Overview
Brian Armstrong presents a unified thesis: AI, crypto, and biotechnology are becoming infrastructure for expanding human and machine agency. Coinbase is evolving beyond cryptocurrency trading into an “everything exchange,” a global stablecoin payment layer, and an agentic-finance platform where both people and autonomous agents can hold funds, trade, and purchase services. Micropayments are central to this vision because most observed agentic-commerce transactions fall below the fixed fee charged by conventional card networks. Inside Coinbase, AI agents are also changing software development through repository-specific knowledge bases that retain lessons from incidents, experiments, controls, and human corrections, creating a recursive improvement loop. Armstrong expects this productivity gain to accelerate existing organizations while enabling unusually capable small teams, rather than simply eliminating workers. The conversation then broadens to tokenized securities, prediction markets, and New Limit’s AI-guided search for epigenetic therapies that restore cellular function. Armstrong frames New Limit’s disease-specific trials as an entry point toward broader rejuvenation treatments. His longer-term outlook includes cognitive enhancement, embryo editing, and regulatory sandboxes where high-risk technologies can be tested. Across these domains, his recurring principle is that technological progress requires builders who identify foundational constraints, run experiments, and convert accumulated knowledge into scalable systems.
Sections
Deeper Implications
Synthesis of the recurring strategic patterns across finance, software, biotechnology, and governance.
- Armstrong’s projects repeatedly target infrastructure-level constraints: transaction costs in machine commerce, fragmented financial rails, lost organizational knowledge, biological aging, and regulatory barriers to experimentation.
- The proposed agent economy depends on a three-part stack: financial identity, low-cost settlement, and markets where specialized agents can buy capabilities from one another.
- Coinbase’s internal AI system treats human review as training data for future operations, turning software development from isolated task completion into cumulative institutional learning.
- New Limit follows a platform strategy similar to the everything-exchange thesis: build a reusable underlying system first, then apply it across many assets or cell types.
Technical and Quantitative Details
Specific architectures, data points, protocols, and experimental stages described in the interview.
- Card transactions typically carry a flat fee of about $0.30 plus an additional percentage, while approximately 76% of observed agentic-commerce transactions are reportedly below $0.30.
- Coinbase plans both human-linked agent accounts with segregated funds and self-custodial agent wallets capable of holding stablecoins.
- The X42 agent-payment protocol was incubated at Coinbase and contributed to the Linux Foundation, with Google, Cloudflare, AWS, and others described as participants.
- Coinbase’s internal Toshi harness can ingest team and repository knowledge, invoke external vendors, make agentic payments, and coordinate multiple coding agents in parallel.
- New Limit’s pipeline moves from AI-selected transcription-factor hypotheses to pooled screens, functional assays, animal models, non-human primates, and human clinical trials.
- Armstrong reports that 88% of Coinbase revenue comes from non-Bitcoin trading and that prediction markets reached roughly a $100 million revenue run rate within months, though he qualifies the precise growth figures as requiring verification.
Forecasts
Future developments Armstrong expects or considers plausible.
- The number of AI agents will exceed the number of humans, and the agentic economy will eventually become larger than the human economy.
- Specialized, fine-tuned agents will form a marketplace and outperform frontier generalists on selected narrow tasks.
- Most major financial assets—including stocks, private credit, treasuries, bank deposits, and potentially real estate—will move onto tokenized rails.
- New Limit expects to launch its first phase-one clinical trial next year and produce approximately three to ten additional drug candidates from the platform in the coming years.
- Embryo screening and editing for disease prevention will become more socially normal and could eventually make unassisted reproduction seem unusually risky.
Memorable Quotes
Statements that capture Armstrong’s central worldview and ambitions.
- We don't want the AIs to be unbanked. You know, we want to bank the AIs.
- people are not being eliminated. I think tasks are being eliminated.
- crypto was really really good for humans and it's going to be essential for AI
- It's easy to be a critic. It's easy to poke holes what other people are doing
- I think humans need to at some point take control of our evolution and get smarter and smarter.