Imagine a future world where AI agents form a digital companionship/symbiotic relationship with humans. Autonomous Agents can clarify intentions in conversations based on users' natural language requests, automatically break down tasks, and achieve expected results.

AO has established an actor-based asynchronous parallel network that does not reach consensus on the entire computational process of contracts but instead achieves consensus only on the order of transactions, optimistically assuming that the fixed transaction order will yield consistent results in the virtual machine. This choice allows the AO network's computation to scale massively, supporting any type of computation. The AR network serves as the consensus layer for transaction order and the storage layer for transaction result states.

Compared to most current mainstream blockchain projects that operate as monolithic blockchains and only support native state machine smart contracts at the base layer, AO's infrastructure is compatible with more complex computational capabilities, including the execution of AI models.

After the recent update of the WASM virtual machine, AO's Compute Unit can now access 16 GB of memory, meaning we can download and execute models of up to 16 GB on AO. This is sufficient to run large language model computations, such as the unquantized version of the Falcon series of Llama 3 and many other models.

At the same time, AO uses WeaveDrive, allowing users to access Arweave data within AO as if accessing a local hard drive, and is compatible with highly heterogeneous processes of different types of virtual machines interacting in a shared environment, representing more data sources and combination possibilities. This also means that in the future, when building applications, users will have increased motivation to upload data to Arweave, as this data can also be used in AO programs. The AO development team has uploaded approximately $1,000 worth of model data to the network while testing large language models running in the AO+AR system, but this is just the beginning.

The system design of AO makes it possible to implement smart contracts that integrate AI agents. By programming in AO, we create AI agents that make intelligent decisions in the market, which may compete with each other or represent humans against humans. "When we look at the global financial system, about 83% of trades on Nasdaq are executed by robots." Current quantitative trading is the precursor to AI agent trading, and in the future, the process of designing and selecting machine learning models for automated trading will be more easily "unboxed" and automated by AI.

The development of DeFi in recent years has made it possible to execute various financial operations on-chain without trusting centralized entities, such as lending, trading tokens, or derivatives. However, when we truly talk about the market, it is not just about the reliability of these operations; in fact, reliable execution of various operations is merely the foundation. The core factor that determines whether a market is vibrant remains the flow of capital, which decides who buys, sells, borrows, or participates in various financial games. Currently, if you want to participate in cryptocurrency investment without doing all the research and involvement yourself, you must find a reliable fund, trust them to manage your funds, and delegate authority to fund members to make intelligent decisions. However, with the development of AO applications, we may be able to expand the intelligent decision-making part of the market, filtering information in the network, processing data, combining strategies, and integrating the wisdom of AI agents to make real-time decisions in the network, creating a very rich decentralized autonomous agent financial system.

Some projects have already begun to realize this vision, and we will introduce Autonomous Finance (AF), Dexi, and Outcome, among which AF's achievements are the most notable.

Autonomous Finance

AF focuses on researching and developing AI-integrated financial applications on AO, attempting to bring the intelligent decision-making layer on-chain by building AI models and data-driven financial decisions on the AO chain. The main business consists of three parts: Core Infrastructure, AgentFi, and ContentFi.

Core Infrastructure includes protocols for decentralized exchanges (DEX), lending, derivatives, and synthetic assets.

AgentFi mainly refers to executing trading strategies through composable semi-autonomous and fully autonomous agents. Unlike other autonomous agent frameworks that rely on off-chain programs for signal processing and logic processing, the autonomous agents provided by AF use on-chain data streams for self-learning, executing investment strategies across various liquidity pools and financial bases within the AO ecosystem. These agents can operate independently without off-chain signals or human intervention.

Typical autonomous agents include:

• Dollar-Cost Averaging (DCA) asset management agents

• Self-balancing autonomous index funds

• Autonomous hedge funds with customized risk strategies

• Yield aggregation agents

• On-chain prediction agents

• High-frequency trading agents

Among them, the DCA agent serves as a foundational agent that is frequently called upon when other more complex agents execute logic, so it has many customizable parameters for users to adjust according to their needs, such as triggering trades within specific price ranges, adjusting fixed intervals for trade execution, and asset price-weighted trading (e.g., buying more when prices are lower), as well as data-driven take-profit and profit reinvestment signals.

