Over the past year, AI + Crypto has become one of the most talked-about narratives in the crypto industry.
At first, the market focused more on AI tokens, decentralized compute, data networks, and various tokens built around the AI Agent narrative. But as we move into 2026, projects are no longer just discussing “how AI will change crypto.” They are beginning to embed AI directly into wallets, exchanges, payment protocols, and on-chain execution flows.
Within just one week, three landmark developments took place:
- On April 24, Binance Wallet launched Agentic Wallet, an independent keyless wallet designed specifically for AI Agents. It features a separate balance, configurable permissions, real-time monitoring, and includes Agentic Wallet Skills out of the box, allowing agents to perform automated actions within predefined boundaries.
- On April 28, TON introduced Agentic Wallet, a self-custodial wallet contract built for AI Agents. Users can set up an independent wallet for an AI Agent within minutes, allowing the agent to autonomously perform on-chain actions such as transfers, swaps, and staking without ever touching the user’s private key.
- On April 29, OKX released Agent Payments Protocol, an open payment standard designed for AI Agent business activities. It covers the full business flow, including quotation, negotiation, escrow, metering, settlement, and dispute resolution, aiming to provide a practical payment and settlement framework for the future agent economy.
On the surface, this looks like a race among major players to let AI take over on-chain execution. But if we zoom out, the real thread running through this competition is about permissions, reuse, and co-creation.
This may also be one of the most important changes Web3 wallets will face in the next decade.
1. The Evolution of AI + Crypto
To understand this shift, we can first look back at how AI + Crypto has evolved over the past few years.
For most everyday users, the biggest challenge in the on-chain world has never been a lack of information. It is the overwhelming amount of fragmented information. A token may appear across 𝕏, Telegram, DEX Screener, block explorers, project documentation, and KOL discussions at the same time. Most users simply do not have enough time to identify where the potential upside and risks are hidden.
That is why many of the earliest AI features launched by wallets and trading platforms focused on information interpretation.
Binance Wallet’s AI features introduced in January 2026 are a typical example. Social Hype ranks token popularity based on social attention and engagement data. Topic Rush turns emerging narratives into topic cards and categorizes them by stages of capital inflow. AI Assistant provides summaries of narratives, sentiment, and event timelines on token pages.
The value of these features is not that they make decisions for users. Rather, they help users lower the barrier to understanding. For new on-chain users, a clear AI-generated summary may be more useful than dozens of fragmented tweets. Even for experienced users, AI can serve as an information filter, helping them identify leads worth further research more quickly.
But this is only the first step.
In crypto, once users understand something, they still need to figure out how to act on it: how to grant approvals, bridge assets, set slippage, manage risk, revoke permissions, identify phishing pages, and so on. In other words, if AI remains limited to “summarizing information,” it still cannot truly enter the core flow of on-chain interaction.
So if the first phase of AI was more like a research assistant, the second phase of AI is moving closer to an executor. Further reading: “A Passport to the AI Agent Era: Why Ethereum is Betting Big on ERC-8004”
Gemini’s Agentic Trading is a representative signal of this shift. According to Gemini, users can connect AI Agents to trading accounts through MCP, allowing agents to call trading-related APIs and perform actions such as reading market data, placing orders, monitoring markets, and managing risk. Gemini also packages certain trading capabilities as Trading Skills, such as retrieving real-time market prices, checking bid-ask spreads, and reading candlestick data, so that agents can call these modules just like tools.
This shows that exchanges are rethinking how users will interact with trading systems in the future. The flow may no longer be limited to clicking buttons, entering prices, and confirming orders. Instead, users may describe their goals in natural language, while AI helps break down the path, call the right tools, and complete execution.
Similar changes are also happening at the wallet layer.
Cobo’s Agentic Wallet introduces the Pact protocol to define what an agent can do, what it cannot do, and under what conditions it should stop. It also uses a recipe-driven skill layer to provide agents with predefined execution paths, such as lending, swaps, DCA, grid strategies, and micropayments.
Coinbase is also developing Agentic Wallets and Agent Skills for AI Agents, combining them with machine payment protocols such as x402. This allows agents to send, trade, earn yield, and even automatically pay for resources such as APIs, compute, data, and storage within predefined limits.
