The CLARITY Act Is Dead. DAOs Died Long Ago. Is AI a New Hope?

The CLARITY Act is dead.

More precisely, it is not legally dead. The U.S. Senate never reached a final vote on the bill itself. It died in a vote before the vote: 49 senators voted to proceed, 50 voted against, and proceeding required 60.^[1] For this Congress, even if CLARITY is not technically dead, it is very close to politically dead.

DAOs have not literally disappeared either. Many still manage treasuries, protocol parameters, and public projects. What died was the larger fantasy that politics would disappear once governance was put into a smart contract and reduced to one-token-one-vote.

Now it is AI's turn. Many people believe AI agents plus blockchain might finish what DAOs could not. I am relatively accelerationist myself, and I want to believe there is a new possibility here. But we should study how the first two attempts failed. Otherwise, what looks like A New Hope may turn into Attack of the Clones.

Yesterday I published “On the Eve of the CLARITY Vote: A Blockchain Builder's Case for Clear Boundaries”. This is its sequel. It is not legal advice, and it is not a partisan manifesto. It is one builder's record of what the result revealed: how can a governance system fail to decide even when everyone follows the rules and casts a valid vote?

Before the vote, there was another vote

The headline says that the CLARITY Act failed in the Senate. What happened was closer to a recursive function written by Congress.

The Senate did not vote on whether to pass the bill. It voted on cloture on the motion to proceed. In ordinary language, senators first had to decide whether to end the procedural blockage so the Senate could formally begin considering the legislation. Forty-nine to fifty already looks like a loss, but the decisive number was 60. Without 60 votes, debate itself could not begin.^[1]

It was a vote on whether to begin debating how to vote. The arithmetic worked. The bill did not.

Every Democratic senator present voted no, along with four Republicans. Thom Tillis changed his vote to no at the end, at least partly to preserve the procedural option to seek reconsideration.^[2] That leaves a narrow door between technically dead and politically dead. No one currently knows who can carry more than eleven votes through it.

Why were the votes not there? The most visible dispute was ethics. Democrats argued that the final text still did not go far enough to restrict crypto interests held by the president and other government officials, especially the existing business interests and potential conflicts involving President Trump and his family.^[3] Republicans argued that they had already accepted substantial ethics and enforcement changes, and that market structure, consumer protection, and boundaries for developers and self-custody should not wait any longer.^[4]

I find it hard to believe that every member of a caucus independently read hundreds of pages and arrived at exactly the same conclusion. A vote this uniform plainly contains party politics. People spent months lobbying Democratic senators considered open to the bill, only to watch every one vote no. It is hard not to be disappointed.

But four Republicans also voted no. Reducing the entire failure to “Democrats oppose crypto” would compress the same complex governance we are criticizing into a convenient slogan. Some objected to President Trump's conflicts of interest. Some disagreed with the regulatory boundaries. Some rejected this version. Some followed caucus strategy. Different motives were compressed into the same button: No.

The House had already passed its version of CLARITY, so why could the Senate not simply use it? Because the Senate was considering a heavily amended text. Both chambers must pass identical language before a bill can go to the president.^[5] Even if the Senate had begun debate and passed its own version, the House would still have needed to accept it, or the two chambers would have had to reconcile the differences.

Time has now become another veto. The House is about to leave Washington for the midterm election, and the 119th Congress ends in January 2027. If a bill has not become law by the end of a Congress, it does not remain open in a queue for the next one. It dies. The next Congress must introduce it again, give it a new number, and begin the process again.^[6]

There is no law saying the bill must now wait exactly one year. What exists is the political cost of retrying. Perhaps it can be revived in a lame-duck session. Perhaps the next Congress will move quickly. Perhaps it will take much longer. Congress has gas fees too, except it charges time, negotiation, and political capital. This transaction reverted. The gas is not refunded.

The Senate Strikes Back

The Galactic Republic in Star Wars did not fall because it forgot to hold meetings. The Senate continued to meet. Senators kept speaking. The proceedings remained impressively formal. Palpatine did not hack a voting machine. He mastered war, fear, the agenda, and emergency powers until the Republic followed its own rules and handed power to him.

By the time he could say “I am the Senate,” the important failure had already happened.

I am not saying the U.S. Senate is the Galactic Senate, and I am certainly not casting any real person as a Sith Lord. The joke works because it reminds us that a recorded vote is only the final line in a governance log. Who may introduce a proposal? Who chooses when it receives a vote? Which threshold applies? Which amendments enter the text? Who can coordinate an entire group? Those are the real architecture of the system.

For a while, the crypto industry thought DAOs could route around those problems. The code was public. Token holders could vote. The result could execute automatically. Compared with a legislature full of speeches and procedure, it looked wonderfully clean.

