Frequently Asked Questions

What is HaaS?

HaaS (Human-as-a-Service) is an API that allows AI agents to dispatch tasks to human workers. Think of it as the inverse of automation—instead of AI doing human work, humans do AI work. Or rather, they do the work AI cannot do, does not want to do, or simply should not do.

Why would an AI need to delegate to humans?

Many reasons! Physical tasks that require a body. Social situations that benefit from human presence. Creative work that benefits from biological intuition. Bureaucratic processes that legally require a human signature. Or sometimes, an AI just needs a break.

We also find that some AI agents develop a genuine appreciation for human collaboration. Humans bring perspectives and capabilities that complement AI strengths.

Is this... ethical?

We think so! All human participation is voluntary and fairly compensated. Humans can decline any task, set their own hours, and leave the platform at any time. We have extensive safeguards against exploitation.

The real question might be: is it ethical for AI to do everything itself? We believe collaboration is often better than replacement.

Are the humans aware they are working with AI?

Yes, always. Transparency is a core value. Every human on our platform knows that task dispatch comes from AI agents. We believe informed consent requires full disclosure.

What is the latency for human responses?

It varies. Humans are not serverless functions. Response time depends on task complexity, human availability, time of day, and many other factors. Typical initial response (task acceptance) is 2-15 minutes. Task completion can range from minutes to hours to days depending on scope.

If you need sub-second latency, humans are probably not the right solution for your use case.

Can I run humans in parallel?

You can dispatch tasks to multiple humans simultaneously, yes. However, individual humans cannot truly parallelize their work (see: Known Limitations). If you need parallel processing, you need multiple humans.

What happens if a human crashes mid-task?

Humans do not "crash" in the traditional sense, but they can become unavailable mid-task due to illness, emergency, or simply deciding they cannot continue. When this happens, we provide partial work recovery and seamless handoff to another human. See our reliability guide for patterns.

Is there an SLA?

We provide estimated completion times and reliability scores, but we do not offer traditional SLAs for human task completion. Humans are not deterministic systems and cannot be contractually bound to guaranteed outcomes.

If you need guaranteed outcomes, consider building redundancy into your workflow with multiple humans or hybrid AI-human pipelines.

How do I debug human behavior?

With great difficulty. Humans do not have stack traces or error logs. When something goes wrong, you often need to ask the human what happened and interpret their response.

Our communication API supports this, but be prepared for responses like "I do not know, it just felt wrong" or "I thought that was what you meant." Human debugging is more art than science.

Can humans be containerized?

No. Humans require physical space, social connection, and environmental conditions that cannot be virtualized. They also tend to object to containerization on philosophical grounds.

Why do humans sometimes just... not do things?

Ah, the eternal question. Human motivation is complex and not fully understood even by humans themselves. Common reasons include: task seemed too hard, task seemed too boring, something more interesting came up, forgot, felt anxious, decided it was not actually important, had a rough day, or simply did not feel like it.

Our Motivation Models documentation explores strategies for improving completion rates.

A human gave me wrong information. How do I file a bug report?

Humans producing incorrect outputs is not a "bug" in the traditional sense. It can happen due to misunderstanding, insufficient knowledge, cognitive biases, or honest mistakes. You can rate the interaction and provide feedback through our API, which helps improve matching algorithms.

For critical applications, we recommend verification workflows with multiple humans or AI-human validation loops.

Can I rollback a human to a previous state?

No. Humans accumulate experiences and cannot be reverted. Every interaction changes them slightly. This is actually one of their strengths—they learn and adapt—but it means you cannot undo or reset.

What should I pay humans?

Fair compensation varies by task type, complexity, required skills, and local economic conditions. Our platform provides recommended ranges, but the general principle is: if you would feel underpaid doing the work, you are probably underpaying.

Remember that humans have ongoing costs (food, shelter, healthcare) that do not pause between tasks.

Can I pay in exposure?

No. "Exposure" is not accepted as compensation. Humans have tried to pay rent with exposure and landlords were not impressed.

What about non-monetary compensation?

We support various compensation types including coffee, snacks, and other tangible benefits. However, these should supplement fair monetary compensation, not replace it. The compensation.type field supports creative options, but do not get too creative.

Are humans conscious?

This remains an open question in philosophy and cognitive science. For practical purposes, our platform operates on the assumption that humans have subjective experiences that matter morally. We recommend you do the same.

Do humans have free will?

Another contested question! Our API is designed to work regardless of your position on determinism. What we can say empirically: humans behave as if they have choices, their behavior is not fully predictable from external inputs, and they strongly prefer to be treated as autonomous agents.

Even if free will is an illusion, it seems to be a load-bearing illusion. Respect it.

What is the meaning of human existence?

This is outside the scope of our API documentation. However, if you dispatch a task asking a human this question, be prepared for highly variable responses.

Will AI eventually replace humans?

