Coinbase’s Everything Exchange: Agentic Finance, Stablecoins, and Tokenization with CEO Brian Armstrong

Summary

Brian Armstrong says Coinbase’s core strategy is shifting toward agentic finance: giving AI agents financial accounts, wallets, and payment rails so they can transact in micro-payments that card networks handle poorly. He also frames Coinbase as an “Everything Exchange” where tokenized stocks, private credit, Treasuries, bank deposits, stablecoins, and prediction markets can all live on-chain or in wallet-native rails. Armstrong argues AI will eliminate tasks more than jobs, and says Coinbase is building internal AI “brains” and orchestration systems to improve engineering and operations. The conversation then turns to New Limit, his longevity company, which uses AI-driven epigenetic reprogramming screens to test transcription-factor combinations and push toward first-in-human trials for diseases like alcohol liver disease. He closes with broader views on social progress, cognitive enhancement, brain-machine interfaces, and special economic zones as tools for accelerating innovation.

Notes

• Coinbase’s near-term thesis is agentic finance, stablecoin payments, and an “everything exchange” model.
• He says crypto adoption is still tiny versus the global economy.
• Coinbase wants AI agents to be banked, not “unbanked.”
• He argues card rails break down for microtransactions because of flat fees.
• Coinbase is building both human-linked agent accounts and self-custodial wallets for agents.
• He frames these agent rails as useful for both consumer and business payments.
• Coinbase says agentic payments are flowing through X402, a protocol it incubated and pushed into the Linux Foundation.
• He expects specialized open-weight models to matter more than frontier models for many internal tasks.
• Coinbase is experimenting with a per-team or per-service “brain” that stores institutional context.
• He wants agent edits and human corrections fed back into that memory system.
• Toshi is an internal harness that can call vendors, pay over agentic finance rails, and dispatch multiple agents in parallel.
• Armstrong thinks AI will remove tasks more than people.
• He says Coinbase’s revenue mix is now dominated by non-bitcoin trading.
• Tokenized stocks are one-to-one claims on the underlying security held in custody, and are currently outside the U.S.
• He extends tokenization to private credit, Treasuries, and bank deposits.
• Armstrong argues tokenization will let assets move directly in wallets rather than via legacy intermediaries.
• He says prediction markets are growing extremely quickly inside the Coinbase app.
• He sees prediction markets as useful for policy analysis and broad uncertainty, not just sports or politics.
• New Limit uses AI to search a huge hypothesis space for transcription-factor combinations.
• He says New Limit built a frontier model for epigenetic reprogramming that proposes the next experiments.
• The company has already shown reprogramming success in at least one human cell type in humanized mouse models.
• New Limit is now testing in non-human primates before a planned Phase 1 trial.
• Armstrong expects several more drug candidates from the platform over time.
• The first three target cell types are hepatocytes, vasculature cells, and T cells.
• The first clinical indication is alcohol liver disease.
• He says ALD has a very poor prognosis after diagnosis.
• Armstrong estimates a successful ALD drug could be worth about $20 billion.
• He sees the long-term goal as rejuvenating many cell types, including liver, immune system, vasculature, brain, muscles, and skin.
• Armstrong says cognitive enhancement should be explored across adults, children, and embryos.
• He cites nutrition as one way to raise IQ in development.
• Armstrong says embryo editing for disease prevention is supported by most Americans, citing Pew.
• He predicts embryo screening and editing will become normalized.
• He argues gene editing could increase diversity rather than reduce it.
• Armstrong says AI progress creates a race in which humans may need to improve too.
• He links that future to brain-machine interfaces.
• Armstrong is highly interested in special economic zones as a way to unlock progress.
• He says Coinbase invested in Prospera in Honduras to learn from it.
• He points to Freedom Cities as a federal-land sandbox for nuclear, crypto, drone delivery, and data centers.
• He says these zones would reduce red tape and allow safe, contained experimentation.
• He notes that the U.S. already behaves like 50 special economic zones because of federalism.