The AI narrative continues to gain momentum, yet the recent performance of related stocks varies dramatically—some are basking in the peak of summer, while others are frozen in the depths of winter. Circle—the issuer of USDC, the largest compliant stablecoin—epitomizes the former; Tencent epitomizes the latter.

Circle’s price surge stems from multiple factors: for instance, its prior drop—from $300 down to $49—provided ample room for correction and built strong rebound momentum. Additionally, USDC’s market size recently rebounded, hitting a new all-time high. But the biggest catalyst may well be the explosive popularity of OpenCLAW, which has left people awestruck by the wave of Agents. This tangible, visceral impact has ignited demand for stablecoins perceived as especially suitable for Agent settlements—and Circle stands today as the only pure-play stablecoin issuer among global public companies.

Tencent’s decline, meanwhile, is more intriguing. Previously criticized for moving too slowly in the AI era, its stock had been in sustained correction from a high of HK$680. Even after releasing a stellar earnings report and announcing at its earnings call that it would reduce share buybacks while increasing AI investment, the stock fell further—not rose. Market concerns about Tencent’s AI spending have thus shifted from “insufficient” to “excessive.”

So in the AI era, who offers greater certainty: Tencent or Circle? Charlie Munger once said investors shouldn’t tackle hard problems. His exact words were: “One of our secrets is that we don’t try to know too much… I constantly assign things to the ‘too hard’ category. Occasionally, an easy decision comes along—and I just do it. That’s my system: most things go into ‘too hard,’ and only the rare, glaringly obvious, simple decisions get acted on immediately.” So—between Tencent and Circle, which one is the “easy problem” on our investment desk, and which is the “hard problem”?

Let’s start with Tencent. Currently, Yuanbao AI inside WeChat serves as an experimental prototype of WeChat’s future native Agent. It already handles most of the highest-frequency AI tasks ordinary users perform—namely, information retrieval and Q&A. However, Yuanbao today interacts only with content-type information within WeChat—text, images, WeChat Channels (video), and Official Accounts. At present, it cannot directly interact with Mini Programs.

Yet predicting the general shape of WeChat’s native Agent isn’t difficult: via natural-language dialogue, users will be able to directly instruct WeChat’s Agent to invoke Mini Programs and execute diverse tasks—including ordering food delivery, booking hotels, summarizing videos and voice recordings, setting calendar reminders, and generating work reports. Mini Programs themselves constitute a massive application ecosystem: most apps we use daily already have Mini Program versions—Didi Chuxing, Meituan, Pinduoduo, McDonald’s, Huolala, WPS—even Alibaba-affiliated Xianyu and Ele.me.

More importantly: WeChat’s Agent holds system-level authority over Mini Programs—akin to (or even stronger than) Apple’s control over iOS app ecosystems. Mini Program developers must adhere to WeChat’s specifications. Unlike DouBao’s mobile assistant—which gets outright blocked by major apps and locked out behind walled gardens—WeChat’s Agent will integrate systemically with Mini Programs. And Mini Program developers across the board will inevitably embrace WeChat’s official Agent to deliver better user experiences.

Compared to OpenCLAW—still hot but torturing early adopters to the point of despair, driving countless ordinary users from “first-time trial” straight to “giving up”—WeChat’s Agent, targeting 1.4 billion users, promises far superior usability and practicality. Its rollout would mark the true dawn of “mass-market Agent adoption.” This form may not feel particularly geeky or flashy—but it’s precisely the kind of product 90% of the population can actually use. The launch of such an Agent is highly probable—perhaps even inevitable—at some point in the future. Yet until it happens—and until markets see it with their own eyes—Tencent’s share price keeps falling. Under these conditions, “Mr. Market” grudgingly refuses to price in this future.

But for value investors, “Mr. Market” is there to be exploited. We invest in events that are highly probable—or even certain—to occur, yet haven’t happened yet. Amid the profound uncertainty of the AI era, is Tencent’s operational certainty really that low? I think it’s fine. So—is buying/holding Tencent right now an “easy problem” or a “hard problem”? I believe it’s far simpler than market sentiment suggests.

Now, Circle: Some things, viewed over the long term, reveal only fuzzy certainty. Will Circle’s business scale smoothly expand to hundreds of billions—or even trillions—of dollars? Bullish voices say yes. Their logic runs like this:
1. AI is the future; Agents will become the internet’s primary users and dominate massive online transaction volumes.
2. AI transactions will require blockchain-based stablecoin settlement.
3. USDC is currently the largest compliant stablecoin; AI-driven settlement demand will directly boost USDC’s scale.
4. Stablecoins exhibit strong network effects—scale begets self-reinforcing growth.
5. USDC’s scale will grow in lockstep with AI transaction volume—to several trillion dollars or more.

This chain of reasoning appears tightly woven—but several critical links harbor substantial uncertainty:
1. Will Agents truly become the internet’s dominant users and drive enormous transaction volumes? Over the long term, this claim’s certainty is overstated.
2. Will AI transactions require stablecoins? Not necessarily. While everyone sees #1, existing payment infrastructure providers—and internet/AI giants holding user bases—are also pushing their own standards, many of which are not blockchain-based. For example, Google is promoting the UCP protocol for agentic commerce.
3. Is USDC the largest compliant stablecoin? Yes—for now. But Tether holds formidable influence and is actively pursuing compliance; its political resources are also robust. Moreover, numerous ambitious traditional finance groups—including PayPal—are poised to enter the space.
4. Do stablecoins have strong network effects? In the face of frictionless, ruthless Agents, stablecoin differentiation will shrink dramatically. As long as they’re all viable for payments, Agents will select the option with lowest friction and fees—whether it’s USD1, USDT, USDC, or a bank-issued stablecoin. Does USDC currently enjoy stronger merchant-side resources? Visa, Stripe, and other payment gateways command merchant networks and distribution channels vastly larger than Circle’s—and they can equally enable other stablecoins. As long as settlement is convenient and frictionless, merchants won’t care which stablecoin they receive. Thus, stablecoin network effects may prove surprisingly weak in an Agent-dominated payments era.
5. Will USDC’s scale grow in lockstep with AI transaction volume—to several trillion dollars or more? When assumptions #2 through #4 each carry high uncertainty, stacking them together to derive conclusion #5 renders the final outcome deeply ambiguous.

So amid the profound uncertainty of the AI era—is Circle’s operational certainty really that high? I think it’s fine. So—is buying/holding Circle right now an “easy problem” or a “hard problem”? I believe it’s far harder than market sentiment implies. This is merely one person’s view—offered for reference only.

[Alex Xu]