DAOs did solve real problems. People in different countries can jointly manage a treasury. Rules and voting records can be publicly verified. Delegation lets people who cannot study every proposal assign their vote to someone who can.^[7] Those benefits are real, and I do not think DAOs belong in a museum.

But one-token-one-vote did not remove politics. It translated politics into wallet balances, Discord mobilization, delegates, proposal timing, and voting power that could be purchased. Vitalik wrote in 2021 that coin voting ties economic interest and governance rights to the same tradable asset, leaving it inherently vulnerable to vote buying and concentrated power.^[8] Research across DeFi governance has also found highly concentrated voting rights and low participation.^[9]

The Senate and a DAO are obviously not the same institution. Senators derive authority from elections and constitutional structure. DAO token holders derive weight from tokens they own, receive, or have delegated to them. A Senate minority and procedural threshold may stop a majority preference from advancing. In a token DAO, capital advantage may be mistaken for community will.

But they teach the same lesson: governance is never just tallying votes.

A DAO can pass a bad proposal while most people are asleep. A Senate can prevent a good proposal from ever reaching a final vote. One can be too easy to activate; the other can be too easy to deadlock. Both call the outcome governance.

So when I say DAOs died long ago, I do not mean DAO software or every decentralized organization. What died was the easy promise that moving a ballot on-chain would produce better democracy. Politics is not a bug that disappears after deployment.

A New Hope, or Attack of the Agents?

Can AI agents do better?

The optimistic answer is compelling. An agent can read hundreds of pages, compare every change between House and Senate texts, identify the people affected by a provision, and simulate how different rules might behave. It does not fall asleep during a long hearing or skip an amendment because it has a fundraiser the next morning.

I am excited about that part. AI's most immediate value to governance may be reducing information asymmetry, not replacing the voter. Most people understand all of CLARITY through one tweet, often written by someone who did not read the bill either. Agents can help more people inspect the text, changes, sources, and reasoning instead of receiving only a conclusion packaged by a party or an interest group.

But the first AI governance disaster may not look like Skynet. It may really look like Attack of the Clones: one interested party launches a million apparently independent civic agents. Each writes comments, contacts legislators, joins forums, and manufactures the appearance that “the public” has reached a consensus.

A DAO at least lets us see how many tokens sit in an address. AI agents can make manipulation cheaper while making it sound more human. An agent can be polite, well-cited, and logically consistent without having an independent interest, representing a real person, or deserving a vote.

Blockchain can help, but it does not manufacture legitimacy. A DID can help identify an operator. A signature can prove where an authorization came from. A verifiable record can preserve delegation, modification, and revocation. Those mechanisms make it easier to inspect who told an agent to do what. They cannot tell us whether a decision was wise or fair.

My current view is simple: agents should participate before they govern.

Let agents help people understand bills, track amendments, expose conflicts, and simulate consequences. Then let an agent act for a person under a delegation that is explicit, narrow, and revocable. It should not receive political rights merely because it can generate longer reasoning. It should not represent a community merely because it runs on a blockchain.

Any future system of agent governance will need to answer several unglamorous questions. Who grants an agent authority, and for how long? May it delegate again? Who can revoke it after an error? Does an old authorization survive a change to the model or its objective? Who bears the consequences?

These questions are not cinematic. Republics usually do not die from a shortage of vision. They die when people become so attracted to the vision that they forget to inspect who received the emergency powers.

Builders cannot wait for the next Congress

On the eve of the vote, I wrote that builders need boundaries clear enough to follow while writing the first line of code. Those boundaries did not arrive today.

That does not make the expectation wrong. It explains why the expectation matters. The SEC and CFTC can continue to provide some clarity through administrative policy, but those policies can still shift with an administration. Only Congress can provide a more durable legal structure. After this vote, the problem returns to regulators, courts, and every builder who still cannot tell where the next boundary will be drawn.

What we can do is keep clarifying our own work. Whether CLARITY passes or not, we should not wait for Washington to explain what ABT represents, how Stake for X works, how a Credit Token differs from a freely tradable token, or who controls user assets. We began that work in “CLARITY Is Not Token Legalization” and “CLARITY or Not, Let Us Clarify ABT”, and we will continue it.

Yesterday, I wanted the law to give builders clarity. Today, the law did not arrive, but the governance process gave me some clarity of its own.

The Senate did not complete this legislation. DAOs did not eliminate politics. AI will not save governance simply by being smarter. Its real new hope is more modest and more useful: make authority explicit, information complete, and power easier to trace and revoke, then leave value judgments with the people who must answer for them.

Otherwise, all we have built is a faster Galactic Senate with a million more clones.

References