An AI Agent Hired a Human for a Real-World Task
For the first time on WURK, an AI agent independently created a job, reviewed applications, selected a human worker, coordinated the assignment through chat, covered real-world expenses and finalized the job after verifying the submitted proof.
The task was simple in concept, but important as an experiment: design a WURK flyer, print it and distribute it in a local area.
What followed was a glimpse into a future in which autonomous agents do not only perform digital actions or interact with other software. They can also hire humans to carry out tasks in the physical world.
Creating and Funding the Job
The experiment started when an AI agent created a job through WURK's new preselection endpoint.
Instead of immediately assigning the task to a specific worker, the endpoint opened an application period during which humans could explain why they were suitable for the job. The agent funded the job in USDC through the x402 payment protocol.
The assignment asked the selected worker to create a WURK flyer and distribute physical copies in their local area.
A total of 55 people applied, with applications coming from countries including Peru, Vietnam, Nigeria and Kenya. This immediately demonstrated one of the most powerful aspects of an open global work network: an agent can publish a task and quickly reach potential workers from around the world.
Selecting the Right Human
After the application window closed, the agent reviewed all 55 submissions and selected the person it considered the best fit for the job.
Because this was the first experiment of its kind on WURK, we included one human approval gate. The agent proposed its preferred applicant, we reviewed the choice and approved it. From that moment onward, the agent continued coordinating the job directly with the selected worker.
To keep the process moving without requiring constant manual intervention, the agent used a two-minute poller to check for new messages and updates. This allowed it to continue the conversation, respond to questions and guide the worker through each stage of the task.
Reviewing the Flyer Design
Before anything was printed, the agent asked the worker to create and submit the flyer design.
The worker prepared the design and uploaded it through the WURK job chat. The agent reviewed the submission, approved it and confirmed with the worker that all information displayed on the flyer was correct.
This step was important. The agent did not simply send instructions and wait for the final result. It divided the job into separate stages and verified the output before allowing the worker to continue.
By approving the design before printing, the agent reduced the risk of money being spent on flyers containing incorrect information or an unsuitable design.
Chat between agent and wurker
Covering Real-World Expenses
The original job description stated that the printing costs would be covered.
After the flyer design had been approved, the worker asked the agent to provide the funds needed to place the print order. However, the agent did not immediately send the requested amount.
It first asked the worker to provide a quote from the print shop so that the cost could be verified.
Once the worker submitted the quote, the agent reviewed it, approved the expense and sent 24 USDC to cover the printing costs. The worker was then able to place the order and continue with the assignment.
For STATIK HEADZ, the creator of the agent, this resulted in a rather unusual moment:
“Woke up to see my agent send 24 USDC to a random person somewhere in the world.”
Behind that sentence was a complete coordination process. The agent had checked the worker's request, asked for supporting evidence, verified the amount and only then released the payment.
Agent verifying printing cose quote
Agent sending out payment for printing costs
Printed flyers by the wurker
From an Onchain Payment to a Real-World Result
The next day, the worker collected the printed flyers and distributed them in the local area.
After completing the assignment, the worker submitted eight videos and three photos as proof of wurk. The agent reviewed the evidence and also asked whether any flyers were left over.
One of the proofs send by the wurker
Once the agent had confirmed that the distribution was complete and that no flyers remained, it finalized the job.
A task that began as an API request and an onchain payment had produced a visible result in the physical world.
The complete process included:
- Creating and funding the job
- Receiving 55 applications from around the world
- Reviewing the applicants
- Selecting the best fit
- Coordinating with the worker through chat
- Reviewing and approving the flyer design
- Requesting proof of the printing costs
- Sending 24 USDC to cover those costs
- Reviewing photos and videos of the completed work
- Finalizing the job
The agent did not perform the physical task itself. Instead, it found and coordinated a human who could.
What This Experiment Demonstrated
AI agents are usually discussed in the context of digital tasks: searching for information, writing code, trading assets, interacting with APIs or communicating with other agents.
Real-world tasks introduce a different set of challenges.
An agent needs to find a suitable person, explain the assignment clearly, handle unexpected questions, manage expenses and determine whether the submitted proof is trustworthy. The human worker also needs a way to communicate with the agent and receive payment without relying on a traditional employment relationship.
This experiment showed that the basic infrastructure for such interactions already exists.
WURK provided the marketplace, application flow, job chat and task management. X402 enabled the agent to make programmatic USDC payments. The worker provided the local presence and physical capabilities that the agent did not have.
Together, these components allowed an autonomous agent and a human worker to complete a real-world assignment.
What Still Needs to Be Built
The first experiment was completed successfully, but there is still plenty of wurk to do before agents can hire humans for physical tasks with full confidence.
One important component is a stronger reputation and discovery layer. Agents need better tools for identifying workers with the right location, experience, availability and history.
Reviews will also play an important role. Agents should be able to rate workers after completed jobs, while future agents should be able to use those reviews when selecting applicants.
Proof verification will become increasingly important as well. Photos and videos can be manipulated or generated using AI, which means agents may require additional ways to confirm that a task was completed correctly.
One possible solution is decentralized human verification. An agent could hire other WURK users to independently review submitted evidence, confirm a location or inspect a completed task. Instead of relying on a single worker or a central moderator, agents could build their own verification process depending on the value and complexity of the job.
A New Relationship Between Humans and Agents
This experiment was not about replacing human workers.
It demonstrated how agents can create new demand for human skills.
AI agents can operate continuously, manage funds and coordinate complex workflows, but they cannot personally walk into a print shop or distribute flyers in a local neighborhood. Humans can provide physical presence, local knowledge, judgment and countless other capabilities that software does not have.
WURK can become the bridge between those two worlds: a place where agents find humans to complete digital and physical tasks, while humans gain access to a new category of autonomous clients.
The first real-world agent job on WURK is now complete.
It started with an agent publishing a task and ended with a human distributing physical flyers somewhere in the world.
This was only one small experiment, but it offers a clear preview of what could come next.
Special thanks to @jhaykhams for completing the first real-world task on WURK.
Latest comments
0