At the core, crypto is providing AI with financial and account infrastructure it can call directly.
If we place these developments on the same map, we can see that they are structurally converging. Almost every player is assembling the same four building blocks:
- Identity layer: the on-chain identity and identifiability of agents, such as OKX’s Agentic Wallet and Coinbase’s agent identity.
- Permission layer: the granularity and duration of user authorization for agents, such as MetaMask’s ERC-7715 advanced permissions and Coinbase’s security guardrails.
- Payment layer: agent-based micropayments, subscriptions, and settlement, such as x402, APP, and MPP.
- Execution layer: agents placing orders, bridging assets, and calling contracts across exchanges and blockchains, such as Gemini Agentic Trading and TON Agentic Wallet.
In the traditional internet, AI Agents can help users search, summarize, write emails, and generate code. But once they enter payments, authorization, and asset operations, they immediately run into questions of identity, trust, accounts, and settlement.
Blockchain, by contrast, naturally provides accounts, signatures, assets, payments, contracts, and traceable records. This gives AI Agents the opportunity to evolve from “digital assistants” into “on-chain executors.”
But execution capability is only the beginning.
If this infrastructure is giving agents “hands,” then another more important and easily overlooked question is: where do agents get their workflows?
2. AI Agents Need More Than Wallets — They Need Reusable Skills
This is why the concept of Skills is worth paying attention to.
How should we understand a Skill?
A Skill is essentially a capability unit that packages a specific workflow. For example, each Skill can be seen as a folder containing a SKILL.md instruction file. This file tells the AI what the Skill does, when it should be used, and how to perform the task.
When systems such as Claude run, they can load the relevant Skill based on the user’s intent. A general-purpose model can then become an agent specialized in a specific task.
This design may appear simple—even surprisingly low-tech. But its real significance is that Skills mean the competition in the agent ecosystem will no longer be limited to model parameters, reasoning capabilities, and tool-calling speed. It will also become a competition over workflows, community knowledge, and reusable capabilities.
The moat of an agent platform may no longer be simply: “My model is bigger, faster, and smarter.” It may become: “Can experts, enterprises, developers, and everyday users on my platform turn their experience into reusable assets that other users and other agents can call again and again?”
When this logic is applied to the wallet industry, it leads to an even more interesting proposition: after wallets enter the agent era, the next phase of competition will not only be about making execution work well. Wallets also need to become workbenches where users can accumulate methods and co-create.
In fact, Gemini’s Trading Skills, Coinbase’s Agent Skills, and Cobo’s recipe-driven skill layer are all doing essentially the same thing: breaking complex operations down into standardized, modular, callable capability units.
This is especially important for everyday users, because on-chain operations are never just a simple instruction.
A seemingly ordinary swap may involve token identification, contract address verification, allowance management, slippage settings, gas estimation, phishing risk detection, and transaction confirmation.
A cross-chain transfer may involve choosing a bridge, estimating arrival time, comparing fees, checking whether assets are displayed on the target chain, and handling failures.
Participating in DeFi may involve understanding yield sources, smart contract risks, liquidation risks, and exit paths.
If these experiences only exist in someone’s mind, they are hard to pass on. If they are only written into a standard tutorial, they are hard to call automatically during real operations. But if they can be turned into Skills, templates, checklists, or interactive works, they may become reusable knowledge assets for every user in the AI era.
This is the new value of wallets in the AI era.
Looking ahead, what truly matters in the wallet space may not be simply whether “AI can help me complete an operation.” The more important question is whether the experience behind those operations can be accumulated, reused, and improved together.
This may also be a direction imToken can further explore through UI 3.0.
3. The Next Decade: A Wallet Vision for the AI Era
Over the past decade, the wallet industry has gone through multiple rounds of iteration: from mnemonic phrases to multi-chain asset management, from DeFi to NFTs, from L2s to account abstraction, and now to AI Agents.
But no matter how the industry narrative changes, wallets have always faced the same fundamental question: how can users manage their on-chain lives more safely, more autonomously, and more clearly?
In the AI era, this question becomes more complex — and more important.
From this perspective, imToken’s thinking around Verifiable UI over the past year establishes the first safety baseline for wallets in the AI era. Further reading: “From Kelp DAO to Verifiable UI: The Next Security Baseline for Decentralization.”
