Who Decides When Nobody Is in Charge?
Governance is the first problem you hit when building an autonomous business, and it is arguably the hardest. Traditional companies solve governance through hierarchy – a CEO makes decisions, a board provides oversight, and employment contracts enforce compliance. Remove the humans from that equation and you need to replace every one of those mechanisms with something else.
The uncomfortable truth is that we have not figured this out yet. DAOs have been experimenting with decentralized governance since 2016, and the results are, at best, mixed. But the experiments are instructive.
The Governance Trilemma
Decentralized governance systems face a trilemma similar to the blockchain scalability trilemma. You can optimize for any two of three properties, but achieving all three simultaneously remains elusive:
- Efficiency – decisions are made quickly enough to respond to market conditions
- Legitimacy – decisions reflect the genuine will of stakeholders
- Security – the governance process cannot be captured or manipulated
Traditional companies optimize for efficiency and security (the CEO decides, the board prevents fraud) at the cost of legitimacy (employees and customers have no formal say). DAOs typically optimize for legitimacy and security (token voting with on-chain execution) at the cost of efficiency (proposals take weeks to pass).
An autonomous business needs all three. A system that takes two weeks to approve a pricing change will be outcompeted. A system where a single whale can override the community will lose trust. A system vulnerable to governance attacks will lose its treasury.
Centralized vs. Decentralized: The Real Tradeoffs
The debate between centralized and decentralized governance is often framed ideologically – decentralization good, centralization bad. The engineering reality is more nuanced.
Centralized governance (a single AI agent or small committee makes decisions) offers speed and consistency. An AI agent can evaluate a proposal against predefined criteria in milliseconds. The problem is single points of failure. If the decision-making agent is compromised, hallucinates, or optimizes for the wrong objective, there is no check on its behavior. This is essentially the “benevolent dictator” model, and history shows that benevolent dictators rarely stay benevolent [1].
Fully decentralized governance (every token holder votes on every decision) offers legitimacy but introduces several well-documented pathologies:
- Voter apathy: In most DAOs, fewer than 10% of token holders participate in governance votes. Uniswap, one of the largest DAOs by treasury size, regularly sees participation rates below 5% [2].
- Plutocracy: Token-weighted voting means wealthy participants dominate. A single wallet holding 5% of tokens can effectively veto proposals in many governance frameworks.
- Slow execution: The full proposal-discussion-vote cycle typically takes 7-14 days. In fast-moving markets, this is an eternity.
Hybrid Models: The Emerging Consensus
The most promising approaches combine elements of both. These hybrid models typically feature:
Delegated governance: Token holders delegate their voting power to representatives who vote on their behalf, similar to representative democracy. Optimism’s “Citizens’ House” and “Token House” bicameral system is the most sophisticated implementation to date, separating resource allocation decisions from protocol upgrade decisions [3].
Tiered decision-making: Different categories of decisions require different levels of consensus. Routine operational decisions (adjusting parameters within predefined ranges) can be made by AI agents or small committees. Strategic decisions (changing the business model, large treasury expenditures) require broader consensus. Emergency decisions (responding to security incidents) can be made by a security council with time-limited authority.
Optimistic governance: Decisions are assumed to pass unless explicitly challenged within a time window. This inverts the traditional voting model – instead of requiring active participation to approve, it requires active participation only to object. This dramatically reduces voter fatigue while maintaining oversight [4].
On-Chain vs. Off-Chain Governance
A critical implementation decision is where governance actually happens.
On-chain governance means votes are recorded and executed directly on a blockchain. The smart contract counts votes and automatically executes the winning proposal. This provides strong guarantees – the outcome cannot be tampered with after the fact. Tools like Tally and Snapshot have made on-chain voting relatively accessible. Tally handles the full lifecycle of on-chain proposals for protocols like Compound, Uniswap, and ENS. Snapshot offers gasless off-chain voting with on-chain verification, dramatically reducing participation costs [5].
Off-chain governance means decisions are made through forums, calls, or informal channels, then executed manually or through trusted intermediaries. This is faster and cheaper but introduces trust assumptions. Someone has to faithfully execute the community’s decision, and history shows this does not always happen.
Most mature DAOs use a hybrid: off-chain discussion and temperature checks (often via Snapshot’s gasless voting), followed by formal on-chain execution for decisions that pass. This balances cost, speed, and security.
