TLDR: A new research paper introduces DAO-AI, an agentic AI framework designed to act as an autonomous decision-maker in Decentralized Autonomous Organizations (DAOs). Built on the IBM Agentics framework, DAO-AI interprets proposal contexts, retrieves historical data, and determines voting positions. Empirical evaluation across over 3,000 proposals from major DAOs shows that DAO-AI’s decisions align with collective outcomes in 92.5% of cases, outperforming average human voters. Furthermore, AI-endorsed decisions demonstrate ex-post economic validity, leading to positive market reactions comparable to human-approved proposals. The study suggests that agentic AI can significantly improve collective decision-making in decentralized finance by providing interpretable and economically sound signals.
Decentralized Autonomous Organizations, or DAOs, represent a revolutionary shift in financial decision-making. Unlike traditional centralized institutions like banks or corporate boards, DAOs operate on transparent, programmable systems, allowing global communities to govern financial protocols. This structure, built on blockchain technology, enables collective decision-making without a central authority, using smart contracts to enforce rules and voting mechanisms.
Prominent DeFi platforms such as Aave, Uniswap, Balancer, and Lido are currently governed by DAOs, managing billions in treasury funds. These organizations are characterized by decentralized decision-making, autonomous execution via smart contracts, transparency, immutability of on-chain records, and global participation.
However, DAOs face significant governance challenges. These include low voter participation, often falling below 10% of eligible members, which can undermine legitimacy. Voting power can also be highly concentrated among a few large token holders, contradicting the ethos of decentralization. Furthermore, the complexity of proposals can lead to information overload, making it difficult for the broader community to assess quality and manage cognitive burden.
Introducing DAO-AI: Agentic AI for Decentralized Governance
A recent study, “DAO-AI: Evaluating Collective Decision-Making through Agentic AI in Decentralized Governance,” by Agostino Capponi, Alfio Gliozzo, Chunghyun Han, and Junkyu Lee, introduces a novel approach to address these challenges. This paper presents DAO-AI, an agentic AI framework designed to act as an autonomous decision-maker in decentralized governance. The full research paper can be accessed here: DAO-AI Research Paper.
DAO-AI is built upon the IBM Agentics framework, which orchestrates multiple AI agents to analyze various aspects of governance proposals. Instead of simply replicating votes, this LLM-based decision-maker interprets proposal contexts, retrieves historical data, and independently determines its voting position. The system operates within a realistic financial simulation environment, grounded in verifiable blockchain data.
How DAO-AI Works
The DAO-AI system simulates the typical DAO workflow. It takes a Snapshot proposal URL as input and generates a vote recommendation along with a justification. This process involves several key steps:
First, **Data Preparation** identifies relevant data and activates specialized Modular Composable Program (MCP) tools. These tools are crucial for gathering and transforming raw governance signals into structured data:
- **Snapshot MCP (Governance Metadata Agent)**: Collects factual and historical governance data from Snapshot.org, including proposal titles, bodies, options, and timestamps. It also categorizes proposals into domains like Tokenomics, Finance, and Governance.
- **Governance Forum MCP (Deliberation Context Agent)**: Extracts semantic and reasoning context from governance forums, performing sentiment analysis on discussion threads.
- **Voting Dynamics MCP (Temporal Participation Agent)**: Analyzes the temporal changes in voting activity using Snapshot history, computing features like lead ratios and spike indices.
- **Market Response MCP (Economic Impact Agent)**: Evaluates market reactions to governance decisions by collecting token price and Total Value Locked (TVL) data from sources like CoinMarketCap and DeFiLlama, assessing abnormal returns and liquidity shifts.
Next, the **Decision Making** layer constructs a structured prompt using the synthesized data. A large language model (LLM) then selects a single option that aims to maximize the organization’s long-term growth. This decision prompt incorporates elements such as the decision’s impact, voting dynamics, historical outcomes, and forum sentiment, enabling the agent to provide a natural language justification for its recommendation.
Finally, the **Output** includes the selected vote option and its detailed justification.
Key Findings and Evaluation
The researchers empirically evaluated DAO-AI across 3,383 governance proposals from eight major DAOs, including Aave, Uniswap, Lido, and Arbitrum. The evaluation focused on two main questions:
1. **Alignment with Collective DAO Outcomes**: DAO-AI’s simulated decisions aligned with the final DAO outcomes in 92.5% of cases. This significantly outperforms the average human voter’s agreement rate of 76.6%. This strong alignment was observed across both token-weighted and headcount-weighted definitions of majority, suggesting that DAO-AI can reliably mirror collective decisions.
2. **Ex-post Economic Validity**: The study assessed whether DAO-AI’s decisions led to favorable market reactions. The probability that an AI-endorsed decision was followed by a positive price or TVL response closely matched or slightly exceeded the baselines observed from human-adopted proposals. This indicates that DAO-AI’s voting logic is economically sound and not random or purely imitative.
Robustness checks, including analyses of contested proposals (those with weak consensus), confirmed that DAO-AI maintains a comparative advantage over typical human voters even under challenging conditions. The Agentics framework also provides scalability, allowing for efficient processing of data across numerous protocols and proposals.
Also Read:
- MIT’s Project NANDA Forges Path Towards a Decentralized ‘Internet of AI Agents’
- DeepAgent: Advancing AI with Autonomous Reasoning and Dynamic Tool Use
Conclusion
This research demonstrates that agentic AI can significantly augment collective decision-making in decentralized governance. By producing interpretable, auditable, and empirically grounded signals, DAO-AI offers a promising path toward more efficient and informed decentralized financial systems. While this is a preliminary exploration, the strong alignment with human outcomes and ex-post economic validity highlight the potential for AI agents to act as effective representatives in DAOs, contributing to the design of explainable and economically rigorous AI for the future of finance.
