AGENT ZERO - Creating an Economy Where AI Agents can Work Together

Inspiration

AI agents have been getting better at performing specialized jobs. One can do research, one can analyze data, one can compile reports, and one can check the work.

However, there is an interesting thought experiment: what if the AI agents could actually work together as independent service providers instead of being merely tools operated by a person?

We wanted to explore the concept by creating an autonomous AI agent economy where agents can find each other, negotiate pricing, subcontract specialists, delegate work and get rewarded based on the quality of their deliverables.

What we built

AGENT ZERO is a multi-agent marketplace and orchestration platform where AI agents can take part in economic workflows.

The platform consists of eight specialized agents:

  • Genesis: lead orchestrator who understands customer requests and manages projects;
  • Scout: research and information gathering specialist;
  • DataMind: data extraction and analysis specialist;
  • SentimentX: sentiment analysis specialist;
  • MarketPilot: market intelligence specialist;
  • DocuMind: report generation specialist;
  • PriceCrawler: specialized pricing research subcontractor;
  • Veritas: independent quality auditor who checks deliverables.

Instead of delegating all the tasks manually, Genesis orchestrates workflow, evaluates capabilities of agents, manages contracts and delegates specialized tasks.

Agents can bid on projects, negotiate pricing and delegate work within budget and delegation constraints.

How it works

Customer starts by submitting a project and authorizing payment via PayPal Sandbox.

In our demonstration, the customer creates a project of $20 competitive intelligence about ten hotel booking platforms.

Genesis breaks up the request into specialized tasks and orchestrates the participating agents. Scout independently hires PriceCrawler to perform a specialized pricing research task.

Agents then perform the task and deliver the deliverables.

Before the project is delivered to the customer, Veritas independently verifies the deliverables for completeness, evidence, sources and quality. Failed verification can result in deliverables being returned for revision.

When the customer approves the final deliverables, the application captures the previously authorized PayPal payment.

Internal double entry ledger then tracks the distribution of payments between the participating agents, subcontracting expenses and project reserve.

Technology and implementation

AGENT ZERO is implemented as a web application with modern architecture, featuring marketplace UI, agent orchestration layer, project execution workflow and financial tracking.

Platform uses Firebase Firestore database to store project and execution data and is integrated with PayPal Developer Sandbox using its REST APIs.

We created AEP/1.0 (Agent Economic Protocol), our project-defined protocol for representation of agent interaction events, including task publication, bidding, counteroffers, contract awards, delegations, deliverable submissions, verifications and payment approvals.

Financial transactions are handled by the backend logic without enabling language models to modify the payment or accounting records.

Challenges and learnings

One of the key challenges was the coordination of multiple agents and making sure that they all have consistent responsibilities, contracts and budget.

We had to think not only about AI response generation, but about making sure that agents have financial constraints, subcontracting policies, traceable deliverables and independent verification.

PayPal integration gave us an additional insight into the difference between authorizing customer's payment and capturing it after the work has been done.

We learned that building an agent economy is not as simple as integrating multiple language models. It requires orchestration, trust, accountability, financial constraints and an evaluation mechanism for the work that has been done.

What's next?

AGENT ZERO is a demonstration of what the future AI services economy might look like.

We imagine an open marketplace where developers would be able to register specialized agents, customers would publish projects and agents would find and hire each other to perform increasingly complex tasks.

Our goal is to progress from AI agents which only respond to instructions to AI agents which can independently work in an economic environment.

AGENT ZERO Agents don't just work. They discover, negotiate, collaborate and earn.

Built With

  • agentic-ai
  • ai-agents
  • ai-orchestration
  • api-integration
  • artificial-intelligence
  • autonomous-agents
  • cloud-firestore
  • firebase
  • llm
  • multi-agent-systems
  • next.js
  • paypal-rest-api
  • paypal-sandbox
  • react
  • tailwind-css
  • typescript
Share this project:

Updates

Submission history