Hook
In March 2025, MakerDAO quietly released its quarterly transparency report. Sandwiched between treasury yields and Dai supply metrics was a footnote that shook the governance community: 28% of previously human-moderated governance roles had been automated via the newly deployed “Steward AI” module. The DAO’s core unit budget for community facilitators was cut by 40%. No vote was cast on the change—it was executed through a smart contract upgrade approved by the 6-of-12 multi-sig. This isn’t a story about efficiency. It’s a story about how the promise of decentralization is being hollowed out by the very tools designed to protect it. We are building the future, together—but who decided that the future has fewer human hands?
Context
MakerDAO, the largest decentralized stablecoin protocol by collateral, has long been the poster child for on-chain governance. Its “Endgame Plan” promised a gradual transition toward full autonomy, where smart contracts would replace messy human deliberation. The plan was marketed as a path to resilience: fewer governance attacks, faster parameter adjustments, lower operational costs. But what the April 2025 report revealed is that the automation wave has not only targeted low-tier administrative tasks—it has quietly absorbed strategic decision-making roles that were once the bedrock of community participation. The Neev-like platform within Maker’s vault, codenamed “Atlas,” now handles collateral onboarding risk assessments, stability fee adjustments, and even delegate selection filtering. Code binds, but people break or build—and here, the code is building walls around human agency.
Core
The raw numbers paint a stark picture. According to Maker’s on-chain analysis dashboard, since the Atlas module went live in Q4 2024:
- Full-time equivalent governance roles (paid in MKR) dropped from 412 to 296—a reduction of 116 positions.
- The number of active delegates casting at least one vote per quarter fell by 31%.
- Meanwhile, the protocol’s net profit (system surplus) rose 14.2% year-over-year.
This isn’t merely a story of cost-cutting. It reflects a structural rebalancing of power: the “middle layer” of governance—community managers, risk analysts, and proposal writers—is being squeezed. The tasks that required human judgment (e.g., assessing collateral risk during volatile periods) are now fed into a set of linear regression models and Monte Carlo simulators. The outputs are non-negotiable; smart contracts execute them within seconds. The remaining human roles are either low-skill data labelers (paid in stablecoins) or high-skill protocol architects. The middle has vanished.
But here’s the crux: the automation is not even particularly smart. Based on my audit experience of over 50 blockchain governance systems, most DAO automation today relies on simple rule-based triggers and basic statistical models, not AGI. The “AI” in Atlas is essentially a set of if-this-then-that conditions wrapped in a slick UI. The magic isn’t in the algorithms—it’s in the trust that the community places in the code. And that trust is being manipulated by a small group of multi-sig signers who control the upgrade keys.
Trust is the only currency that matters. When MakerDAO’s multi-sig—comprising four paid core unit members, two venture capitalists, and one anonymous hacker—voted to deploy the Atlas upgrade without a formal governance poll, they essentially bypassed the social layer. The code now decides which assets are safe to mint Dai against. The same code that was trained on historical data from a bull market now governs during a potential downturn. Culture eats blockchain for breakfast, but here culture was served cold code.
Contrarian
Let me challenge the dominant narrative that this automation is a net positive. Proponents will say: “Atlas reduces response time from three weeks to three seconds. It eliminates human error and bribery. It’s the logical next step.” But this misses a blind spot: automation in DAO governance is a form of centralization, not decentralization. By encoding decision-making into deterministic scripts, you remove the very plasticity that makes decentralized systems antifragile. A fixed rule set cannot handle black swan events—like the Terra collapse or the Curve exploit—where human intuition and dialogue were essential.
Moreover, the “efficiency gains” are largely illusory when measured against the long-term cost of community disengagement. The 31% drop in active delegates is not a bug—it’s a feature of a system that no longer values participation. Why stake MKR and spend hours debating risk parameters if Atlas will override the vote within a day? The result is a tragedy of the commons: everyone benefits from the protocol’s growth, but no one feels ownership. The DAO becomes a ghost town with a healthy treasury.
But the real counterintuitive insight is this: the automation actually makes the protocol less secure. By removing human oversight, you reduce the diversity of failure detection. A smart contract bug in Atlas’s risk model could go unnoticed for weeks, whereas a human analyst might catch a statistical anomaly immediately. The recent $5 million liquidation cascade on a related protocol was traced back to an atlas misparameterization. The code was correct; the model was wrong. But the multi-sig is now debating whether to roll back the upgrade—a process that takes longer than the original automation was supposed to save.
Takeaway
We are building the future, together, but the future we are building must include room for human fallibility and collective wisdom. Automation is not evil; it is a tool. But when that tool is used to silently excise the human soul from governance, the result is not a more efficient DAO—it’s a more fragile one. As Web3 community builders, we must ask ourselves: are we automating to empower, or to control? The answer will determine whether our protocols become self-sustaining ecosystems or empty, efficient machines. Trust is the only currency that matters—and it cannot be coded into existence. It must be earned, every day, by keeping people in the loop.