Coinbase Is Testing AI Advisors Modeled on Its Own Alumni
Coinbase has built and is testing internal AI agents designed to simulate the strategic thinking of two of its most influential former executives, giving employees access to simulated senior counsel at any hour.
Brian Armstrong, Coinbase's CEO, disclosed on April 20 that the company is running internal tests on AI agents built to replicate the advisory styles of Fred Ehrsam, the exchange's co-founder and former President, and Balaji Srinivasan, its first Chief Technology Officer. The agents are not consumer products. They are internal tools intended to give Coinbase staff access to the kind of high-level strategic perspective that typically lives only with a company's most senior leaders.
The move is the latest signal that Coinbase is treating AI integration as an operational priority, not a marketing posture. Armstrong previously disclosed that roughly 40 percent of Coinbase's daily code is now generated by AI systems. Separately, the company has deployed RAPID-D, a four-agent decision support system built on Anthropic's Claude 3.7 Sonnet. RAPID-D assigns distinct roles to each agent: one analyzes, one retrieves context, one plays devil's advocate, and one synthesizes the outputs into a recommendation. The system also integrates real-time stakeholder feedback, distinguishing it from a static advisory loop.
Coinbase's own engineering documentation, sourced from the ZenML LLMOps Database, describes the system's purpose in full: "RAPID-D is not designed to replace human oversight but to augment it with a scalable, AI-driven diligence engine."
Who the Agents Are Modeled After
The choice of Ehrsam and Srinivasan is deliberate. Ehrsam co-founded Coinbase in 2012 and later co-founded Paradigm, now one of the most influential venture firms in crypto. His worldview centers on open, permissionless infrastructure built for the long term. In 2026, he was appointed by President Trump to the President's Council of Advisors on Science and Technology. He has also stepped back from day-to-day work at Paradigm to pursue Nudge, a brain-computer interface startup, signaling a pivot toward adjacent frontier technology.
Srinivasan joined Coinbase after the company acquired Earn.com and holds four Stanford degrees: a BS, MS, and PhD in Electrical Engineering, plus an MS in Chemical Engineering. He authored "The Network State" and has argued publicly, across multiple forums and over an extended period separate from the Coinbase announcement, that crypto's "strict deterministic patterns" serve as a necessary safeguard against AI systems that are "extremely good at faking things."
These are not generic executive archetypes. They represent specific, substantive positions on how financial infrastructure should be built and who it should serve.
The Agent Economy Coinbase Is Building Toward
The advisor agents fit inside a larger strategic frame Armstrong has been articulating for months. On March 9, 2026, he said publicly: "Very soon there are going to be more AI agents than humans making transactions. They can't open a bank account, but they can own a crypto wallet."
The structural reason is straightforward: AI agents cannot satisfy Know Your Customer requirements, which means they cannot access traditional banking. Crypto wallets currently carry no equivalent restriction, a conclusion Armstrong's argument implies and that holds under existing regulatory frameworks.
Coinbase has already begun building the rails for that future. Its x402 protocol revives the dormant HTTP 402 "Payment Required" status code and redesigns it for machine-to-machine crypto transactions on Base, Coinbase's Ethereum Layer 2 network (a secondary blockchain built to process transactions faster and at lower cost than Ethereum's main chain). On February 11, 2026, Coinbase launched Agentic Wallets built on the x402 protocol. By March 9, x402 had processed more than 50 million transactions. Coinbase, Cloudflare, and Stripe have since formed a nonprofit to govern x402 as an open-source standard. The company's AgentKit developer toolkit lets any developer give an AI agent a crypto wallet and on-chain capabilities, with support for multiple large language models including OpenAI, Anthropic Claude, and Llama. The toolkit is also framework-agnostic, meaning developers can integrate it across a wide range of development environments without being locked into a single stack.
What This Means Outside the United States
Armstrong's core argument, that AI agents need crypto because they cannot access traditional banking, lands differently in markets where millions of human users face the same structural exclusion. India ranked first in the 2026 Global Crypto Adoption Index, with Nigeria in second place. Ethiopia ranked tenth, Kenya thirteenth, and Ghana twentieth.
Sub-Saharan Africa recorded stablecoin growth exceeding 180 percent year over year, according to Crypto News Navigator citing the 2026 Global Crypto Adoption Index, driven primarily by remittances, merchant payments, savings, and savings dollarization. That last driver carries particular weight in Nigeria and Ethiopia, where currency instability makes dollar-pegged stablecoins a practical hedge rather than a speculative instrument.
Nigeria's Binance Wallet has 30 million users. Kenya's BitPesa serves 6.5 million remittance users.
The architecture Coinbase is building for AI agents reflects a financial access problem that residents of these markets have navigated for years without alternatives. The tools designed for non-human actors may, in practice, benefit human ones in underserved regions first.
Regional infrastructure is beginning to adapt. South African exchange VALR launched an AI service on April 10, 2026, explicitly designed to serve both human users and autonomous AI agents as independent market participants. VALR operates within a perimeter licensed by the FSCA (Financial Sector Conduct Authority), South Africa's primary financial regulator. That regulatory grounding matters: for readers in Nigeria and Kenya navigating their own evolving frameworks, the FSCA designation signals substantive compliance rather than marketing language, and positions VALR among the first licensed African exchanges to formally accommodate non-human financial actors.
What Comes Next
The global AI agent market was valued at approximately 8 billion dollars as of 2025 and is projected to reach 50 billion dollars by 2030. Estimates for the total economic footprint of AI agents by that same year reach as high as 3 to 5 trillion dollars, according to projections cited by CoinDesk that include figures tied to Coinbase's own AgentKit addressable market analysis. Readers should weigh that figure with its provenance in mind.
Whether Coinbase's internal advisor agents prove useful as a staff tool matters less than what their existence signals: the company is now building both the philosophical framework and the technical infrastructure for a financial system in which AI agents are routine participants. The question for regulators, developers, and users in emerging markets is not whether that system arrives, but whether they will have shaped it before it does.