Mesh’s CTO says an AI agent can’t act without a funded wallet. Trust Wallet splits its approach: copilot for consumers, autonomous agent kit for developers.
At Consensus Miami on Thursday, executives from Trust Wallet and Mesh laid out a thesis that is quietly rewiring crypto infrastructure: wallets are being rebuilt for AI agents, not just humans. The shift is not a distant roadmap item. It is a response to what Mesh CTO Arjun Mukherjee called the cold-start problem, a constraint that makes autonomous software useless until it can hold value and sign transactions on-chain.
“An agent can’t do anything until it has a wallet funded,” Mukherjee said. “It’s very difficult for the agent to act until it has a wallet to do something, and it has value to transact with. And suddenly, enter crypto. Crypto has found its kind of niche, its killer app.”
The statement is more than conference-stage optimism. It signals a functional marriage between two technologies that have largely operated in parallel. AI agents can reason, plan, and execute multi-step tasks, but they lack a native financial layer. Blockchains provide programmable value transfer, identity, and settlement. The wallet becomes the bridge, and the race to build agent-compatible wallets is now a competitive differentiator for crypto infrastructure firms.
The naive market read treats AI agents as just another user type that will eventually adopt existing wallets. That framing misses the mechanical constraint. A human can open a wallet, fund it from a bank account, and manually approve transactions. An agent cannot. It arrives with zero balance, no credit history, and no way to bootstrap its own activity. Without a pre-funded wallet and permission to spend, the agent is inert.
Mukherjee’s cold-start framing is precise. The agent needs three things simultaneously: a wallet address, a balance, and the ability to route payments across chains, networks, and tokens. Mesh’s product response, called Smart Funding, is a connectivity layer that handles that routing for both human and agent users. It abstracts away the fragmentation of exchanges, wallets, smart contracts, and decentralized exchanges so that an agent can receive a funding instruction and execute it without manual bridging or token selection.
For traders, the implication is that agent-driven flow will not look like retail flow. It will be programmatic, multi-leg, and cross-chain by default. Wallets that cannot handle that routing complexity will be invisible to the agent economy. The infrastructure layer that solves the cold-start problem–funding, routing, and settlement–becomes the gatekeeper for a new category of on-chain volume.
Trust Wallet CEO Felix Fan described a deliberately split approach. On the consumer-facing crypto app, where users hold their own keys, the agent acts as a copilot. It simplifies navigation, explains on-chain actions, and speeds up processes, but it never takes custodial control. Every step requires explicit user consent.
“Users always hold the keys and all these permissions. Every single step, they need to give consent,” Fan said. The agent’s role is to “speed up the process and also help them to better understand how to navigate on-chain.”
That is the safe, incremental path. It reduces friction without introducing new trust assumptions. The more aggressive path sits on the developer side. Trust Wallet recently launched an agent kit that lets autonomous agents make trades, transfers, and other on-chain actions without human approval. The kit is paired with the implementation of EIP-8004, an Ethereum proposal that gives agents an on-chain identity and a credit-style score.
The bifurcation is not just a product decision. It is a risk-management architecture. The consumer side preserves human custody and consent, limiting liability and regulatory exposure. The developer side embraces full autonomy but ties it to an identity primitive that makes agent behavior auditable and, potentially, reputationally constrained. A trader evaluating which wallet infrastructure will capture agent flow should watch whether EIP-8004 gains adoption across other wallet providers. If it does, it creates a standard for agent accountability that could become a prerequisite for institutional deployment.
When an autonomous agent makes a bad trade, drains a wallet, or violates a sanctions rule, who is liable? The panelists gave answers that reveal how differently crypto-native firms are thinking about this compared to traditional finance.
Mukherjee said Mesh is wary of importing traditional finance’s friction into agent payments. “AI should augment human judgment, not replace human responsibility or accountability,” he said, adding that responsibility for an agent’s actions sits with the institution that deploys it.
That is a clean, hard-line position. The deploying institution–whether a trading firm, a DAO, or an AI lab–owns the risk. The wallet infrastructure provider does not intermediate the liability. This is the opposite of how banks and payment networks handle fraud and errors, where intermediaries absorb and redistribute losses through chargebacks, insurance, and compliance overhead. In the agent-wallet model, the infrastructure is a rail, not a guarantor.
For market participants, this has two consequences. First, agent-driven strategies will require explicit risk parameters set by the deployer, not assumed by the wallet. Second, the absence of intermediary liability means agent activity will likely concentrate in venues and protocols that offer on-chain enforcement of rules–smart contract escrows, reputation staking, and slashing conditions–rather than legal recourse. The infrastructure that supports those enforcement mechanisms will capture volume that cannot exist in a chargeback-dependent system.
Both panelists said they expect major AI labs to launch their own wallets. Fan noted that X has already been vocal about X Money, and “Grok will very likely have a wallet within.” He added that “Claude and all these players, they can run on-chain maybe just tomorrow. So we are open for that challenge.”
Mukherjee said Mesh’s strategy is to remain agnostic across wallets, networks, and tokens. “If there’s Web3-based e-commerce on any network, on any token, and any connected funds, we all win,” he said.
This is the competitive landscape that matters. If AI labs launch their own wallets, they will control the user interface and the agent’s default financial layer. That threatens standalone wallet providers unless those providers become the invisible plumbing that AI labs plug into. Mesh’s agnostic connectivity play and Trust Wallet’s agent kit are both bets on becoming that plumbing. The trade for investors watching this space is not which wallet wins the consumer app battle, but which infrastructure layer becomes the standard for agent funding, routing, and identity.
The discussion comes as BlackRock filed paperwork to expand its tokenized fund lineup, with real-world assets growing 200% year over year. That parallel development deepens the infrastructure need for agent-ready wallets. Tokenized funds require on-chain settlement, and agents that can autonomously allocate into those funds need wallets that can handle institutional-grade compliance checks, identity verification, and multi-chain settlement. The cold-start problem Mukherjee described is not just a retail agent issue; it is the same problem an institutional agent faces when it needs to deploy capital into a tokenized money market fund.
The practical watch item for traders and infrastructure investors is the adoption curve of EIP-8004. If the proposal gains traction beyond Trust Wallet, it creates a standard for agent identity that could become a prerequisite for any wallet that wants to serve autonomous agents. The second catalyst is the first major AI lab wallet launch. When Grok, Claude, or another large model provider ships a native wallet, it will test whether the agnostic infrastructure thesis holds. If the lab builds on top of existing connectivity layers like Mesh, the plumbing play is validated. If the lab builds a vertically integrated stack that cuts out third-party wallet providers, the standalone wallet thesis weakens.
For now, the signal from Consensus Miami is clear: the wallet rebuild is not a speculative narrative. It is an engineering response to a specific, solvable constraint. The cold-start problem is real, and crypto’s programmable value layer is the only solution that does not require an agent to have a bank account, a credit score, and a human sponsor. That is a structural advantage that will pull agent activity on-chain, and the wallets that solve funding, routing, and identity first will capture the flow.
Drafted by the AlphaScala research model and grounded in primary market data – live prices, fundamentals, SEC filings, hedge-fund holdings, and insider activity. Each story is checked against AlphaScala publishing rules before release. Educational coverage, not personalized advice.