The World Cup offers a practical scene for observing how wallets are evolving. Topics that football fans discuss every day, such as whether a team can qualify or how championship odds are changing, become tradable and priced behavior in prediction markets such as Polymarket. When mainstream Web3 wallets integrate World Cup prediction activities, the activity is not only about sports or entertainment. Over a longer timeline, it becomes a lightweight entry point for users to participate in on-chain interactions.
In this World Cup-related exploration, imToken has started to place AI Agents into real usage scenarios. Its agents on the web and in Discord are designed around specific prediction needs, helping users complete betting-related transactions in a more natural way. Users no longer have to operate only inside the wallet app. They can participate in prediction markets from Discord or a web page, while being guided back on-chain by the Agent when an on-chain step is required. This can be seen as an early form of the Agentic Wallet: the Web3 wallet of the future is not limited to a wallet application, but appears as a more widely distributed AI wallet experience.
From an Asset Gateway to an Intent Interpreter
For the past decade, the core questions that wallets addressed were relatively clear: where assets are stored, who controls the private key, and who signs the transaction. A user opening imToken was mainly checking balances, making transfers, interacting with protocols, or managing multi-chain assets. At that stage, the wallet functioned as an asset gateway and a signing gateway. As long as the user already knew what they planned to do, the wallet only needed to complete the final step.
The Agentic era changes that starting point. Users do not always know what to click at the beginning. In a World Cup scenario, an ordinary user may not first think, “I need to open Polymarket, find a specific market, evaluate the odds, and complete a trade.” The questions are more likely to be: “What is interesting about tonight’s match?” “I think Portugal will qualify; are there related markets?” “Is this price already too low?” “If I only want to participate with a small amount, what is the process?”
In the past, those questions were scattered across WeChat groups, social platforms, and search engines. Each user had to piece together the information and then execute the steps one by one. After an Agent becomes part of the flow, the interaction changes. The user only needs to express a general intent. The Agent decomposes the path, and the wallet turns that path into a series of on-chain actions.
This is not simply a matter of adding a chat box to a wallet. The deeper change is that the wallet begins to move from a “function menu” toward an “intent interpreter.” Traditional wallets mainly ask users to choose whether they want to transfer, swap, stake, or connect to a DApp. A future wallet moves closer to a natural-language entry point, where users describe what they want to accomplish and the wallet, together with the Agent, breaks the task into executable steps.
This is why a mass event such as the World Cup works well as an entry point for an Agentic Wallet. It already has a shared context, users naturally want to express opinions, and the activity requires judgment. The Agent does not need to begin by managing complex asset portfolios, especially because complex asset management carries higher risk. It can first help users find an interaction path within a specific scene, then return final control to the wallet and to the user.
Wallet Interfaces Spread Beyond the App
The web agent and Discord agent mentioned in imToken’s exploration are examples of wallet capabilities moving into lighter entry points. Users do not necessarily open the app first or enter a traditional DApp browser first. They can encounter the Agent on an activity page, in a World Cup context, or inside a community channel, and then find a path toward the relevant on-chain interaction. This means the boundary of the wallet is expanding outward.
Previously, the wallet entry point was relatively clear: open the app, enter the asset page, click a function, and connect to a DApp. In the next stage, wallet entry points can be distributed across web pages, Discord, Telegram, AI chat boxes, activity pages, developer tools, and even lightweight wallet interfaces generated by users themselves. From this angle, the key point is not World Cup prediction alone. The more important shift is that it allows the wallet to stand in front of user intent in a more natural way.
If viewed only from inside the crypto industry, Agentic Wallets can easily be treated as a narrative concept: AI helps users watch markets, find opportunities, and trade. That sounds like an extension of the previous AI Agent wave. But Mastercard’s launch of Agent Pay for Machines on June 10 placed the topic into a broader context. It made the issue no longer belong only to Web3.
Mastercard and WeChat Pay Bring Trusted Agents Into Payments
Mastercard’s definition of Agent Pay is clear: trusted AI Agents can participate in payments under user authorization. The design covers how an Agent is identified, how it is authorized, how it is verified within the payment network, and how merchants, issuers, and users know that a transaction was assisted by an Agent. These are highly similar to the questions that Web3 wallets face.
When AI only helps write copy, the cost of an error is usually manageable. Once AI begins to participate in asset interactions, the problem changes. Does it truly have permission? Has it accurately understood the user’s intent? Is the service it calls trustworthy? Does the transaction it initiates go beyond the permitted boundary? If the result is inconsistent with what the user expected, who handles the outcome? Mastercard’s answer is to redesign the identity, tokenization, authorization, risk control, and dispute handling of trusted Agents inside the payment network.
Domestic payment giants are also moving in the same direction. WeChat Pay is working with Tencent’s agent product WorkBuddy to test AI payment functions, including the launch of an “AI Exclusive Card” inside WeChat Wallet. Based on disclosed information, the core of such products is not to let AI spend money without limits. Instead, they set boundaries for agent payments through recharge limits, payment authorization scope, and password verification confirmation. This follows the same line of logic as Mastercard Agent Pay: AI can participate in payments, but it must be identified, authorized, limited, and audited.
On-Chain Wallets Need Stricter Boundaries
Web3 wallets face the on-chain version of the same problem. The difference is that traditional payment systems place more emphasis on networks, merchants, issuers, and compliance responsibilities, while on-chain wallets place more emphasis on private keys, signatures, approvals, contract calls, and user self-custody. For that reason, Agentic Wallets cannot simply copy the path of traditional payments. In traditional payments, users can rely on banks, card networks, merchants, dispute handling, and risk-control systems. In the on-chain world, once a transaction is on-chain, there is often no undo button.
The more efficiency an AI Agent brings, the more important the wallet becomes as the last safety boundary. Future wallets should not only allow Agents to “do things.” They also need to ensure that Agents can do only what users have permitted. This is why Web3 is a particularly suitable environment for discussing Agentic Wallets. When many people discuss AI Agents, they instinctively move toward extreme automation: AI automatically trades, manages wealth, finds airdrops, or arbitrages. That direction is attractive, but for a wallet, the hardest issue is not automation itself. The harder issue is boundary awareness.
A wallet is not an ordinary application. In an ordinary app, if AI recommends the wrong song, writes the wrong sentence, or clicks the wrong page, the result is mostly an experience problem. In a wallet, if AI misunderstands an instruction, calls the wrong contract, or grants excessive authorization, it can become a real asset risk. Therefore, the first-principles question for an Agentic Wallet is not “how much can AI do for the user,” but “how does the user know what the AI is doing?”
This is also why imToken has emphasized control over the past few years. From self-custody wallets to multi-chain asset management, and now to AI co-creation and Agent exploration, the continuous thread is not simply that there are more functions. The thread is that users should always be able to understand, confirm, and control their own digital world.
In the Agentic era, that principle becomes more concrete. Wallets need to let users understand who the Agent is, which capabilities it can call, how long the authorization lasts, whether it can operate across DApps, when another confirmation is required, and whether the user can pause or revoke it with one click. These questions may sound detailed, but they form the foundation for an Agentic Wallet to truly work. The power of an AI Agent comes from its ability to turn complex processes into simpler experiences. A single user sentence can be decomposed into more than ten execution steps. That improves usability, but it also creates a security challenge. The longer the path and the more intermediate steps there are, the more the wallet needs to bring key checkpoints back in front of the user. The World Cup prediction activity is an event-based entry point, while imToken’s web Agent and Discord Agent are community-based entry points. The shift toward Agentic Wallets has already started in these small scenes.