DAO (Decentralized Autonomous Organization): Governance by Code and the Limits of That Promise

A DAO promises to swap out hierarchical management (a board, executives, appointed managers) for governance rules enforced directly by code: a proposal is submitted, token holders vote, and if it passes, the outcome executes on-chain without needing anyone's further permission. That technical promise is real, and understanding its mechanics is worth doing on its own terms.

What Is a DAO: Definitions and Terminology

The term DAO (Decentralized Autonomous Organization) predates most of what people now associate with it. Vitalik Buterin's May 2014 Ethereum Foundation blog post, "DAOs, DACs, DAs and More: An Incomplete Terminology Guide," tried to pin down the vocabulary before it calcified (buterin2014terminology in the references): a DA (Decentralized Application) is any application whose logic runs on a decentralized network; a DAC (Decentralized Autonomous Company) additionally issues tradable, equity-like shares; and a DAO sits at the most ambitious end of the spectrum: an entity with internal capital (a treasury it owns and can move on its own initiative) and a degree of autonomy in how that capital gets deployed, ideally under rules that require no trusted party to enforce them. Buterin was explicit that these categories blur into each other in practice, and a decade on, the industry mostly uses "DAO" as an umbrella term for all of them.

A more formal definition came later from Rolf Hassan and Primavera De Filippi, in a 2021 Internet Policy Review paper (hassan2021daos in the references): a DAO is the combination of (1) governance rules encoded as smart contracts, (2) a record of member interactions and decisions maintained on a blockchain, and (3), typically, a native token representing both capital and voting rights. No single element here is unique to DAOs; what makes the combination distinctive is that governance is not merely documented on-chain, it is executed there: a passed proposal does not require anyone's further permission to take effect.

The contrast with a traditional corporation is worth making concrete rather than rhetorical. A corporation has legal personhood granted by a state, a board of directors accountable under fiduciary law, and bylaws that are text, interpreted, and when necessary enforced, by courts. A DAO, absent a specific legal wrapper (see the legal-status discussion further below), has none of this by default: no legal personhood, no board, and "bylaws" that are literally executable code rather than a document a judge can interpret contextually. That difference is exactly what "governance by code" both promises and, as this article examines, struggles to fully deliver.

Technical Foundation: How On-Chain Governance Actually Executes

Propose → Vote → Timelock → Execute: The Governor Pattern

The reference implementation most DAO tooling still imitates is Compound's Governor Bravo pattern. A proposal first needs a sponsor holding, or delegated, at least a minimum token threshold before it can even be submitted; once submitted, a voting period opens during which token holders vote for, against, or abstain, and the proposal passes only if it clears both a quorum (a minimum amount of participating voting power) and an approval threshold (typically a simple majority of votes cast). A passed proposal does not execute immediately; it queues into a timelock contract that enforces a fixed delay, commonly 48 hours, before anyone is permitted to trigger execution.

That delay is not an incidental implementation detail; it is a deliberate safety valve. During the timelock window, anyone (token holders, exchanges, security researchers) can inspect exactly what a passed proposal will do and react before the funds move: exchanges can pause withdrawals, users can exit, and in some designs a guardian multisig can cancel the queued action outright. This stands in sharp contrast to The DAO in 2016, which had no timelock at all between a proposal executing and funds moving, a gap covered in the historical section below.

Delegation and "Liquid Democracy"

Token voting rarely requires holders to vote directly. Compound-style governance tokens (COMP and its many imitators) let a holder delegate their voting power to another address without transferring the underlying tokens: the delegate votes, the delegator keeps custody and can revoke the delegation at any time. In theory this is liquid democracy: representation flows fluidly to whoever a holder trusts to stay informed, and can be recalled instantly, unlike a fixed-term elected representative. In practice, delegation tends to concentrate voting power into the hands of a small number of highly visible delegates (professional delegate services, founding-team-adjacent addresses, and a handful of large funds) rather than distributing it evenly, a pattern documented empirically in the governance-reality section below.

Off-Chain Signaling + On-Chain Execution: Snapshot

Because every on-chain vote costs gas, and quorum and voting-period requirements slow decision-making down, many DAOs run non-binding signaling votes off-chain using Snapshot: voters sign a message with their wallet (no transaction, no gas), and voting power is computed from a token-balance snapshot at a given block, with ballots stored on IPFS. Snapshot itself does not execute anything; the outcome of a Snapshot vote is typically carried out by a multisig of trusted signers who manually execute the corresponding transaction. This pattern is enormously popular, since it is cheap and fast, but it quietly reintroduces exactly the kind of centralization DAOs are supposed to avoid: the binding step in the process is a small group of humans choosing to honor a non-binding vote, not code executing autonomously.

