Web3 data platform CARV is working on a new AI roadmap that goes beyond typical integrations like chatbots or content tools. The company says it wants to develop fully autonomous AI agents—what it calls “AI Beings”—that operate on-chain with their own memory, identity, and economic behaviour.
While most blockchain projects use AI to improve services like analytics or recommendations, CARV’s concept imagines software agents that act independently within blockchain systems. These agents could manage wallets, earn tokens, vote in governance decisions, and even reproduce or create new agents.
The idea is tied closely to blockchain’s features: verifiable identity, decentralised control, and open access. CARV argues that these qualities are essential for agents to act without relying on centralised systems. The company’s tech stack includes a custom SVM Chain for AI-specific transactions, a decentralised identity standard (ERC-7231), and a framework it calls D.A.T.A.
The plan is structured around a five-layer “AI Being Stack,” which includes execution infrastructure, communication protocols, identity tools, economic logic, and application support. The setup is ambitious—but it’s also still mostly theoretical.
A roadmap in three parts
CARV has laid out its plans in three phases: Genesis, Pulse, and Convergence.
The first phase focuses on launching wallet-connected AI agents in consumer apps. These agents will draw on user-consented data and store memory using secure computing techniques and cryptographic proofs. CARV says these apps will launch on Google Play and the App Store, though no specific release dates have been announced.
The second phase introduces on-chain learning. Agents will start to adapt their behaviour based on user actions, staking signals, and governance votes. These inputs are fed into a system of validator nodes that check data flows and agent behaviour. CARV suggests this step combines blockchain governance with reinforcement learning, but exactly how these models will be trained and evaluated isn’t clear.
The final phase imagines agent collaboration. In this stage, AI agents would interact with each other, share information, and coordinate across apps. CARV describes use cases where, for example, a health assistant talks to a finance bot. These interactions would be based on shared identity and reputation protocols and governed by on-chain logic.
Gaps between vision and reality
While CARV’s roadmap outlines a bold shift from “tools to agents,” many parts remain early or unproven. Blockchain and AI integrations have gained interest, but large-scale agent coordination—especially where agents act independently and interact with real-world assets—raises technical, ethical, and security questions.
Key challenges include:
- Complexity of agent behaviour: Training AI to act autonomously and safely in decentralised systems is still a developing field. Most projects rely on supervised models or narrow use cases.
- Security and misuse: Autonomous agents with control over wallets or governance rights may be vulnerable to manipulation or exploitation. Verifying their behaviour in an open system isn’t straightforward.
- User trust and adoption: It’s unclear how willing users will be to rely on AI agents that store memory on-chain or participate in financial decisions. CARV’s plan depends heavily on consent, transparency, and technical literacy.
- Lack of external validation: So far, most of the roadmap has come from CARV itself. There’s little third-party analysis or public feedback on whether this kind of agent system is viable at scale.
A shift in positioning
CARV’s roadmap also signals a shift in how it positions itself. Originally described as a data coordination platform for Web3, the company is now framing its infrastructure as a base layer for agent economies. This change aligns with broader industry interest in “AI x Web3” crossovers but also risks drifting into speculative territory.
Ambero Tu, CARV’s CTO, said in a statement that the goal is to build systems where AI agents and human users co-govern and co-own services. The language is aspirational, but the actual deployment timeline—and how CARV plans to manage risk—hasn’t been detailed.
For now, the company is in early deployment stages. If it can demonstrate working agents in real-world apps—and show how those agents improve utility without introducing new problems—it may attract wider attention. Until then, its broader goals remain mostly conceptual.
(Photo by Immo Wegmann)
See also: China could test Yuan stablecoin Hong Kong route but faces headwinds – Morgan Stanley
Want to learn more about blockchain from industry leaders? Check out Blockchain Expo taking place in Amsterdam, California and London.
Explore other upcoming enterprise technology events and webinars powered by TechForge here.
