2017’s dream is today’s regulation.
Tencent just flipped the switch on Miora, an “AI creative agent” boasting memory, demand comprehension, and multi-agent collaboration. The announcement landed with the subtlety of a marketing release: three bullet points, no benchmarks, no pricing. But for anyone tracing the convergence of AI and crypto, this is not a product launch—it’s a stress test. Miora is a closed-source, centrally orchestrated system designed to serve Tencent’s advertising empire. It will generate banners, write slogans, and even draft video scripts—all within the Great Firewall. And it does so without a single on-chain transaction, without a verifiable token economy, and without any user-owned data custody.
This is exactly the kind of centralized chokehold that the crypto-native AI movement was built to break. Yet the market reaction—silence from DeFi Twitter, a flicker of interest from AI tokens—betrays a deeper confusion. We are still treating AI agents as isolated software, not as the future backbone of autonomous economic activity. Miora is a wake-up call: if crypto does not build the rails for machine-to-machine commerce, centralized giants like Tencent will fill the void with their own proprietary systems.
The core insight is not about Miora’s technology—it’s about what it reveals about crypto’s missing infrastructure.
Let’s start with the memory claim. Miora purportedly remembers user preferences across sessions. In a centralized database, that’s trivial. But in a trustless environment, memory requires persistent, verifiable, and portable storage—vectors, not just logs. The crypto space has tools like Ceramic Network or Tableland, but no standardized agent memory layer. Every DeFi bot today either runs stateless or stores memory in a centralized Redis cache. That’s not sovereignty; that’s a security breach waiting to happen. Miora’s multi-agent collaboration is even more revealing. Tencent likely uses a coordinator agent (plan-then-execute loop) that delegates tasks to specialized sub-agents—image generator, text writer, compliance checker. This mirrors the architecture of projects like AutoGPT or MetaGPT, but with one key difference: Tencent’s agents are not autonomous. They are controlled by a single corporate entity, with no ability to transact, negotiate, or settle payments independently.
Here is the contrarian angle: Miora’s launch actually validates the crypto thesis—and it does so by exposing the dead end of centralization.
Decentralized agent frameworks (e.g., Olas, Fetch.ai, Autonolas) have struggled to gain traction because they lack immediate utility. Tencent just provided that utility: a billion-dollar advertising market that desperately needs AI-generated content. But the centralized version is a trap. Tencent owns the agents, the data, and the monetization pipeline. Users get convenience; Tencent gets GDP-level returns. Compare that to a hypothetical decentralized agent network where creative agents compete for tasks, settle payments in stablecoins, and store reputation on-chain. The cost for a brand to generate one banner might drop from $50 (agency) to $0.50 (AI agent), with the agent pocketing $0.40 and the network fee $0.10. That is not just cheaper—it is a new asset class: agent productivity tokens.
The problem is that crypto is not ready for this volume. Miora’s deployment targets millions of small-to-medium businesses actively using WeChat and Tencent Ads. If those businesses were to switch to a decentralized alternative tomorrow, the entire current Layer 2 ecosystem would buckle under the transaction volume. A single Miora instance might process 10,000 creative tasks per second—each potentially requiring multiple on-chain microtransactions for agent collaboration, memory storage, and payment finality. Ethereum, even with Layer 2, handles roughly 1,500 TPS on Optimism and 2,000 TPS on Arbitrum combined. Solana peaks at 5,000 TPS in production. The infrastructure gap is not a decimal—it’s an order of magnitude.
This is where my experience as a CBDC researcher comes into play. During the Fed’s stress tests on our digital dollar prototype, we simulated 10,000 TPS with zero-knowledge privacy. That is exactly the throughput an entire ecosystem of AI agents would require. And yet the crypto community is still debating whether to scale via rollups, validiums, or channels, while centralized players ship products that will create the demand for that scale. If crypto does not deliver a production-ready, private, high-throughput agent transaction layer within the next 12 months, Tencent will not wait. They will deploy their own digital payment rails (already have WeChat Pay) and turn Miora into a closed-loop economic engine—no crypto required.
The regulatory angle is even sharper. Miora operates under China’s Generative AI regulations, which require content audits, watermarking, and copyright checks. This is not a bug; it is a feature. As a regulatory opportunity, Tencent is framing Miora as compliant by design. Crypto’s decentralized agents, by contrast, face legal gray zones: who is responsible if an autonomous agent generates defamatory content? Who pays the fine? Without a clear legal framework for agent liability, institutions will hesitate to adopt decentralized versions. The contrarian play is to recognize that regulation is not an enemy—it is a forcing function. If crypto can build agent frameworks that are compliant by default (e.g., on-chain identity for agents, auditable memory, content moderation via verifiable computation), they can leapfrog centralized offerings that are still shackled to human oversight.
The takeaway is stark: Miora is not a product to ignore or to fear—it is a signal to act.
We are witnessing the first major deployment of a multi-agent system with real economic activity (advertising). The centralized version will succeed in the short term because it has infrastructure, compliance, and distribution. But the long-term winner will be the network that allows these agents to transact autonomously, settle cross-border micro-payments, and retain ownership of their own memory and reputation. Crypto has the architectural advantage; it just lacks the execution speed. Every day that goes by without a scalable agent-to-agent payment layer, Tencent and its peers solidify their moat.
So ask yourself: in 2027, when AI agents account for 40% of all digital content production, will they be running on a permissioned WeChat server or on a permissionless blockchain? The answer depends on what we build today.