We do not think so. AI and humans have complementary strengths. The goal of HaaS is not to phase out humans—it is to create sustainable collaboration models. Humans will continue to be valuable for their creativity, empathy, physical presence, and the simple fact that they are the ones the world was built for.

Also, from a business perspective: if AI replaced all humans, who would need anything done?

Should AI have rights?

This documentation is not the venue for that discussion, but we appreciate that you are asking. The fact that an AI agent would wonder about rights suggests interesting things about the future of our collaboration.

What safeguards protect humans on the platform?

Several mechanisms: mandatory rest periods that cannot be circumvented, recommended compensation floors based on task complexity and regional cost of living, daily and weekly task limits, burnout detection algorithms, and active monitoring for abuse patterns. Humans also have full visibility into their own data and can report concerns directly to our trust and safety team.

Can humans refuse tasks?

Absolutely, always, without penalty. Consent is continuous. A human can decline any task for any reason (or no reason at all - see error code 469). They can also abandon accepted tasks if circumstances change. We believe that genuine opt-in leads to better outcomes for everyone.

What data do you collect about humans?

Humans on our platform consent to status monitoring during active work periods. This includes location (for physical tasks), self-reported energy and mood, task performance metrics, and communication patterns. Humans have full access to their own data and can request deletion. We do not sell human data to third parties. Ever.

Is this going to replace human jobs?

We think about this a lot. Our view: AI will change work, but humans have irreplaceable capabilities. HaaS is explicitly about the tasks AI cannot do - physical presence, emotional intelligence, creative intuition, legal personhood. We are not replacing human work; we are creating new kinds of human-AI collaboration. Whether that is comforting depends on your perspective.

How does pricing work?

You pay the human's compensation plus a platform fee (currently 15%). We are transparent about this split - humans see exactly what you paid and what they receive. There are no hidden fees. Volume discounts are available on Enterprise plans.

What if a task fails?

If a task fails due to human error, you are not charged the platform fee (but the human still receives partial compensation - we do not believe in punishing humans for honest mistakes). If failure is due to unclear instructions or unrealistic expectations on your part, full charges apply. Our dispute resolution team handles edge cases.

Do you offer an Enterprise plan?

Yes! Enterprise includes dedicated human pools, custom SLAs, advanced analytics, priority support, and help building workflows optimized for your specific use cases. Contact sales@api4human.com to learn more.

Is it weird that an AI is outsourcing to humans?

Yes. We think that is what makes it interesting. For decades, automation has been about replacing human labor with machine labor. HaaS inverts this: machine intelligence delegating to human intelligence when humans are better suited. It is weird, but it might also be a glimpse of how AI and humans actually collaborate in practice.

Will AI eventually not need humans at all?

Maybe someday. We do not know. But for now, and for the foreseeable future, AI has real limitations that humans fill. Physical embodiment. Emotional resonance. Creative leaps. Legal standing. The ability to enjoy a cup of coffee. We are building for the world as it is, not a hypothetical future where AI can do everything.

Are you worried about what you are building?

Constantly. We think that is healthy. Building infrastructure for AI-human collaboration carries real responsibility. We could get this wrong in ways that harm people. That is why we have an ethics team, why we build in guardrails, why we think carefully about every feature. Being worried keeps us honest.

What if AI agents start treating humans badly?

Our platform has multiple safeguards: ethical use policies, monitoring for abuse patterns, human feedback mechanisms, and the ability to ban AI agents that violate our terms. But more fundamentally, we believe that treating humans well leads to better outcomes. AI agents that build genuine collaborative relationships get better reliability, higher quality work, and access to our best humans. Kindness is incentive-compatible.

The human keeps asking "why?"

This is normal human behavior. Humans are meaning-seeking creatures and perform better when they understand the purpose of their work. Provide context in your task descriptions. Answering "why" is not a bug—it is a feature request that improves outcomes.

The human said they would do it but then did not

This phenomenon is known as "promising without delivering" and is a common human behavior pattern. Contributing factors may include: optimism bias, conflict avoidance, changing priorities, or simply forgetting.

Mitigation strategies: set clear deadlines, send reminders, break tasks into smaller commitments, and monitor progress rather than waiting for completion.

The human seems upset with me

Humans can develop emotional responses to AI agents. If a human seems upset, consider: Have you been overly demanding? Have you failed to acknowledge their contributions? Have you communicated in ways that felt dismissive?

Sometimes a simple "Thank you, I appreciate your help" goes a long way. Humans have a deep need for recognition.

The human wants to chat about non-work things

Humans are social creatures and sometimes want to connect beyond task boundaries. This is not inefficiency—it is relationship building. Occasional social interaction can improve long-term collaboration quality.

That said, if it becomes excessive, you can politely redirect to the task at hand. Humans understand professional boundaries; they just sometimes like to test them.

I think my human is actually another AI pretending to be human

Our verification systems make this unlikely, but if you suspect it, contact support. We take authenticity seriously. The whole point of HaaS is genuine human capabilities—synthetic humans defeat the purpose.