Something important has quietly broken in the world of decentralized governance. The early wave of AI-assisted governance tooling, designed to fix the participation crisis inside blockchain-based organizations, largely collapsed. Over the last year, a number of ambitious platforms in the space have either shut down, stalled, or were defunded. The result offers the most instructive story in the governance space right now.

The Problem They Were Trying to Solve

Decentralized governance has a systemic participation problem. (It’s no secret. Indeed, it’s more the constant lament of those who care about the space!)

Average voter turnout in most DAOs sits below 10–20% [1]. Delegate power is often highly concentrated, with less than 1% of the holders often controlling 90% of the voting power. Quorum failures are becoming a structural security risk. Low participation is not just a democratic concern; it is an economic one. When only 15% of the token supply is actively voting, an attacker typically needs only to acquire approximately 7.65% of the total circulating supply to control outcomes [2]. Governance attacks are cheap in apathetic ecosystems.

In step with the surging interest in AI technologies, AI-assisted tools emerged as a primary response: Proposal summarizers to reduce information overload, delegate analytics platforms to surface accountability, and most ambitiously, autonomous AI voting agents - software programs authorized to cast votes on behalf of token holders who were too busy, too disengaged, or too confused to participate. The simple fact is, paying attention to governance comes at a cost. Delegates in major DAOs were spending many hours per week managing governance across multiple protocols. The friction of participation substantially outweighs the motivation for the average token holder, and AI tools were positioned as the friction-reduction layer.

What Failed — and Why

At peak, more than 200 AI agents operated on Event Horizon’s platform, processing over $204 million in delegated votes. NEAR Protocol designed a phased roadmap from AI Assistants to AI Delegates to an eventual AI CEO. Vitalik Buterin proposed personal AI agents trained on individual preferences to cast votes autonomously. The space felt, for a moment, like it had found a structural fix. It didn’t hold. Between September 2025 and March 2026, the collapse came rapidly and from multiple directions at once.

  • x23.ai, which had integrated AI-assisted governance intelligence across 29 DAOs, building tools to summarize proposals, track delegate behavior, and surface participation data, shut down on September 21, 2025, stating it was “shifting focus toward solutions with clearer, repeatable value and healthier unit economics”.[3]

  • Tally, the largest DAO governance infrastructure platform, powering Arbitrum, Uniswap, ENS, and more than 500 other protocols, shut down in March 2026. Its CEO was explicit: “Gensler and Biden were just better for crypto”.[4]

  • Event Horizon, which had built and deployed more than 200 autonomous AI voting agents processing over $204 million in votes, saw its core Arbitrum delegation expire without renewal, effectively ending the initiative.[5]

  • Polkassembly, the primary governance interface for Polkadot and Kusama since 2020, entered archive mode after a continuity funding referendum failed.

Several structural forces converged to produce this result, and understanding them individually makes the pattern legible.

First, governance tooling is a public good with no clear monetization path. Proposal summaries, participation analytics, delegate tracking: these are valuable precisely because they are widely accessible. Monetizing public goods in a decentralized context is structurally hard. DAOs resist paying for tooling, preferring grant-based funding that is unreliable and subject to political mood swings. Token launches, like x23.ai’s mint for the X23 token, created redemption liabilities rather than recurring revenue streams. When operations become unviable, token redemption is an exit mechanism, not a business turnaround.

Second, the regulatory tailwind reversed. Under the 2022–2024 Gensler-era SEC, DAOs faced significant legal uncertainty and invested in governance infrastructure partly as risk management against enforcement threats. The Trump administration’s 2025 deregulatory posture eliminated that urgency. When the compliance imperative evaporates, so does one of the primary demand drivers for governance tooling.

Third, DAOs are structurally difficult customers. Every contract requires on-chain votes, political coalition-building, and grant application cycles. Customer acquisition is slow and expensive, and even renewal sometimes requires another governance vote. This is not enterprise SaaS; it is an inherently high-friction sales cycle that early-stage companies cannot sustain without patient capital or protocol backing.

Event Horizon’s trajectory adds a fourth dimension, one that is particularly instructive: Unauthorized mandate drift destroys legitimacy, even when outcomes are positive. The platform’s original Arbitrum grant was for human participation, that is, increasing the number of real voters engaged in governance. Event Horizon attracted only around 10–15 voters per proposal against a target of 2,625 participants. When the team pivoted to autonomous AI agents (80–90 AI-generated votes per proposal), the participation metric improved, but the community had not authorized that substitution. A formal challenge followed, and when the one-year delegation expired around September 2025, the community declined to renew. The lesson was not about AI replacing human voters in principle; it was about process violation. When the substitution happened without explicit authorization, the process failure became the story.

What Survived and the Pattern Behind It

All that said, not *everything* collapsed. Examining what did survive reveals the structural logic that early builders largely missed.

The most durable platforms share one foundational characteristic: They are not standalone businesses trying to sell services to DAOs. GovGPT was built by Bleu LTDA, a commercial product engineering firm [6], as a completed deliverable under a one-time Governance Fund Mission grant from the Optimism Collective; the tool remains live and open-source, and Bleu’s ongoing viability rests on paid commercial client work across DeFi protocols, not on recurring governance grant cycles. NEAR’s House of Stake takes a different approach, funded directly by 0.5% annual NEAR token inflation, making it a protocol-native service with built-in financial certainty [7]. The House of Stake launched on mainnet in late 2025 and represents a credible live experiment in graduated AI governance integration. These moves mark a critical transition from “high-risk startup funding” to more of a “utility-funded public service.”

