No PriorsCoinbase’s Everything Exchange: Agentic Finance, Stablecoins & Tokenization with CEO Brian Armstrong
CHAPTERS
- 0:00 – 0:50
Agentic commerce needs micropayments: why cards fail under $1
Brian explains why traditional debit/credit rails break down for agentic commerce: fixed per-transaction fees make sub-dollar payments uneconomical. He frames the goal as giving AI agents real financial access so they aren’t “unbanked.”
- •Card rails often have a ~$0.30 fixed fee plus percentage, making tiny payments impractical
- •A large share of agentic transactions are below the card fee floor (many under ~$0.30)
- •AI agents will need their own accounts to pay and get paid
- •Coinbase is building simple tooling to provision agent financial accounts quickly
- 0:50 – 1:57
Brian Armstrong’s scope: Coinbase, AI adoption, and New Limit
Elad introduces Brian and sets the agenda: crypto’s evolution at Coinbase, AI’s impact on finance and internal operations, and Brian’s longevity biotech company New Limit. The conversation is positioned as a cross-domain look at major technology shifts.
- •Brian leads Coinbase and has been aggressive about AI adoption
- •New Limit focuses on longevity/anti-aging using modern biology and AI methods
- •Discussion will span crypto markets, payments, AI, and biotech entrepreneurship
- 1:57 – 2:43
Coinbase’s “Everything Exchange”: one venue for every asset
Brian lays out the thesis that all asset classes are moving on-chain, enabling a single exchange with liquidity and cross-margin. He also flags regulatory clarity as a key enabler for expanding product coverage and geography.
- •Tokenization enables trading of stocks, commodities, crypto, and derivatives in one place
- •Cross-margin and unified liquidity become possible when assets live on-chain
- •Perpetual futures and prediction markets fit into the same unified venue
- •Regulatory clarity is expanding the set of countries/products Coinbase can serve
- 2:43 – 3:13
Stablecoin payments as the new global money transport layer
Brian argues stablecoins are growing even through crypto downturns because they solve a different problem: cheap, fast, global transfers. Coinbase is leaning into helping businesses integrate stablecoin payments for both humans and agents.
- •Stablecoin payments can be near-instant and very low cost globally
- •Growth persists even when speculative crypto prices fall
- •Coinbase helps companies integrate stablecoins as a payments method
- •Payments infrastructure becomes foundational for broader on-chain finance
- 3:13 – 5:22
Agentic Finance (AI-Fi): AI advisors + AI-owned accounts
Brian describes two prongs of AI-Fi: consumer-facing AI advisors that optimize personal finance, and agent-facing financial accounts that let AI transact autonomously. He explains why crypto rails and self-custody help overcome identity/KYC constraints for non-human agents.
- •AI advisors can democratize wealth-management-like guidance (portfolio, taxes, rebalancing)
- •Agents need their own spend and trading accounts to operate end-to-end
- •Coinbase is exploring both linked (human-owned) agent accounts and fully self-custodial wallets
- •Crypto rails avoid KYC hurdles that agents can’t satisfy (no government ID)
- 5:22 – 7:19
Protocols and marketplaces for agent payments (X402, agentic.market)
Brian details how agents get “stuck” at paywalls and checkout flows, motivating native agent payment protocols. He highlights X402 (incubated at Coinbase, now in the Linux Foundation) and the agentic.market dashboard tracking this emerging economy.
- •Agents frequently fail on tasks due to payment/checkout friction (cards, paywalls, AWS spend)
- •Agents will need dedicated spend accounts (and possibly stablecoin-backed cards for compatibility)
- •X402 aims to standardize agentic payments; multiple large infra players participate
- •agentic.market tracks transaction stats and service providers; Coinbase Business supports merchants
- 7:19 – 9:11
What agents buy for pennies: data, tools, and specialist agents
The discussion drills into what sub-$0.30 transactions represent—often information retrieval and agent-to-agent tool calls. Brian predicts a market of specialized agents fine-tuned on proprietary workflows that can outperform general frontier models on narrow tasks.
- •Micropayments often purchase information: research, paywalled data, scraping, financial data
- •Agents increasingly pay other agents for specialized capabilities (like tool calls)
- •Small fine-tuned/open models can beat frontier models on domain-specific tasks
- •A future marketplace will contain many specialist agents alongside generalists
- 9:11 – 11:45
AI scale, scarcity, and why crypto complements an agent economy
Elad raises concerns about AI resource aggregation; Brian responds by arguing scarcity persists even with automation. He contends money will remain necessary and that crypto rails are well-suited for machine-speed commerce.
- •Safety concerns include potential AI resource aggregation and manipulation risks
- •Brian argues scarcity (land, energy, compute) persists for a long time
- •Even with deflation from automation, a medium of exchange remains useful
- •Crypto is positioned as essential infrastructure for AI-to-AI transactions
- 11:45 – 14:17
AI inside Coinbase: building a “company brain” and recursive self-improvement
Brian explains Coinbase’s internal AI evolution from basic copilots to a structured knowledge system (“brains”) per team/repo/service. The key is feedback loops: corrections must be written back into the system so future agents improve, enabling recursive self-improvement.
