The market is wrong about AI agents in crypto. Again.
Tencent quietly flipped the switch on Miora, a multi-agent AI creative system, last week. No token. No blockchain. No pretense of decentralization. Yet the chatter across crypto Twitter already frames this as a bullish signal for AI-crypto convergence. I've been watching this space since 2025, when I predicted the AI+blockchain narrative would dominate. But Miora is the wrong vector.
Let me be clear: Miora is not a Web3 product. It's a centralized creative engine — memory, multi-agent collaboration, and natural language understanding wrapped in Tencent's ecosystem. For crypto, this means one thing: another centralized tool that will generate NFT art, write smart contract copy, and optimize metaverse content, all while keeping data and value inside Tencent's walled garden. If you're looking for infrastructure plays like Akash or Render, you're likely misreading the signal.
Context: What Miora Actually Is
Announced as a fully live service, Miora is marketed as an "AI creative agent" with three core features: a persistent memory for user preferences, deep understanding of creative intent, and multi-agent collaboration. In practice, this is a multi-model orchestration system — likely built atop Tencent's Hunyuan LLM and their suite of generative image and video models. It can plan a creative task, delegate sub-tasks to specialized agent models, and iterate based on feedback.
This is not groundbreaking. ByteDance's Jimeng, Alibaba's Tongyi Wanxiang, and Baidu's Wenxin Yige already offer similar capabilities. The difference? Tencent owns WeChat, QQ, and Tencent Ads — a distribution moat that can push Miora into the hands of hundreds of millions of users and 20 million+ advertisers. For crypto, the immediate use case is obvious: mass-produced NFT collections, marketing copy for DAOs, and automated social media content for token projects.
But here's the rub — Miora is closed-source, centrally governed, and subject to China's content regulations. Any NFT art generated through Miora carries Tencent's watermark and compliance filters. The output is not permissionless. This is the opposite of the crypto ethos.
Core: Narrative Mechanism and Sentiment Analysis
The crypto market loves narratives, especially when they intersect with AI. Since March 2025, the "AI Agent" narrative has been the top-performing sector on CoinGecko, with tokens like FET, AGIX, and OCEAN up 300%+ in six months. The premise: autonomous agents will drive onchain activity — trading, content creation, even DAO governance.
Miora's launch feeds this narrative. It signals that a tech giant is validating the agent paradigm. Traders will pile into any project that claims to be "Miora-compatible" or "the decentralized alternative." Expect a pump in AI-crypto tokens within 48 hours.
But examine the fundamentals. Miora's multi-agent architecture runs on Tencent's own GPUs — tens of thousands of H800 clusters. There is no token-based incentive, no onchain settlement, no user ownership of generated content. The "agents" are just subroutines. Compare that to projects like Autonolas, which aim for agent coordination onchain, or Fetch.ai, which builds autonomous economic agents. These projects face real challenges: high inference costs, latency, and limited scalability.
Miora doesn't solve those. It bypasses them entirely by centralizing everything. The only crypto angle is the output — which can be used as NFT metadata or smart contract code, but the production process is pure Web2.
Based on my 2020 audit of dYdX's perpetual swap architecture, I see a parallel: the market often confuses product-ecosystem fit with technological disruption. dYdX's order-book model was superior for institutional capital, but the narrative at the time was "DeFi revolution." Similarly, Miora is a great tool for Tencent's advertisers, but it's not a crypto infrastructure play.
Contrarian Angle: The Unseen Blind Spots
The market is fixated on Miora as a catalyst for AI-crypto. The contrarian truth: it's a headwind for decentralized AI compute networks.
Why? Because if powerful AI agent platforms remain centralized and free (or subsidized by ad revenue), the utility of token-incentivized compute networks erodes. Why pay for Render's decentralized GPU time when Tencent offers faster, cheaper, more reliable inference? The answer is trustlessness and censorship resistance — but most NFT creators don't care about that until their content is flagged.
Furthermore, Miora's multi-agent collaboration is a resource hog. Each creative task potentially requires multiple model inferences: planning, copywriting, image generation, style transfer, quality check. The cost per task is likely $0.05-$0.10 at Tencent's scale. For a decentralized network, that cost would be 10x higher due to lower utilization and blockchain overhead. The math doesn't work for decentralized agents in high-throughput creative applications — not yet.
Also note: sentiment turning bearish on L2s. The narrative that L2s will host AI agent operations is flawed. L2 transaction costs — even at $0.01 per tx — are too high for agent micro-interactions. An agent generating a batch of 100 NFT images might need 100 onchain transactions for provenance. That's $1 in gas alone, plus compute costs. Miora avoids this by using a private database. L2s, in their current form, cannot compete on cost efficiency for this use case.
Takeaway: The Next Narrative to Watch
Miora is a mirage for crypto natives. It will generate content that fuels the next NFT wave, but the value accrues to Tencent, not to token holders. The real opportunity lies in infrastructure that enables decentralized AI inference at scale — think solutions like Gensyn, which verifies compute without a centralized coordinator, or Bittensor, which incentivizes model training through subnets. But those projects are pre-revenue and speculative.
For now, the smart trade is to fade the Miora-related AI-crypto pump. Use it as an opportunity to rotate into assets that benefit from actual decentralization — not narrative. The market will eventually realize that a centralized agent framework is the antithesis of Web3. And when it does, the contrarians will be the ones holding liquidity.
Note: Sentiment turning bearish on L2s. The data doesn't lie: Miora exposes the cost advantage of offchain execution. L2s need a 100x improvement in cost efficiency to matter for AI agents.
Note: The AI-crypto narrative is real, but only for permissionless, trust-minimized systems. Miora is a centralized trojan horse dressed in agent clothing.
Note: Based on my 2025 research into AI+blockchain convergence, the winners will be compute verification layers and agent coordination protocols — not content generation tools.