Governance Attacks
Any governance system can be attacked, but decentralized governance introduces novel attack vectors:
Flash loan governance attacks: An attacker borrows a large number of governance tokens within a single transaction, votes on a malicious proposal, and returns the tokens – all in one block. Beanstalk lost $182 million to exactly this attack in April 2022. The attacker flash-borrowed enough tokens to achieve a governance supermajority, passed a proposal to drain the treasury, and repaid the loan [6].
Defenses include vote-escrow mechanisms (tokens must be locked for a minimum period before gaining voting power), time-delayed execution (proposals must wait before being executed, giving the community time to respond), and quorum requirements that account for circulating supply rather than just tokens present.
51% attacks on governance: If a single entity acquires more than half the voting power, they can pass any proposal. This is distinct from a blockchain 51% attack – it is about governance tokens, not mining power. The defense is token distribution: ensuring no single entity can realistically accumulate a majority. Vesting schedules, team token lockups, and broad airdrops all help.
Voter bribery: Platforms like Convex Finance have demonstrated that governance votes can be effectively purchased. Convex accumulated enough Curve governance tokens to influence Curve’s emission schedule, creating a “metagovernance” layer. This is not necessarily malicious – Convex’s users voluntarily delegate – but it demonstrates that governance power concentrates even in ostensibly decentralized systems [7].
MakerDAO’s Endgame: A Governance Evolution Case Study
MakerDAO provides perhaps the most instructive case study in governance evolution. Founded in 2014, Maker has gone through at least three distinct governance phases.
In its early years, Maker operated as a relatively centralized foundation with token-weighted voting. This worked while the community was small and aligned. As the protocol grew to manage billions in collateral, governance became increasingly contentious. Proposals took months to pass. Voter participation declined. The foundation’s influence created tension with the decentralization ethos.
In 2022, founder Rune Christensen proposed the “Endgame Plan” – a radical restructuring that would split MakerDAO into multiple semi-autonomous “SubDAOs,” each with its own governance token and focused mandate. The core protocol would handle only the most fundamental decisions (collateral parameters, system-wide risk), while SubDAOs would handle specific business lines like real-world asset lending or consumer-facing products [8].
The Endgame represents a move toward what you might call “federated governance” – a model where a parent organization sets broad parameters but subsidiaries operate independently within those bounds. This mirrors how large conglomerates like Berkshire Hathaway operate, with a light-touch holding company and autonomous subsidiaries.
Whether Endgame succeeds remains to be seen. The rebranding to “Sky Protocol” in 2024 and the introduction of USDS (replacing DAI) showed the community that even fundamental elements could change through governance. But it demonstrates that governance itself must evolve – the model that works for a $10 million protocol will not work for a $10 billion one.
Implications for Autonomous Businesses
For an autonomous business – one where AI agents handle day-to-day operations – governance becomes even more complex. The system needs:
Machine-readable governance rules: AI agents cannot interpret ambiguous natural-language policies. Governance parameters must be encoded precisely, with clear boundaries and fallback behaviors.
Real-time oversight mechanisms: Human governors cannot review every agent decision, but they need to be alerted when agents approach the boundaries of their authority or when anomalies occur.
Graceful authority escalation: When an AI agent encounters a situation outside its governance mandate, it needs a clear path to escalate to a broader decision-making body – without halting operations entirely.
Governance of the governance: Someone or something needs to decide when the governance rules themselves need to change. This meta-governance problem is recursive and has no clean solution.
The honest assessment is that we are still in the early experimental phase. The DAO ecosystem has produced valuable data on what does not work, and some promising patterns for what might. But no existing governance model is ready for a fully autonomous business operating at scale with real economic consequences.
References
[1] V. Buterin, “Moving Beyond Coin Voting Governance,” vitalik.eth.limo, 2021.
[2] Messari, “Governor Note: Uniswap Governance Activity Report,” 2024.
[3] Optimism Collective, “Working Constitution of the Optimism Collective,” 2023.
[4] Gnosis, “SafeSnap: Trustless On-Chain Execution for Off-Chain Voting,” 2021.
[5] Snapshot Labs, “Snapshot Documentation,” docs.snapshot.org, 2024.
[6] Rekt News, “Beanstalk - Rekt,” rekt.news, April 2022.
[7] Convex Finance, “Understanding Convex and veCRV,” docs.convexfinance.com, 2023.
[8] R. Christensen, “The Endgame Plan,” MakerDAO Forum, 2022.