Coinbase CEO Brian Armstrong announced Saturday that the exchange is testing AI agents modeled after co-founder Fred Ehrsam and former CTO Balaji Srinivasan, deployed directly into Slack and email as virtual teammates, according to Decrypt. Armstrong predicted the company will have “more agents than human employees at some point soon.”
The two agents serve distinct roles. The Fred agent functions as a strategic executive assistant, helping staff “refine documents, strategies, and concepts,” per Crypto Briefing. The Balaji agent is designed to “spark innovation and new perspectives,” leaning toward creative problem-solving. Both appear in the same communication channels employees already use, with no separate interface required.
Early Internal Feedback
Coinbase engineer Travis Bloom reported discussing a new idea with the Balaji agent, saying it “really helped crystallize the vision.” Armstrong described both agents as modeled after “legendary former Coinbase employees” who shaped the company’s early growth, according to Decrypt.
Armstrong added that the next step involves making it “easy for any employee to spin up other agent employees.” Future agents will likely not be “digital twins” of specific individuals, suggesting Coinbase sees the current experiment as a proof of concept for broader internal agent adoption.
Coinbase’s Agent Infrastructure Stack
The workplace agents sit on top of infrastructure Coinbase has been building for months. In February, the exchange launched Agentic Wallets enabling AI systems to hold funds, execute trades, and process onchain transactions autonomously. Those wallets connect to Coinbase’s x402 protocol for autonomous crypto payments without human intervention.
The numbers around crypto-native agents are striking. AI agents now account for 58% of all cryptocurrency trading volume, according to Crypto Briefing. Daily active on-chain AI agents exceeded 250,000 in early 2026, a 400% increase from the prior year. Over 435,000 buyer agents and 90,000 seller agents were operating on Coinbase’s x402 protocol alone as of early March.
From Personas to Workforce Planning
The experiment raises a practical question for every company deploying internal agents: should agents have identities? Armstrong’s choice to model them after real people creates instant legibility for employees. Staff know what “ask Fred” or “ask Balaji” means because they know (or know of) those people. That shortcut sidesteps the blank-slate problem where employees don’t know what an AI agent is good at or when to use it.
But Armstrong signaled this approach won’t scale. Custom agents “tailored to their team’s needs” suggests Coinbase is heading toward role-based agents, not personality-based ones. For a company with 4,279 employees as of mid-2025, the math on “more agents than human employees” implies thousands of purpose-built AI teammates running across departments.