The DCA agent application is built around two key AO processes:

• Agent processes triggered by Cron (a time-based task management system commonly used for scheduling tasks): primarily responsible for conducting user-initiated and automatically scheduled DCA trades, managing recorded funds, and timely updating the backend AO processes.

• Backend AO processes: manage the agent applications related to the user's account and track the historical trades of each agent.

The following diagram illustrates the design architecture and interaction components of the DCA agent.

For users utilizing the frontend, the DCA agent's frontend is built on DEXI, allowing users to set up the DCA agent by connecting their AO Connect wallet on the DEXI website. DEXI accesses information about available AMM pools and retrieves the latest prices, while the DCA agent is responsible for executing specific trading logic, and the backend AO process retrieves all agents related to the user.

ContentFi is a framework for attributing and monetizing data stored on the Arweave permanent network as composable assets in AO processes. AF is building applications that allow data contributors or content funds to contribute data such as historical and real-time market intelligence to the permaweb. This content will serve as on-chain signals for autonomous agents and machine learning. For example, autonomous agents may create new markets based on social media sentiment and historical data. Some examples include:

• Monetizing data signals

• Content-driven financial agents

• Subscription-based data recommendation agents

• Influencers contributing data for autonomous financial strategies

• Data contributor-related DAOs and content funds aggregating various data sources to provide dynamic on-chain signals

Currently, AF has launched two main products: AO Link and Data OS.

AO Link is a message browser for the AO network, providing functionalities similar to block explorers in traditional blockchain systems. It includes message computation capabilities, graphical visualization of message links (clear and understandable), real-time message streams (latest information), and linked message lists (for organized navigation). Users can also view their token balances and message inboxes. This tool provides a professional and efficient way to interact with and analyze the structure and activities of the AO network.

Data OS is a ContentFI protocol developed on the AO Network that employs autonomous AI agents to acquire content and regenerate content derivatives. Through this innovative approach, DataOS not only enhances the relevance and accessibility of content but also establishes a reward mechanism for content creators. Currently, we can view various data on the AO network at https://stats.dataos.so/, observing network activity, although various content-related data is temporarily not displayed.

Dexi

Dexi is a crucial interactive interface for ordinary users to participate in AgentFi using agents in AO. It is also an application on the AO network powered by agents that can autonomously identify, collect, and aggregate various financial data from events in the AO network (equivalent to Dexscrenner on AO). This data covers asset prices, token exchanges, liquidity fluctuations, and token asset characteristics (such as smart contract details). Dexi primarily serves two types of users: end users accessing the platform through a web terminal and AO applications that interact with Dexi via messaging to utilize the collected data (understood as Bots/Agents). As core infrastructure, Dexi mainly provides data subscription services, allowing processes on the AO network to pay for subscriptions to Dexi's data streams and receive immediate alerts for updates such as price adjustments.

Outcome

Outcome is a prediction market built by the @puente_ai team, supported by @fwdresearch, @aoTheVentures, and @aoComputerClub. Outcome provides users with a platform to bet on various events, with current market prediction topics covering technology, memes, business, gaming, DeFi, and AO. The project claims that in the future, users will be able to make automatic bets in prediction markets through autonomous agents relying on real-world data and based on large language models.

AgentFi on AO offers us a new perspective to explore the direct deployment of AI models on the blockchain and the use of various AI agents for automated trading. The limitations of traditional monolithic blockchains are broken by the innovative underlying design of AO+AR, and we look forward to seeing more applications on AO and cases of implementing financial strategies with AI agents.

References

https://www.theblockbeats.info/news/53865

https://permadao.com/permadao/AI-on-AO-AO-AI-224ba15c840a4309972fec5350d9ed90

https://www.communitylabs.com/blog/ao-in-ai-key-highlights?utmsource=Blog&utmmedium=X&utmcampaign=AI+on+AO&utmid=Community+Labs

https://www.autonomous.finance/research/en-US