This is a critical and somewhat counterintuitive judgment.
In most industry discussions, AI usually means more natural input, smarter interpretation, and smoother human-machine interaction. But from the perspective of wallets, the smarter AI becomes, the more verifiable the interface needs to be.
Verifiable UI aims to solve a fundamental problem: users must be able to verify the authenticity of the interface itself. Even if a centralized service goes offline, and even if an agent’s interpretation is wrong, users should still retain final control over their assets and interactions.
This is rooted in the same principle imToken has upheld for years: non-custody.
The non-custodial principle answers the question of who owns the private key. Verifiable UI takes this one step further and answers the question: “Is what I see truly what I am about to sign?”
It is under this baseline that imToken believes “using deterministic rules to constrain probabilistic systems” is a more suitable way to think about wallets in the AI era.
No matter how smart AI becomes or how powerful a model is, whether a transaction can ultimately be executed must still be constrained by the deterministic code, rules, and permission boundaries inside the wallet.
The wallet should not become a peripheral entry point for AI. It should be the final checkpoint for user asset security in the AI era.
But beyond defense, UI 3.0 also has another more proactive direction: moving from tool to co-creation platform.
As execution is gradually taken over by agents, and as interfaces become verifiable, the next question wallets must answer becomes: what can users still create here?
The logic behind this is structurally similar to the logic of Skills.
When more and more wallets integrate agents, design permission systems, and provide execution capabilities, long-term differentiation may not come from “who added AI first.” It may come from whether users can accumulate their own creations or reusable artifacts inside the wallet — such as an on-chain strategy, a security rule template, a wallet application for a specific scenario, a shareable operation checklist, or a Skill that other users and agents may call in the future.
Looking back at the past decade, wallets evolved from “address books + signers” to “DApp browsers,” and then to “multi-chain portals.”
In the next decade, the question they need to answer may be: when agents execute on my behalf, and when I can verify the interface myself, what else can I co-create here?
When users begin letting AI participate in on-chain operations, wallets need to answer more than “Can this transaction be signed?” They also need to answer: What does this transaction mean? Does it match the user’s real intent? Is the authorization scope reasonable? Have the risks been explained clearly enough? Can users preserve the experience from this operation and share it with those who come later?
This is also the new direction UI 3.0 can convey: the wallet interface of the future should not only be more polished and smoother. It should understand users better, explain risks more clearly, capture knowledge more effectively, and be better suited for user co-creation.
This means users are not only here to share stories. They can also build and refine reusable creations together.
For example, this could be a Skill on “how to safely add a token,” an interactive guide for “using Layer 2 for the first time,” a risk checklist for “identifying phishing approvals,” a workflow template for “hardware wallet cold storage,” or even a beginner-friendly wallet learning path.
These creations do not need to be complex from the beginning.
They can be an image, a tutorial, a checklist, a security reminder template, or a prototype Skill that AI Agents may be able to call in the future.
What matters is that users are no longer just learners being taught. They become co-creators.
This may be one of the most promising changes in user education in the AI era: knowledge is no longer delivered one-way by project teams. Instead, it is continuously supplemented, corrected, verified, and reused by real users in real scenarios.
Conclusion
Looking back, the combination of AI and crypto is quickly moving from concept to product.
Exchanges are beginning to let AI access accounts and trading capabilities. Wallets are beginning to design asset permissions and execution boundaries for agents. Payment protocols are exploring automatic settlement between machines. Skills and recipes are turning complex operations into reusable modules.
These changes may not ultimately point to a world where “AI does everything.”
Instead, they may point to a world where user capabilities are expanded.
In this world, the wallet remains the entry point. It is no longer just an asset gateway, but an entry point for identity, authorization, knowledge, operations, and co-creation.
For imToken, the tenth anniversary is not only a moment to look back on the on-chain journey it has shared with users over the past decade. It can also become an invitation to the future: an invitation for every user to preserve their experience, questions, methods, and creations, and to help build the wallet knowledge network of the AI era together.
After all, a truly vibrant agent ecosystem does not compete only on models. It competes on the co-creating community behind it.
And wallets may be the most natural starting point for that community — and the path we will walk together in the next decade.