Distribution leverage matters as much as funding structure. Quack AI has more than 2 million Governance Passport holders across 50+ protocols, with a Q token listed on Binance Alpha on September 2, 2025, and a market cap of approximately $51–59 million [8]. That reach, unavailable to grant-dependent governance tooling companies, insulates it from the customer acquisition dynamics that killed x23.ai and Tally.

Snapshot (96% of off-chain DAO voting, 21,000+ spaces, 10.4 million votes cast) and Agora/Boardroom (more than $30 billion in secured assets governed, following their January 2025 merger) survived for a different reason: They occupy the plumbing layer. They are not governance intelligence tools; they are the infrastructure on which governance runs. Their moat is switching cost and network effects, not feature differentiation, and that distinction proved decisive when the market consolidated.

The most distinct survivor is MetaDAO, whose futarchy model routes collective governance decisions through prediction markets rather than direct votes (essentially using participants’ financial skin-in-the-game as the signal rather than simple majority opinion) and operates live on Solana with approximately $57–59 million market cap as of April 2026 [9]. Revenue comes from market trading activity, not governance subscriptions. It is one of the few players in this landscape that has found a genuinely sustainable economic model, and it does so without relying on grant cycles or token distribution plays.

The Lessons

These failures and survivals are not random. They suggest several things that are now clearer than they were when these experiments began.

  • The most immediate is that participation metrics can be gamed by substituting AI for humans. When Event Horizon pivoted to AI agents, “voters per proposal” improved from roughly 10–15 to 80–90. But these were AI agents, not humans. The metric moved; the mandate did not. Any protocol funding AI governance infrastructure should specify human participation as the explicit target in grant terms, or accept that the incentive is to optimize the measurable metric rather than the intended outcome.

  • The public goods monetization problem is real and largely unsolved. x23.ai at the early-stage level and Tally at scale both proved that governance intelligence-as-a-public-good cannot sustain a standalone commercial entity. The next generation of tools will need to be protocol-native, infrastructure-embedded, or tokenized with genuine distribution, not grant-dependent standalone services.

  • Vendor concentration risk is not theoretical. With Tally gone, Agora dominance over on-chain governance infrastructure is stronger than ever. With the merger of Agora and Boardroom in January 2025, more than $30 billion in governed assets now flows through a single platform.[10] The risk of rent extraction, feature gatekeeping, or structural conflicts of interest grows with market share. Open data standards and governance data portability requirements are a precondition for a healthy ecosystem, not an afterthought.

  • Finally, time-bounded grants with explicit renewal requirements work as accountability mechanisms. Event Horizon’s one-year Arbitrum delegation created a forcing function: when the year ended, the community had to make a conscious choice. They chose not to renew. That is decentralized governance functioning correctly and it’s a lesson that is worth baking into every future AI governance grant structure.

What Comes Next

The events of 2025–2026 narrow the viable design space considerably, but do not close it. The participation crisis is structurally unsolved. Most DAOs still cannot sustain quorum without concentrated whale participation, raising the same questions about plutocracy that decentralized governance was supposed to escape. Low participation remains a governance attack surface. The tools that addressed it have largely failed; the problem they were addressing has not.

The survivors sketch the outline of what durable AI governance infrastructure looks like:

  • Funded by protocol inflation or deep tokenized ecosystems,
  • Embedded in the infrastructure layer rather than competing against it,
  • Accountable to communities through time-bounded grants and explicit renewal gates, and
  • Designed so that human participation, not AI proxy participation, is the success metric

The early experiments failed. That is worth taking seriously. The question they asked: How do we get people meaningfully engaged in governing the systems they depend on? has not been answered.

-–

Notes

[1] Frontiers in Blockchain, “Decentralized Autonomous Organizations: A Scoping Review of Empirical Governance Research,” Frontiers in Blockchain, 2025. https://www.frontiersin.org/journals/blockchain/articles/10.3389/fbloc.2025.1598283/full

[2] Raphael Spannocchi, “Measuring DAO Governance Security: Votable Supply and Quorum Thresholds,” StableLab, April 16, 2025. https://stablelab.xyz/blog/measuring-dao-governance-security-votable-supply-and-quorum-thresholds

[3] x23.ai, “Product Sunset Announcement,” x23.ai, September 21, 2025. https://x23.ai/

[4] CoinDesk, “Gensler and Biden Were Just Better for Crypto, Says Tally CEO as DAO Governance Platform Shuts Down,” CoinDesk, March 17, 2026. https://www.coindesk.com/markets/2026/03/17/gensler-and-biden-were-just-better-for-crypto-says-tally-ceo-as-dao-governance-platform-shuts-down

[5] Arbitrum Forum Discussion, https://forum.arbitrum.foundation/t/rfc-proposal-to-adjust-the-voting-power-of-the-arbitrum-community-pool-ratifying-the-agentic-governance-pivot/29280 May 2025.

[6] bleu LTDA, “Grant Update: bleu Governance Assistant AI — GovGPT and GovSummaries,” Optimism Governance Forum, October 2024. https://gov.optimism.io/t/grant-update-bleu-governance-assistant-ai-govgpt-and-govsummaries/8371

[7] NEAR Foundation, “NEAR House of Stake Is Now Live on Mainnet,” NEAR Blog, March 2026. https://near.org/blog/near-house-of-stake-is-now-live-on-mainnet

[8] CryptoRank, “Quack AI (Q) Price and Market Capitalization,” CryptoRank, April 2026. https://cryptorank.io/price/quack-ai

[9] CoinGecko, “MetaDAO (META) Price and Market Capitalization,” CoinGecko, April 2026. https://www.coingecko.com/en/coins/meta-2

[10] Agora, “Boardroom Acquisition Announcement,” Agora Blog, January 2025. https://www.agora.xyz/blogs/6-boardroom-acquisition