AI Agents in DeFi: Infrastructure Builds Lag Behind Hype

AI agents have been a prominent topic in the Web3 community, with ambitious narratives suggesting that these autonomous, intelligent entities could manage capital, risk, and strategy across decentralized protocols. The idea was that these systems would not only outperform humans in execution but also free users from the constant monitoring and micromanagement of their digital assets. At the peak of this excitement, bold predictions were made, such as the claim that within a year, the majority of all DeFi Total Value Locked (TVL) would be managed by AI agents.

However, as time has passed, the initial buzz around AI agents has given way to a more realistic assessment. The current most popular AI agents are often limited to profiles with a token, which does not align with the grand vision for these agents. With their infrastructure still in its nascent or development stage, the concept of DeFi-native agents remains somewhat abstract. The AI economy is currently in a holding pattern, waiting for the build to catch up to the narrative.

While the term “AI agent” has taken many forms, embedding these agents into blockchain environments seems to carry a particular charge, amplifying expectations. Over the past six months, it has become clear that the combination of AI agents and Web3 has astronomical potential, but only if the sector can move past early speculation and build for long-term value for end-users, a vision shared by many. Early optimism was expressed by projects like Fetch in late 2023, who highlighted the immense opportunity for both companies and people. However, real use cases remained limited at that time, demonstrating that this is far from a passing trend. It’s a subject of ongoing extensive research, showing that AI agents can reconstruct how value is created and distributed across decentralized systems.

A handful of foundational projects—Giza, Axal, and Theoriq, among others—are architecting the primitives for agent-dedicated infrastructure in DeFi, each with a distinct approach. Giza is advancing verifiable on-chain inference through zero-knowledge machine learning, enabling agents to act with cryptographic accountability. Axal prioritizes execution integrity, developing systems for runtime verification and constraint enforcement. Theoriq, by contrast, explores decentralized intelligence through AI swarms—simulated collectives of agents coordinating within shared environments. This multidimensional approach addresses a growing issue in DeFi: the fragmentation of AI agents. Handling token swaps, yield strategies, or cross-chain bridging often operates in isolation, with little to no coordination between them. The result is a disjointed user experience that’s difficult to navigate and scale.

The proposed solution—Agentic DeFi—calls for intelligent agent swarms that can collaborate across tasks, chains, and user intents to deliver a unified experience. Theoriq’s model gestures toward this future. By exploring AI swarms, which are simulated collectives of agents that share data and goals, we can establish a core architecture for agent ecosystems that don’t just act independently but operate as synchronized systems. Although ambitious, these initiatives are still in their early stages. Very few are operating at high thresholds, but we can already spot a product-market fit, with accomplishments, such as Giza, being one example.

Notably, each agent framework is solving a different layer of the same problem. This reflects a maturing space, where builders are no longer racing to replicate but instead developing complementary solutions. All of these pieces must ultimately fit together to form a cohesive future. There’s a growing consensus that the bottleneck isn’t intelligence—it’s oftentimes efficient infrastructure. For agents to operate resourcefully within DeFi, they must plug into modular environments that allow them to execute safely, adapt intelligently, and remain accountable to human-defined constraints. What’s needed is a robust foundation of vault frameworks, risk engines, and liquidity systems—each enabling the agent to take actions with safeguards in place. Modules can define what agents are permitted to do with capital, just like risk modules help them assess uncertainty, and liquidity modules allow them to monitor the available liquidity and trigger redemptions if necessary.

The vision of agents running vaults, rebalancing portfolios, or participating in governance is achievable. And we’re getting there. But it won’t be reached through surface-level integrations or overpromised retail bots or memecoins. What you can take away from this is that agents don’t just need intelligence; they need infrastructure. Without DeFi frameworks for agents, dynamic risk controls, and composable liquidity tooling, the road might be rocky. They need interoperability, coordination, and modular environments designed to support dynamic, cross-functional behavior. That’s why differentiated approaches within AI agents matter. Giza’s verifiability layer, Axal’s runtime enforcement, and Theoriq’s coordinated swarms are not competing with each other. They’re complementary.