- •Started with table-stakes AI: coding tools, support automation, fraud/risk ML
- •Created “brains” capturing incident history, controls, experiments, and code review outcomes
- •Enforcing write-back: human corrections become durable context for future agent runs
- •Goal is higher one-shot PR acceptance and compounding productivity gains
- 14:17 – 16:14
Toshi agent harness and parallelized engineering: the CEO ships PRs
Brian describes the internal agent harness (Toshi) that can spawn multiple agents, choose cheaper models, and execute work in parallel. He notes a behavioral shift: instead of pinging teams in Slack, he can generate PRs for review, accelerating iteration.
- •Toshi integrates tools, knowledge bases, and (eventually) external vendor calls and payments
- •Agents can generate multi-phase plans and execute multiple workstreams concurrently
- •Model selection can optimize cost/performance (mix of open-source and hosted models)
- •Workflow shift: leaders can submit completed PRs rather than task requests
- 16:14 – 18:09
Productivity vs. headcount: why companies may not shrink (but speed up)
Elad asks whether AI makes companies much smaller; Brian distinguishes task automation from eliminating people. He expects existing teams to move faster and ship more, while acknowledging new tiny teams may achieve outsized output.
- •AI removes/automates tasks more than it removes people directly
- •Incumbents likely keep headcount but accelerate delivery and scope
- •New startups may achieve scale with very small teams
- •Operating leverage can improve if revenue grows faster than team size
- 18:09 – 21:36
Tokenization and the Everything Exchange expansion: from crypto to tokenized stocks
Brian notes Coinbase revenue is now largely non-Bitcoin trading and outlines tokenized stocks as a major step beyond stablecoins. He emphasizes global access: billions lack brokerage accounts, and tokenization can unlock investment participation and easier transferability.
- •Coinbase reports most revenue now comes from non-Bitcoin activities (per earnings materials)
- •Tokenized stocks (ex-US initially) aim for 1:1 backed, custody-held underlying securities
- •Stablecoins were the first major tokenization wave; stocks are the next
- •Tokenization could extend to treasuries, deposits, private credit, and more; enables simple transfers (e.g., gifting shares)
- 21:36 – 23:58
Prediction markets: beyond sports into policy and ‘big questions’
Brian is bullish on prediction markets’ rapid growth inside Coinbase and argues they can inform society more broadly than sports betting. He imagines markets for policy outcomes and even “opinion markets” on enduring questions that update as evidence changes.
- •Prediction markets are scaling quickly in-product (as described in earnings commentary)
- •Sports is a major on-ramp, but not the endpoint
- •Markets could forecast outcomes of proposed policies (e.g., unemployment impacts)
- •“Opinion markets” could track beliefs on unresolved questions, shifting with new evidence
- 23:58 – 28:16
New Limit origin story: from dinners to epigenetic reprogramming startup
Brian explains why he started incubating New Limit while remaining focused on Coinbase: a desire to fund “hard tech” and pursue under-addressed meta problems. A series of expert dinners led him to epigenetic reprogramming, which became compelling enough to build around.
- •Post-IPO, Brian wanted to incubate select big ideas without leaving Coinbase’s mission
- •He looked for underfunded areas where he could add leverage beyond capital
- •Longevity felt full of hype; expert conversations highlighted epigenetic reprogramming as promising
- •New Limit formed with a strong founding team; Brian serves as investor/board member
- 28:16 – 35:01
New Limit’s platform: AI-guided screens to clinical trials (liver, vascular, immune)
Brian details New Limit’s approach: an AI model proposes transcription factor combinations, validated through high-throughput pooled screens, then functional assays, then clinical progression. He shares initial target cell types and the strategy of starting with high unmet-need indications before broader rejuvenation use cases.
- •AI explores a huge search space of transcription factor sets for reprogramming
- •Workflow: pooled screens → phenotypic hits → functional assays/animal models → human trials
- •Public roadmap starts with hepatocytes (liver), vascular cells, and immune cells (notably T cells)
- •First clinical path aims at severe liver disease as an initial indication, with longer-term rejuvenation goals
- 35:01 – 41:08
Five-year outlook: agentic finance, cognitive enhancement, and human evolution
Brian prioritizes scaling crypto’s share of global GDP via agentic finance, stablecoins, and the Everything Exchange, while also exploring frontier topics like cognitive enhancement. He argues humans should intentionally advance—via biology and potentially brain-machine interfaces—alongside rapidly improving AI.
- •Near-term focus: expand on-chain finance adoption among humans and agents
- •Crypto remains a small fraction of global GDP; goal is orders-of-magnitude growth
- •Frontier curiosity: cognitive enhancement and raising baseline human capabilities
- •Gene editing for disease prevention may normalize; debate over diversity vs standardization
- 41:08 – 45:09
Special economic zones and ‘freedom cities’: governance as a lever for innovation
Brian frames special economic zones as a meta-solution to regulatory drag, enabling safe sandboxes for experimentation. He cites global examples and discusses a U.S. version—freedom cities on federal land—where constrained pilots could unlock progress in energy, biotech, drones, and crypto.
- •Overregulation can slow iteration; sandboxes enable rapid experimentation
- •Examples include Shenzhen, Singapore, Dubai; Prospera in Honduras as a modern attempt
- •Freedom cities concept: carve out zones with streamlined approvals (EPA, FAA, etc.)
- •Federalism already creates partial competition; targeted zones could push further
