The consensus is clear: OpenAI has won. Codex and ChatGPT Work crossed 10 million weekly active users. A five-fold increase in a quarter. The narrative writes itself: AI agents are here, and they are scaling. But consensus is a lagging indicator. It ignores the cost of attention, the fragility of centralized platforms, and the structural flaws that will surface as these numbers compound.
Let me isolate the raw signal from an unverified source—a blockchain news aggregator citing an entity called 'Dongcha Beating.' The data: 10M weekly active users, a milestone tied to a gimmick of resetting usage caps. No technical breakdown. No revenue per user. No churn rate. This is the digital equivalent of a whitepaper promise in 2017. I audited over 200 of those. Ninety-five percent failed on tokenomics alone. The same skepticism applies here: verify the data before making allocation decisions.
From a macro perspective, the 10M figure is not about OpenAI. It is about the structural demand for autonomous agents. That demand is real. Every knowledge worker now expects AI to code, schedule, and negotiate on their behalf. But the infrastructure supporting these agents is a centralized honeypot. Code is law, but capital decides who writes it. Right now, one entity writes both the model and the rules. That concentration is the risk we ignore.
The real insight is hidden in plain sight: agents need trustless settlement.
In 2022, during the Terra-Luna collapse, I watched liquidity evaporate because the anchors were centralized. The same dynamic will unfold in the agent economy. As agents begin to execute transactions—paying for compute, buying data, negotiating contracts—they will demand a settlement layer that cannot be censored or front-run. That is where blockchain comes in.
Consider the oracle problem. In DeFi, oracle feed latency is the Achilles' heel. Chainlink's solution of 'decentralization through centralized node operators' is a joke. It works until it doesn't. For agents, the oracle is the gateway to real-world actions. If an agent relies on a centralized API for pricing or identity, the entire system inherits that node's failure modes. We need a verifiable, on-chain attestation layer for agent decisions.
Now, look at the compute side. Ten million weekly active users require an enormous inference fleet. That fleet is largely controlled by one cloud provider (Microsoft Azure) through one chip supplier (NVIDIA). That is a single point of failure—not just technical, but geopolitical. The next cycle will reward decentralized physical infrastructure networks (DePIN) that allow agents to route compute through distributed, permissionless nodes. We saw the early signs in 2023 with projects like Gensyn and Akash. The 2026 AI-agent economy framework we built at my fund anticipated this shift: agents will autonomously negotiate compute and data trades on-chain.
History doesn't repeat, but it rhymes. The rise of centralized exchanges led to FTX. The rise of centralized AI agents will lead to a similar reckoning. The market is betting on OpenAI's dominance, but the real value accrues to the infrastructure that enables non-custodial agent execution.
Volatility is the fee for admission to the future.
The current sideways market is the perfect environment to accumulate these infrastructure assets. Chop is for positioning. While others obsess over the 10M number, I am looking at the signals beneath: increasing developer activity on decentralized compute networks, a new generation of zero-knowledge proofs for agent verification, and composable identity standards like Verifiable Credentials on-chain.
My contrarian angle is this: the 10M figure will eventually be an anchor, not a tailwind. Once the first major exploit or regulatory clampdown hits the centralized agent platforms, capital will rotate into sovereign agent frameworks. The 2024 Bitcoin ETF onboarding taught me that institutions move slow until they don't. They will demand auditable, permissionless rails for their agent interactions.
The takeaway for cycle positioning is clear: forget the agent count. Focus on the stacks that allow agents to exist without asking permission. Decentralized oracle networks that actually resist censorship. Layer-2s that can handle microtransactions at agent scale. Identity protocols that give agents their own wallets and attestations. That is where the next wave of alpha resides.
Risk isn't a number; it's what you don't know. What you don't know is how centralized agency will fail. I'd rather own the insurance policy than the claim.
