Interview

Brian Armstrong on agentic finance at Coinbase: trading via AI agents growing 20% week over week

Nov 17, 2026 with Brian Armstrong

Key Points

  • Coinbase's AI agent trading feature is growing 20% week over week, with Armstrong positioning permissioned agent access as a competitive advantage over financial institutions that have pulled back.
  • Coinbase received a national trust charter from the OCC, regulatory credibility that Armstrong argues makes institutional adoption of crypto-native AI infrastructure more plausible.
  • Armstrong's longevity biotech New Limit reaches first clinical trials in 2026 using AI to screen hypotheses and high-throughput lab screening for cell reprogramming therapies.
Brian Armstrong on agentic finance at Coinbase: trading via AI agents growing 20% week over week

Agentic finance at Coinbase

Brian Armstrong's central argument is that crypto infrastructure and AI agents are converging faster than most of the finance industry is prepared for. Coinbase for Agents, the product that lets users connect an AI agent to their Coinbase account, is seeing trading volume grow 20% week over week. Armstrong describes use cases ranging from automated limit orders timed to market corrections to portfolio rebalancing and tax optimization — practical, mundane financial tasks rather than exotic strategies.

The growth rate is striking enough that Armstrong says Coinbase is actively meeting agentic finance teams as acquisition targets, adding to roughly 10 acquisitions completed last year. The company received a national trust charter from the OCC, which Armstrong frames as the kind of regulatory credibility that makes institutional adoption of crypto-native AI infrastructure more plausible.

“We're really excited about agentic finance right now. Coinbase for agents is the product. You can connect your agent to your Coinbase account. A lot of people are doing trading with it. That's growing like 20% week over week right now. It's crazy. We want to decentralize AI and make sure that there's not just a couple of big monolithic ones.”

Controls, not openness

Where other financial institutions have pulled back from agent access, Coinbase is leaning in with a permissioning layer rather than a prohibition. The approach lets users segregate funds into designated accounts and set constraints on leverage before an agent touches anything. Armstrong argues this is the firm's core competency — taking frontier technologies and making them trusted enough for real money.

On cybersecurity more broadly, Armstrong says Coinbase operates on the assumption that any developer could have malware on their laptop at any given time, and that core infrastructure has to be resilient to that from the start. Every line of code is reviewed by a security-focused AI model before it reaches production. More than 99% of funds are held offline in cold storage, distributed globally and patented. Armstrong also says Coinbase is actively reducing its vendor list, reasoning that rebuilding certain tools in-house trades short-term cost for long-term auditability.

Quantum risk

Armstrong is not alarmed about near-term quantum risk to Bitcoin's cryptography, but he thinks AI could accelerate the development of quantum computers in ways that compress the timeline. Coinbase is a member of a Bitcoin Quantum Advisory Council anchored by Dan Boneh at Stanford, which is working toward upgrading Bitcoin to post-quantum cryptography via improvement proposals including BIP 360. Armstrong's position is that the risk isn't imminent, but getting ahead of it early is worth the effort.

New Limit

Armstrong's longevity biotech, New Limit, is reaching its first clinical trials in 2026. The company uses AI to screen millions of hypotheses, runs the most promising candidates through high-throughput wet lab screens, and then tests in functional assays and mouse models before advancing to human trials. The core thesis is cell reprogramming — restoring cellular function to a younger state. Armstrong is candid that full biological simulation remains far off; he notes that even simulating a single cell, let alone an organism, is orders of magnitude beyond the protein-folding work that earned DeepMind a Nobel Prize. The near-term bottleneck, once the clinical trial pathway is underway, is physical throughput in the lab — which Armstrong expects humanoid robots to eventually address, though he thinks they need to master simpler tasks first.

Every deal, every interview. 5 minutes.

TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.