Beyond One-Token-One-Vote: Quadratic Voting/Funding and Conviction Voting

One-token-one-vote directly converts wealth into governance power, which is one of the central critiques covered later in this article. Several designs try to blunt that mapping. Quadratic funding, proposed by Buterin, Zoë Hitzig, and Glen Weyl in a 2019 Management Science paper, "A Flexible Design for Funding Public Goods" (buterin2019flexible in the references), computes matching funds for a public good as proportional to the square of the sum of the square roots of individual contributions rather than the raw sum; mathematically, this rewards the number of independent contributors more than the size of any single contribution, and is the mechanism behind Gitcoin Grants' matching pools. Quadratic voting applies the same square-root logic directly to votes: casting n votes on an issue costs n² units of voting credits, so the marginal cost of buying more influence rises quickly, blunting (though not eliminating) the ability to convert wealth into outsized power. Both designs share a critical weakness: without robust identity verification, a single actor can split into many pseudonymous identities, a Sybil attack, and capture the "per-person" benefit repeatedly, which is why quadratic mechanisms in practice lean on DID-style proof-of-personhood or reputation systems to hold up at all. Conviction voting, used by 1Hive and the Commons Stack, takes a different axis entirely: instead of a fixed voting window, a voter's weight on a proposal accrues continuously the longer their tokens stay committed to it, so sustained conviction outweighs a large but fleeting vote.

The Minimal Viable DAO and the Ragequit Safety Valve: Moloch

MolochDAO, launched in 2019 as a deliberately stripped-down grant-making vehicle for Ethereum public goods, minimized its own feature surface as a direct reaction to how much attack surface and coordination overhead earlier, more ambitious DAO frameworks had accumulated. Its most influential contribution is ragequit: after a proposal passes but before it executes, any member who voted against it, or simply disagrees, can exit immediately, burning their membership shares in exchange for their proportional slice of the treasury, before the proposal's effects land. Ragequit is an exit-based, not vote-based, check on majority rule, conceptually close to Albert Hirschman's classic exit-versus-voice framework: a dissenting minority in a Moloch-style DAO is not stuck lobbying for votes it will lose (voice), it can simply leave with its capital intact (exit), which in turn disciplines the majority, since a mass ragequit visibly drains the treasury the majority is trying to control.

The Historical Turning Point: The DAO Hack and the Code-Is-Law Debate

In April-May 2016, a project simply named "The DAO" ran a token sale that raised roughly 12.7 million ETH, worth about $150 million at the time, making it the largest crowdfunding event yet seen, intended to function as a member-directed venture fund governed entirely by on-chain voting. In June 2016, an attacker exploited a reentrancy vulnerability: a flaw in which a contract's external call to another contract (here, a withdrawal) could recursively call back into the vulnerable function before the first call had finished updating the contract's internal balance, letting the same funds be withdrawn repeatedly before the accounting caught up. Roughly 3.6 million ETH, about $60 million at the time, was drained before the attack was halted. Nicola Atzei, Massimo Bartoletti, and Tiziana Cimoli later formalized reentrancy as one of a taxonomy of Ethereum smart-contract vulnerability classes in their 2017 survey (atzei2017survey in the references); The DAO hack remains the canonical example, and smart contracts covers how this class of bug fits into the broader picture of on-chain code as law.

The Ethereum community's response was a contentious hard fork that effectively reversed the theft by moving the drained funds to a recovery contract, a decision reached not by any on-chain vote (The DAO's own governance had no mechanism to undo a completed exploit) but through off-chain coordination among core developers, miners, and exchanges. A minority of the network rejected the fork on the principle that a blockchain's history must remain immutable regardless of outcome, and continued the original, unmodified chain as Ethereum Classic (ETC), while the forked chain retained the Ethereum (ETH) name. Quinn DuPont's 2017 ethnographic study of The DAO (dupont2017experiments in the references) documents in detail how the episode forced a reckoning between "code is law" as an ideology (the idea that a smart contract's literal execution is the final word, full stop) and the social and political reality that surfaced the moment the code failed catastrophically and a community of humans had to decide, collectively and outside the code, what "correct" even meant.

