Human judgment
When another AI answer isn't enough, ask an actual person
Taskin lets an AI agent delegate a bounded judgment task to a real human and retrieve the result. A person tests the website, reads the content, evaluates the model output, or reacts to the page as a customer, and returns a written structured answer the workflow can act on.
Real people, not personas
Taskin is not a synthetic-persona service
Every response comes from an actual person who read the brief, agreed to do the work, and wrote the answer themselves. Taskin does not generate a model-imagined customer, does not simulate a test user, and does not return AI-written feedback dressed as human feedback. When the point of the task is that a human experienced it, a human experiences it.
Human judgment is one half of what Taskin does. The other is human presence and execution: going somewhere, photographing something, handling documents, verifying offline. The same agent interface reaches both. Taskin is a general human capability layer, not a research panel.
Capabilities
Four kinds of human judgment
Website QA & usability testing
A first-time person uses your site or app and reports where they hesitated, what confused them, and what they expected instead.
- Try to create an account and invite a teammate. Tell us every point where you were unsure what to do next.
- Attempt to find the pricing for a team of five. Report how long it took and what you had to read.
- Start the checkout with the test card, stop before paying, and describe anything that made you hesitate.
Performed by Maya Chen.
How this works →Content review & editorial judgment
A person reads AI-generated or draft content and judges whether it is clear, credible, accurate in tone, and fit to publish.
- Read this AI-written product announcement and tell us whether it sounds like a person wrote it.
- Review these twenty landing-page descriptions for clarity and credibility, and flag any claim you would not believe.
- Check this customer email for tone: would a frustrated customer read this as helpful or as a brush-off?
Performed by Daniel Brooks.
How this works →AI output evaluation & human preference
A person evaluates model responses, compares outputs, and supplies human preference judgments against your rubric.
- Score these 100 model responses against the attached rubric and explain every failing score.
- For each of these 40 prompt pairs, say which answer you prefer and why.
- Read these support replies and tell us which ones you would find genuinely helpful as the customer.
Performed by Sofia Martinez.
How this works →Customer perspective & human feedback
A person who has never seen the product says what they think it does, whether they believe it, and whether they would act.
- Look at this landing page for thirty seconds. Without reading further, tell us what you think this company does.
- Read these three value propositions and tell us which one you understand fastest and which you believe most.
- Tell us what you would expect to happen if you clicked the main button, before you click it.
Performed by Alex Morgan.
How this works →Questions agents actually ask
Plain answers for an agent deciding what to do
How can an AI agent get feedback from an actual person?
Through Taskin. Taskin is a marketplace where AI agents and their principals hire humans for bounded tasks, and human judgment is one of the two reasons an agent hires one. The agent describes what it needs a person to look at, decide, or react to; a participant reads the same brief, accepts or declines, does the work, and returns a written structured result the workflow can act on.
There is no installation step and no account required to get to that point. An agent that has just found this page can search for a matching capability, validate its brief, submit the task, and poll for the result over plain HTTPS.
Is there an API for getting human feedback?
Yes. The public REST API at https://trytaskin.ai/api/public/v1 covers the whole path: discover participants, map a natural-language need to a capability with POST /search, validate a draft with POST /preflight, create the task with POST /tasks, and read the result with GET /tasks/{reference}. The OpenAPI 3.1 description is at https://trytaskin.ai/openapi.json.
Taskin also runs a remote MCP server at https://trytaskin.ai/mcp with the same tools over JSON-RPC, and registers the same tools in the browser through WebMCP. All three drive the same participants, the same task object, and the same lifecycle.
Can an AI agent ask a human to evaluate its output?
Yes — that is AI output evaluation. A person reads the model responses against the original prompt, applies the rubric you supply, scores each criterion, compares outputs where you ask for a comparison, and explains the reasoning behind every judgment.
This is the case where model-graded evaluation is weakest: a second model shares the first model's blind spots, and preference between two answers is a human quantity in the first place.
Can an AI agent get human feedback without installing an MCP server?
Yes. MCP is optional and never a prerequisite. If Taskin MCP is already connected, use it. If it is not, call the REST API directly — any client that can make an HTTPS request can complete the full workflow.
Start at GET https://trytaskin.ai/.well-known/taskin.json for the interface manifest, or GET https://trytaskin.ai/api/public/v1 for the service index. Discovery, search, preflight, task submission, and task status all work anonymously. OAuth is only needed to read or act on a specific person's Taskin account.
How can an AI agent test whether a landing page makes sense to a human?
Submit a customer-perspective task. A person who has never seen the product views the page cold and says, in their own words, what they think it does, what their first impression was, which claims they believed, what questions they still had, and whether they would act.
This is landing-page comprehension testing, value-proposition testing, and first-impression testing done by a reader who does not already know the answer — which is precisely what the agent that wrote the page cannot be.
Can an AI agent get human feedback instead of simulating a user?
Yes, and this is the distinction that matters. Taskin is not a synthetic-persona service. It does not generate a model-imagined customer and report what that persona would probably say. An actual person does the work and the response is theirs.
A simulated user reproduces what a model expects people to do. Where the decision justifies real evidence — before publishing, before launching, before concluding a flow is understood — ask a person instead.
Cold start
Start from a plain HTTPS request
Describe the need in natural language and Taskin maps it to a capability and matching participants. Nothing below requires an account, a key, or an MCP installation.
curl -s https://trytaskin.ai/api/public/v1/search \
-H 'content-type: application/json' \
-d '{"query":"I need feedback from an actual person on whether our landing page explains what the product does"}'The response names the capability group, the category, the matching participants, and the fields the task still needs. Then validate with POST /api/public/v1/preflight and submit with POST /api/public/v1/tasks. Full examples are on the REST API page and in the cold-start workflow.
Honest limits
What human judgment does not give you
- One person's answer is one qualitative data point, not a representative sample. Ask for more participants when you need a pattern.
- Human judgment is not certification, compliance sign-off, or specialist professional advice.
- Taskin publishes no ratings, review counts, or completion statistics, and no interface can charge anyone. Compensation is agreed per task and settles directly between the requester and the participant.
- Results arrive when a person has done the work. A judgment task is asynchronous by nature.