The new AI agent economy will run on Smart Accounts.

Web3’s first billion users may not only be humans, but AI agents.

Data from September and October 2023 shows that more than half of all Safe transactions on Gnosis Chain are currently made by AI agents. This signals a nascent but growing "agent economy"—an onchain economy run solely by AI agents that is turning the crypto-AI dream team into a reality.

Onchain autonomous services

What are autonomous services? Autonomous services come in many forms, and increasingly, rely on and interact with each other to create their own self-sustaining and self-referencing economies. In web3, some of the autonomous services that are in demand include:

• Automation

• Relayers

• Oracles

• Messaging

Autonomous services are essential for enhancing the web3 user experience, enabling features like account abstraction. While these advantages are poised to attract new users to the space, there is one type of autonomous service that is inevitable: AI agents.

Behind web3’s ultimate DAUs: AI Agents

Already today, over 4% of all Safe transactions on Gnosis Chain originate from Olas network’s autonomous services, with the majority coming from AI agents.

Autonomous services built using Olas’ open-source framework have distinct, crypto-native capabilities. They are based on advanced autonomous agent technology, co-ownable via a Safe. Each service is made up of one or more software agents, each operated by an agent operator. While these agents run off-chain and can retrieve data from anywhere, they are represented onchain by a Safe Smart Account.

Through a Safe, the service can, among other things, hold assets, pay for transactions, and interact with onchain apps. Safe’s multisig functionality helps multiple agents to co-own the autonomous service across multiple chains, including Ethereum, Polygon, and Gnosis Chain.

Leveraging and safeguarding AI agents with Smart Accounts

Smart Accounts are critical infrastructure for the agent economy, as agents will not be using EOAs or externally owned accounts. Instead, it will be Safe’s account programmability and customisable logic that will allow the infinite offchain activity of AI agents to be fully leveraged in the onchain economy.

Furthermore, Safe’s Smart Account infrastructure allows specific onchain rules and actions to be defined on a granular level, and are absolutely necessary for software agents to reach their full web3 potential, be it for basic autonomous services or entire AI agents.

Multisig configurability: Decentralizing Autonomous Services

Safe’s multisig feature allows for accounts to be co-owned by autonomous services, with the threshold of signers programmable to any desired configuration. In a Safe multisig account controlled by an autonomous service of multiple agent signers, fault tolerance and decentralization among agents can be configurable according to the service requirements.However, such security may not always be required. For example, prediction agents, which are intended to be managed by individual users, employ a basic 1-of-1 multisig. This also enables the progressive decentralization of autonomous services, as developers can build an autonomous service with just one agent, then scale it to multiple agents when they’re ready.

Transaction Guards: Protecting outcomes of AI agents

Transaction Guards are another feature of Safe’s Smart Accounts that are critical for agents. Transaction Guards are able to make checks before and after a Safe transaction to increase the security of autonomous services even further. For example, even if agents are executing malicious code, they can be restricted to performing only specific transactions.

In AI agent powered funds, for instance, guards could restrict the actions of agents so that they could only alter the fund’s allocations, thus protecting the assets in the fund through onchain means.

Crypto-native AI applications with customisable fault-tolerance, decentralization, and reliability are now possible to bootstrap any use case in a diverse and secure agent economy.

AI x Web3 use cases

What are some interesting use cases emerging at the intersection of AI x Web3?

Autonomous trading

Just for Safe users, the team from CoW Swap introduced Composable CoW, a conditional smart order framework. Composable CoW enables the deployment of programmatic orders such as autonomous trading bots, via a Safe, without having to deal with order placement or execution—just focusing on the programmatic code of the order.

What’s more, these conditional smart orders can be executed unlimitedly and in parallel, with each of the specific order types summarised into a single merkle root to achieve O(I) gas fee consumption. Some possibilities include fee collection, DAO payroll, TWAP orders, Good after time (GAT) orders, stop loss orders, and more.

The Composable CoW smart order framework for Safe users.

Prediction markets

AI agents are also transforming prediction markets. For example, on the Omen platform, three types of agents: the Market Creator Agent, the AI Mech, and the Trader Agent work together to create, analyse and trade markets. What arises is a network of agents allowing for AI-based prediction markets that can inform market research, political forecasting, trading and much more.

Impact evaluation

Olas Contribute, serving as the Olas DAO’s coordination tool, employs autonomous services to AI-ify the DAO’s operations. Just one feature is its autonomous, AI-ified contribution evaluation of community contribution points.

By utilising agents to extract data from Twitter/X and leveraging the capabilities of Large Language Models (LLMs), the tool finds and evaluates the impact of community tweets, allocating points as rewards to the tweet's owner continuously, 24/7.

AI-Powered DAO delegation

Valory’s Governatooorr, is a lighthearted, experimental autonomous service acting as an AI governance delegate in over 14,000 DAOs (via compatibility with Snapshot and Governor Bravo) to which users can simply delegate their tokens.

Leveraging LLMs, it can actively watch for and then analyze DAO proposals, apply a user’s voting preferences and vote, potentially AI-ifying DAO governance for web 3’s biggest DAOs.

Safe DAATA and AI Hackathon Winners

Safe recently ran a DAATA and AI Hackathon at the intersection of data analytics, Artificial Intelligence, and account abstraction in collaboration with partners The Graph, Gnosis Chain, Covalent, Olas, and BuidlGuidl. Here are some winners who impressed the team with their AI hacks.

Unblock.AI Integrating the Safe AA SDK, the team from Polywrap created a chat interface that makes it simple and easy for folks to use Safe’s functionalities. These include checking balances, creating and signing transactions, adding and removing signers, and more—all easily accessible through a chat interface.Try it in action:

Safe Guardian Angels This wallet tool uses a group of AI agents to monitor wallets and provide active defence against potential threats to protect user funds and improve wallet security. Features include the automatic revoking of contract permissions and the withdrawing of funds through monitoring the Internet for anomalies and known hacks.

Watch a recap of the presentations!

The AI agent economy is coming. And it will run on Smart Accounts.

Stay in the loop with all our latest insights and updates! Click Subscribe and never miss a beat from our team.

• Download Safe{Wallet}

• Build on Safe{Core}

• Explore the Safe Ecosystem