DAO-AI: Evaluating Collective Decision-Making through Agentic AI in Decentralized Governance

TL;DR

This study presents DAO-AI, an agentic AI framework built on the Agentics platform to simulate autonomous voting in DAOs using 3,383 proposals across eight major DAOs. It integrates modular data tools (MCPs) to collect proposal metadata, forum sentiment, voting dynamics, and market responses, and uses structured transductions to produce a single vote option with justification. Empirically, DAO-AI achieves strong alignment with collective outcomes (AI agreement ≈ 92–93%) and demonstrates ex-post economic validity comparable to or better than human decisions, supporting its potential to augment decentralized governance. The work advances data-centric, explainable AI for DeFi, offering auditable signals and a scalable pathway for evaluating AI-driven governance in open digital economies.

Abstract

This paper presents a first empirical study of agentic AI as autonomous decision-makers in decentralized governance. Using more than 3K proposals from major protocols, we build an agentic AI voter that interprets proposal contexts, retrieves historical deliberation data, and independently determines its voting position. The agent operates within a realistic financial simulation environment grounded in verifiable blockchain data, implemented through a modular composable program (MCP) workflow that defines data flow and tool usage via Agentics framework. We evaluate how closely the agent's decisions align with the human and token-weighted outcomes, uncovering strong alignments measured by carefully designed evaluation metrics. Our findings demonstrate that agentic AI can augment collective decision-making by producing interpretable, auditable, and empirically grounded signals in realistic DAO governance settings. The study contributes to the design of explainable and economically rigorous AI agents for decentralized financial systems.

Figures (2)

• Figure 1: DAO-AI Overview: End-to-end architecture of DAO AI for governance vote recommendation. Given a Snapshot proposal URL as input (1), the system orchestrates a sequence of Modular Composable Programs (MCPs) to fetch proposal metadata (2), analyze forum discussions, assess market responses, and evaluate voting dynamics (3). The collected evidence is synthesized into a structured prompt, guiding the language model to select a single vote option and generate a justification that reflects community sentiment, historical outcomes, and economic impact (4). Finally the system generates a vote recommendation with justification (5).
• Figure 2: Proposal Example from Uniswap and Poll.