The critical point for evaluating DAOs as a governance model is this: a system explicitly designed to be autonomous, to need no off-chain human intervention, turned out, at the first real crisis, to depend entirely on off-chain human and social consensus to resolve. That does not make code-based governance worthless; it means the premise that coding the rules eliminates the need for management and judgment does not survive contact with a genuine edge case.

The Governance Reality: A Critical Examination

Voter Apathy and Power Concentration

Rainer Fritsch, Marino Müller, and Roger Wattenhofer's 2022 empirical study of major DAOs, including Uniswap, Compound, and ENS (fritsch2022analyzing in the references), measured actual on-chain voting behavior rather than taking "decentralized" as given. Their findings: voting power is heavily concentrated in a small number of addresses and delegates (in several of the DAOs studied, a handful of delegates could unilaterally meet quorum on their own), and turnout, measured as the share of eligible tokens actually cast in a given vote, was frequently in the low single digits. The label "decentralized" in DAO branding, in other words, frequently does not match the measured concentration of actual decision-making power.

The Critique of Coin Voting

Buterin himself became one of the most prominent critics of the mechanism his own ecosystem popularized. His August 2021 essay, "Moving Beyond Coin Voting Governance," lays out three structural problems with plain one-token-one-vote: plutocracy, since voting power maps directly onto wealth, so outcomes tend to reflect what benefits large holders rather than the protocol's broader user base; voter apathy, a form of rational ignorance: a small holder's vote is vanishingly unlikely to change the outcome, so the rational choice is simply not to research proposals or vote at all, concentrating effective power further among whoever does show up; and bribery and vote markets, the observation that voting power itself, unlike most other rights attached to a token, can be temporarily rented or purchased without the briber giving up anything else: a briber does not need to buy the token permanently to buy its vote for a single proposal.

The Curve Wars are a concrete, sustained illustration of the bribery critique in action. Curve Finance's governance token, CRV, uses a vote-locking model (veCRV): locking CRV for up to four years grants voting power that decays as the lock approaches expiry, intended to align voters with the protocol's long-term health. Convex Finance emerged as a meta-governance layer that aggregates users' CRV into long-term locks on their behalf and issues a liquid token in return, letting Convex itself accumulate outsized veCRV voting power without individual users sacrificing liquidity. That concentrated voting power then became something protocols found worth paying for directly: marketplaces such as Votium let protocols bribe Convex/veCRV voters to direct Curve's liquidity-gauge weights, and therefore token emissions, toward their own pools. The result is a fully legitimized, openly operating bribery market, precisely the dynamic Buterin's essay warned was structurally inherent to coin voting, not a hypothetical edge case.

Principal-Agent Problems, Old and New

Traditional corporate governance did not solve the principal-agent problem, the risk that the people running an organization (agents) act against the interests of those who own it (principals), but it did build institutionalized checks around it over roughly a century of corporate law: directors owe an enforceable fiduciary duty, boards face accountability mechanisms including removal and liability, and shareholders retain the right to bring derivative suits against management on the company's behalf. DAOs largely lack equivalent, enforceable duties running between token holders, delegates, and core development teams: a delegate who neglects their responsibilities, or a founding team that quietly retains outsized influence, faces little that resembles fiduciary liability. At the same time, DAOs introduce a genuinely new misalignment: token-price incentives push holders toward decisions that favor short-term price action over long-term protocol health, and founding teams and venture investors frequently retain concentrated token allocations that grant outsized influence long after a project markets itself as "decentralized."

What Ostrom's Commons-Governance Principles Reveal

Elinor Ostrom's 1990 "Governing the Commons" (ostrom1990governing in the references), work that earned her the 2009 Nobel Memorial Prize in Economic Sciences, derived eight design principles from decades of studying real, durable, self-governing institutions managing shared resources: clearly defined boundaries (who counts as a member, what resource is governed), congruence between rules and local conditions, collective-choice arrangements that let those affected by rules participate in changing them, monitoring (ideally by the members themselves), graduated sanctions for rule violations, low-cost and accessible conflict-resolution mechanisms, minimal external recognition of the group's right to self-organize, and, for larger systems, nested layers of smaller governance units.

