OpenAI just dropped a quiet bomb: Codex and ChatGPT Work have crossed 10 million weekly active users. The headlines celebrate the growth — a 5x leap from 2 million in weeks. But as a cross-border payment researcher who has spent years tracing the quiet resilience beneath the market, I see something else: the most urgent infrastructure challenge of the next decade.
These 10 million users are not just chatting. They are programming, drafting contracts, coordinating logistics, and soon — sending value. Every time an AI agent books a flight, orders inventory, or pays a freelancer, it needs a payment rail. Today, those rails are centralized bank accounts and credit cards. Tomorrow, they must be programmable, trustless, and global.
Tracing the quiet resilience beneath the market, the crypto ecosystem has been building these rails for years. But the arrival of 10 million agent users changes the calculus. We are no longer debating if agents will need crypto — we are debating if crypto is ready.
Context: The Global Liquidity Map and the Agent Economy
The macro context is simple: global liquidity is shifting from human-driven transactions to machine-driven micro-payments. In 2026, my own research on AI-agent payment integration showed that a single B2B cross-border transaction can be broken into dozens of micro-settlements between autonomous agents. Traditional correspondent banking — with its 3–5 day settlement windows and $30–50 fees per wire — is economically impossible for micro-transactions.
OpenAI's milestone is proof that agents are already embedded in workflows. Codex writes code that compiles. ChatGPT Work drafts memos that get signed. The next step is natural: agents will need to pay for APIs, data access, compute time, and services from other agents. The global payment infrastructure for this economy does not exist yet.
The crypto industry has been attempting to fill this gap. Stellar and Ripple (which I audited in 2018) offer fast cross-border settlement. Bitcoin, post-ETF approval, has become Wall Street's toy — its settlement is too slow and costly for micro-payments. Ethereum's L2 ecosystem has dozens of rollups, but they are slicing already-scarce liquidity into fragments. The user base is the same small group of speculators, not the 10 million agents.
Core: Why Payment Rails Are the Real Bottleneck
Let me ground this in data. Based on my experience in the 2022 bear market bridge preservation, I audited cross-chain bridges that were supposed to enable seamless value transfer. The reality: most bridges had insufficient liquidity reserves for mass withdrawals. When Terra collapsed, three major bridges nearly failed because they could not handle the surge. The lesson was clear: infrastructure must be stress-tested before adoption, not after.
Now apply that to the agent economy. Assume each of the 10 million agents makes just 10 micro-transactions per week — a conservative estimate for a code agent requesting API calls or a work agent purchasing cloud credits. That is 100 million weekly transactions. Existing blockchain networks struggle to handle that volume without congestion or high fees. Solana can process thousands of transactions per second, but its recent outages show fragility. Ethereum L2s are improving but still face fragmentation: an agent on Arbitrum cannot easily pay an agent on Optimism without bridging, which adds latency and cost.
During my 2020 DeFi yield safety investigation, I reverse-engineered Compound's governance interface and found that rapid innovation often comes at the cost of user protection. The same is true here. The race to build agent-specific L2s or sidechains risks repeating the DeFi summer mistakes: launching shiny products without robust liquidity, security, or composability.
My 2024 ETF regulatory harmonization work with ESMA taught me that institutional adoption requires clear rules. Agents cannot sign KYC forms. They cannot hold bank accounts. They need programmable wallets with deterministic compliance. Most current KYC solutions are theater — buying a few wallet holdings can bypass them, and the compliance cost falls entirely on honest users. For agents, this is worse: they cannot even perform the theater. We need new identity models, like decentralized identifiers combined with on-chain reputation, that allow agents to prove trustworthiness without human intervention.
Contrarian: The Decoupling Thesis — Centralized Agents Will Not Wait for Decentralized Infrastructure
The prevailing narrative is that crypto will naturally power the agent economy. I disagree. OpenAI's agents run on centralized servers, pay with credit cards, and settle through Stripe. They do not need crypto today. And they are growing at 10 million users per quarter. By the time the crypto industry finishes debating which L2 is best, these agents will already be deeply integrated into traditional payment rails — making it even harder to switch.
Furthermore, the fragmentation of liquidity across dozens of L2s is not scaling — it is slicing. Agents need a single, global, instant settlement layer. Bitcoin is too slow. Ethereum L1 is too expensive. L2s are too fragmented. The only networks that come close are Stellar and XRP Ledger, but they lack the programmability for complex agent logic. The industry is building solutions for humans, not for machines.
We should also consider the security angle. In my 2018 stability audit of XRP Ledger, I identified latency issues that could be exploited during volatility. Agents operate at machine speed. A vulnerability in a payment rail could drain billions in seconds. The 2022 bridge hacks taught us that centralized points of failure are dangerous. OpenAI's agents are a honeypot — if they adopt a centralized payment rail, one breach could compromise millions of agent wallets.
Takeaway: The Race Is Not About AI — It Is About the Rails
OpenAI's 10 million users are not a victory lap for AI. They are an alarm bell for infrastructure. The crypto industry must stop treating agent payments as a theoretical future use case and start building production-grade, cross-chain payment rails that agents can access programmatically. We need stablecoin-powered, low-latency, composable payment channels that work across L2s and L1s. We need identity solutions that treat agents as first-class economic actors. And we need to learn from past infrastructure failures — the 2018 stability issues, the 2020 governance exploits, the 2022 bridge collapses — to build resilient systems.
Tracing the quiet resilience beneath the market, I see a window of opportunity. Traditional payment rails will not serve agents well — they are slow, expensive, and require human intervention. Crypto has the technical primitives but lacks the product-market fit. The first team to deliver a stable, scalable, agent-native payment rail will not just win a market — they will power the next economy.
The question is not whether agents will need payment rails. The question is whether we will build them before the agents’ human masters decide credit cards are good enough. That decision is being made right now, every day, by 10 million users.