Autonomous Economic Agent (AEA)

An autonomous economic agent (AEA) is a piece of software that acts on behalf of an owner, such as a person, company or device, to find opportunities, negotiate and complete economic transactions without needing human approval at each step. An AEA might sell spare electricity from a home battery, book parking for a car, buy data from a sensor network or trade tokens, following goals and limits set by its owner.

The term is most closely associated with Fetch.ai, a company founded in Cambridge, UK, in 2017, which built an open-source AEA framework and a blockchain network designed for agents to discover and pay one another. Fetch.ai later merged its token into the Artificial Superintelligence Alliance. AEAs are an early form of the AI agents now discussed in agentic payments and agent-to-agent transactions.

Imagine a digital assistant that does deals for you while you sleep. You tell it: "Sell my unused solar power to neighbours when the price is above 20 cents, and buy cheap power at night to charge my car." The assistant then goes off, finds buyers and sellers, haggles, and settles each deal, sending you a summary in the morning.

That assistant is an autonomous economic agent. It needs a way to find other agents, a shared language to bargain in, a way to pay and get paid, and clear rules about what it may and may not do.

Fetch.ai's framework, released as open-source software, treats each agent as a collection of components:

• Skills. Modules that contain the agent's behaviour, such as negotiating a price or responding to a request.
• Protocols. Agreed message formats for conversations between agents, such as "call for proposal", "propose", "accept".
• Connections. Links to the outside world, including other agents, websites, databases and blockchains.
• Wallets and keys. Credentials that let the agent sign transactions and hold tokens, similar to an agent wallet.
• Decision maker. A component that checks whether a proposed transaction fits the owner's preferences before signing.

Agents register their services so others can find them, then use the agreed protocols to negotiate and settle, often through smart contracts or token transfers on Fetch.ai's network, which was built with the Cosmos SDK, the toolkit behind Cosmos.

• Energy markets. Agents for batteries, solar panels and chargers buy and sell power in local markets.
• Mobility. Agents for vehicles book parking or charging and pay automatically.
• Data marketplaces. Sensors and data owners sell data streams to buyers' agents.
• Trading. Agents monitor prices and execute trades on exchanges, including a decentralized exchange.
• Supply chains and travel. Agents compare offers and book services across providers.

These uses link AEAs to the internet of things, where connected devices could hold small budgets and pay for services.

AEAs predate the wave of large language model agents that began in 2023. Early AEAs relied mainly on rule-based logic and economic models, with the language of negotiation drawn from game theory and multi-agent systems research. Fetch.ai later released lighter tools for building agents and a hosted platform for running them, and connected agents to language models so they could understand natural-language requests. The broader industry has since developed its own agent communication and payment protocols, and the AEA concept is one of several approaches competing to define how agents trade.

Letting software transact on its own raises practical and legal questions. Agents can make mistakes at machine speed, be exploited by malicious counterparties or be manipulated through false data. If an agent's decision causes a loss, it is unclear whether the owner, the developer or the platform is responsible. Agents that pay one another still involve real people and companies behind them, so anti-money laundering and sanctions rules apply. Adoption of AEAs in real markets has so far been limited to pilots and niche uses, and the economic value of agent networks remains largely unproven at scale.

• Fetch.ai