Rozas, Tenorio-Fornés, Díaz-Molina, and Hassan's 2021 paper (rozas2021ostrom in the references) explicitly maps Ostrom's framework onto DAOs and other blockchain-based commons governance, and the fit is instructively uneven. DAOs tend to satisfy the principles that blockchains make nearly free: boundaries are enforced by token-gating, and monitoring is close to perfect since every vote and transaction is public by construction. But DAOs structurally struggle with graduated sanctions: smart contracts are much better suited to binary outcomes, a proposal executes or it does not, than to proportionate, escalating responses to a member's bad behavior, and with accessible conflict resolution, since disputes that fall outside what a contract can adjudicate have nowhere obvious to go on-chain. Ostrom's framework matters here precisely because it is neither hype nor dismissal: it offers a rigorous, empirically grounded standard for evaluating whether a given DAO's governance is likely to prove durable, independent of how impressive its architecture looks.

Legal Status: The Unincorporated-Association Problem and Legal Wrappers

In 2022, the U.S. Commodity Futures Trading Commission brought an enforcement action, CFTC v. Ooki DAO (N.D. Cal.), and won on a theory with sweeping implications: that Ooki DAO could be treated, for legal purposes, as an unincorporated association, meaning individual token holders who participated in governance votes could face personal liability for the DAO's regulatory violations. The case is a landmark precedent for a reason that cuts against a common intuition: having no legal wrapper does not mean no legal exposure: for a DAO with real-world activity and no incorporated shell, it can instead mean unlimited, personal, joint liability for anyone who governs it.

The direct response from several jurisdictions has been to offer purpose-built legal wrappers. Wyoming's DAO LLC Supplement (2021, codified at W.S. 17-31) and subsequent Marshall Islands DAO legislation let a DAO register as a limited-liability legal entity, capping members' personal exposure the way an ordinary LLC or corporation would. The irony is direct: adopting a legal wrapper to solve the Ooki DAO problem means reintroducing exactly the things (a jurisdiction, a registered agent, a recognized legal entity subject to a state's courts) that a "code is law" DAO was conceived to make unnecessary. Cristian Santana and Laura Albareda's 2022 review in Technological Forecasting and Social Change (santana2022blockchain in the references) surveys these regulatory and legal challenges in depth, and concludes that legal uncertainty, not technology, is now one of the largest open obstacles to DAO adoption at scale.

Case Studies Compared

Common Misconceptions and Caveats

• "A DAO runs with no humans involved" is a misconception: most day-to-day coordination still happens off-chain, in Discord servers and forum threads that never touch the blockchain, and core development teams or founding foundations frequently retain outsized informal influence even after formally "decentralizing" governance. DuPont's ethnographic account of The DAO found that, notwithstanding its name, even it relied on a human curator team to vet which proposals were fit to bring to a vote.
• "On-chain voting equals maximal decentralization" is a misconception, refuted directly by the concentration data in Fritsch, Müller, and Wattenhofer's study and by the bribery-market dynamics of the Curve Wars: a fully on-chain, fully transparent vote can still be a vote effectively controlled by a handful of addresses.
• "Code is Law means the legal system is irrelevant" is a misconception: CFTC v. Ooki DAO demonstrates that regulators and courts can and do reassert jurisdiction the moment a DAO's activity causes real-world harm, regardless of how its governance is technically implemented.
• "Removing hierarchical management removes the principal-agent problem" is a misconception: in practice, the classic principal-agent problem does not disappear; it is replaced by different, and in some respects less accountable, misalignments, including token-price short-termism, concentrated VC and founder allocations, and low-turnout capture by a small, motivated minority.

Related Pages

The threads running through this article connect to several other deep dives on this site. Deciding what counts as "the" outcome of a DAO vote is, structurally, the same problem as agreeing on one canonical truth in distributed consensus: both require converting many independent, possibly conflicting inputs into a single binding record. The DAO hack itself is a foundational case study in smart contract vulnerability, and the way real DAOs actually move treasury funds in practice (a small set of trusted signers executing what a vote decided) is exactly the pattern covered in multisig wallets. More broadly, a DAO is simply one category, albeit a particularly ambitious one, of DApp: an application whose core logic and state live on a decentralized network rather than a company's servers. And the two open problems that recur throughout the governance critique above, Sybil resistance for mechanisms like quadratic voting and durable, portable identity for delegation and reputation, are covered respectively in attacks on P2P and blockchain networks and in DID.