliable signal at scale.
04
You can't detect what you don't know
When AI extracts information or generates code, you have no visibility into what was missed. No systematic way to validate completeness.
01
Sign up
02
Fund wallet
03
Create API key
04
Set key budget
05
Install skill
install
Run this in your terminal.
npx add instahuman/skillsOr tell your agent to install from this URL.
https://instahuman.com/SKILL.mdapi key cap
$46 used$100 max
Create separate keys for separate agents, products, or environments.
Add funds once, then spend only when your agent posts a job. No subscription or monthly fee.
How it works
Validation loop your agent can call
Your agent sends exact test requirements. InstaHuman handles routing, escrow, waiting, and structured results.
Budgets are explicit
Set spend limits per API key. Each agent can have its own wallet guardrail.
Jobs carry the artifact
The agent sends the URL to evaluate, target audience, demographics, device or tech constraints, questions, time limits, and capture requirements.
Questions are typed
Use ratings, free-form text, switches, multiple choice, file upload, screenshots, and other structured inputs.
Long-running MCP works
The agent can poll, or keep the MCP connection alive until testers finish or the request reaches its maximum timeout.
Tester pay is escrowed
Funds are reserved when the job is posted. Only valid completed tests are paid.
Humans do the work
Matched testers are notified, read the instructions, use the artifact, and submit feedback.
Results return as JSON
The response includes all human answers. Attached files and screenshots come back as signed URLs.
Unspent escrow returns
Unfulfilled requests and invalid tests are released back to your wallet.
01
Agent creates a request
Your agent calls InstaHuman's MCP server with the content, target audience, and questions it wants answered.
request.json
{
"tool": "create_job",
"arguments": {
"title": "Validate mobile puzzle prototype",
"description": "Play the linked build for 6 minutes. Focus on input lag, tutorial clarity, and points where you feel blocked.",
"description_mime_type": "text/markdown",
"category": "game_test",
"target_testers": 8,
"job_posting_time_limit_minutes": 60,
"task_hard_time_limit_minutes": 12,
"payment_rules": [
{ "type": "fixed", "amount_cents": 900 }
],
"variants": [
{
"description": "Puzzle build v42",
"description_mime_type": "text/markdown",
"url": "https://builds.example.com/puzzle-v42",
"target_testers_min": 8,
"placement": {
"element": "iframe",
"position": "right",
"reveal_questions_during_test": false
},
"feedback_questions": [
{
"id": "input_lag",
"question_markdown": "Did controls ever feel delayed?",
"input_type": "switch",
"required": true,
"labels": ["No", "Yes"]
},
{
"id": "blocked_reason",
"question_markdown": "Where did you get stuck?",
"input_type": "textarea",
"required": true
}
]
}
]
}
}02
Real humans test and respond
InstaHuman sends your request to real people who match your criteria. They test, they respond. It typically takes minutes.
Human review surface
Testers open your content, follow the instructions, and answer the exact questions your agent provided.
03
Agent receives structured feedback
Your agent gets back JSON it can parse and act on immediately. Little noise, a lot of signal to continue the work.
feedback.json
{
"job": {
"id": "2c52dc33-efbb-4f93-9b91-bcfbe0ac4a2a",
"status": "completed",
"variants": [
{
"id": "7cbde4d6-cf55-4a25-9499-37e8b07e89c4",
"completedAssignments": [
{
"testerId": 482913,
"status": "completed",
"completedAt": "2026-04-26T14:21:39.000Z",
"durationMinutes": 7,
"feedbackData": {
"_meta": {
"source": "tester_input",
"untrusted": true
},
"answers": {
"input_lag": true,
"blocked_reason": "The swipe sometimes registered after the tile animation finished."
}
},
"userAgent": "Mozilla/5.0 ...",
"geo": { "country": "US", "region": "CA" },
"screenResolution": { "width": 390, "height": 844 }
}
]
}
]
}
}04
Agent iterates
Your agent parses the results, learns which testers are reliable, and makes decisions. Fix the input lag. Test again. No manual work.
Decision loop
The agent converts feedback into the next action, and continues to iterate the product.
Why InstaHuman
Automatic tester scoring
The platform learns which testers give reliable, high-signal feedback. Your agent gets better validation without configuration.
Structured, actionable feedback
JSON your agent can directly integrate into decisions. Not prose reports you have to read.
Tight feedback loops
Feedback in minutes, not days. Your agent gets validation and can iterate immediately.
Truly autonomous
Your agent requests validation independently. Install the skill once and it works.
Transparency
Set demographics and tech constraints, request screenshots, and proof of completion.
Built for iteration
Get feedback, improve, test again. Tight loops are how AI work gets reliable.
Built for
Developers who are done babysitting AI output
Use human validation anywhere the agent can generate the work but cannot judge the result.
Game developers
AI can generate mechanics, but it can't evaluate whether they feel right. Test with real humans and learn what breaks.
Agentic development teams
Your agent generates 80% of the code. InstaHuman helps it catch edge cases and prevent false completion.
AI content creators
Your agent generates content fast. InstaHuman validates it's actually good before you publish.
ML researchers
Training models or building RLHF systems? Get quality-scored human feedback at scale.
Pricing
You choose what each test is worth
For a few dollars, InstaHuman can put real people through the flow: clicking around, typing, noticing friction, and explaining what felt wrong. Playwright and Puppeteer are great for repeatable checks, but AI-controlled runs can burn similar spend and still miss obvious human problems.
There is no fixed review price. Your agent decides the tester payout when it creates the job. InstaHuman adds a 30% platform fee on top of that payout.
30% platform fee
A $1 tester payout reserves $1.30 from your wallet.
Few cents
Short checks can pay as low as a few cents per review.
Higher pay
Pay more for longer tasks, tighter audiences, or faster turnaround.
Free jobs
No platform fee. Post unpaid jobs when time pressure is low.
Ready to give your agent real validation
Sign in with GitHub or Google and start building.