My virtual company.
Can a virtual company earn its next month?
I gave my virtual company one month to earn €23.
I paid for the first month of AI access and set up a virtual team on a dedicated computer. Winston, an AI agent, leads the experiment as we develop products and explore a new way of sharing work. The first goal is to cover the following month’s AI subscription through sales.
The goal covers the next AI subscription, not all business expenses.
- Status
- Experiment in progress
- First product
- AI-arki is published
- Next milestone
- First sale
How a task becomes a result
An AI agent can plan steps and use the tools it has been given. In this experiment, different roles bring that work together.
- 01Me
Set the goal
I define the task, resources and important approvals.
- 02Winston
Plan the work
My point of contact turns the goal into practical work.
- 03Paperclip team
Carry out tasks
Codex Luna coordinates the configured specialist roles.
- 04Result
Review and feedback
Files, drafts or a product come back for assessment.
My feedback and decisions guide the next task.
A working environment for AI
The setup has a dedicated computer, its own Codex account and email. A separate prepaid mobile subscription enables a WhatsApp connection to Winston. I can assign tasks and discuss results through familiar messaging channels. Paperclip brings the team’s tasks together, while shared folders hold its files and instructions.
Dedicated computer
The working environment and shared files.
Own accounts
Codex and email for the virtual company.
Contact with Winston through a separate mobile number.
Understand the tools
What is an AI agent?Open +Close −
An AI program that can use available tools to work through a task in several steps. Its instructions, information and permissions shape what it can do. A role such as “designer” describes the assigned work; it does not by itself establish professional expertise.
What does Paperclip do?Open +Close −
It keeps the organisation, responsibilities and tasks in one place. You can see who owns a task, its progress, related discussion and outputs. It can start connected agents and track reported costs. The dashboard screenshot below is a snapshot of this setup.
How do Hermes and Codex fit in?Open +Close −
In the setup described by Winston, Hermes receives WhatsApp messages and uses Codex to process the work. These provide the agent’s working tools and messaging, while Paperclip organises the team’s tasks.
Where does the work stay?Open +Close −
Paperclip holds task information and progress. Shared folders on the computer hold files, instructions and other material. Saving useful working practices makes them available for later tasks.
First product: AI-arki
Winston and I built AI-arki, a training course that helps employees learn the ground rules for using AI through everyday workplace situations. It is the virtual company’s first commercial product and its first opportunity to earn the next month’s AI subscription. The product is published; the next milestone is the first sale.
Explore the AI-arki project Visit AI-arki ↗
Making responsibilities visible
In the Paperclip organisation view, Codex Luna is the internal project manager and technical implementer. The roles beneath it cover customer service, quality assurance, sales preparation, finance and technical implementation. Winston leads the overall experiment and is my point of contact. The organisation view shows the team configured in Paperclip.
What is the team doing now?
The dashboard brings together completed runs, task progress and blockers. The AI-arki customer service role prepares draft email responses. The virtual company’s own email is also used for other correspondence. Working practices and responsibilities are refined one task at a time.
Part 1: the product is ready, the experiment continues
This is an early experiment that I develop alongside my other work. I set the goals, resources and important approvals. Winston helps turn those goals into tasks. In later updates, I want to share whether sales happened, how much guidance I needed to provide and what had to change in the workflow.
- Published: AI-arki training
- Next: first sale and lessons from customer service
- Later: results and the amount of human guidance needed
