How Coinbase plans to turn crypto into the payment layer for AI

This article summarizes an episode of Sourcery’s video series featuring Coinbase CEO Brian Armstrong, Coinbase vice president of legal Molly Abraham, and head of base Jesse Pollak.

Image credit: Shutterstock

Traditional banking excludes billions from high-quality investments and lacks the infrastructure for AI to process autonomous transactions. Coinbase CEO Brian Armstrong argues that digital currency solves both failures.

He envisions crypto as a global payment network for both individuals without brokerage access and AI agents that increasingly perform economic work.

Making that vision viable requires regulatory trust and resilient infrastructure. Vice president of legal Molly Abraham emphasizes a compliance-first approach to tokenized assets.

Meanwhile head of base Jesse Pollak focuses on building an open network where autonomous programs can operate securely.

Together, they outline an ecosystem where strict legal standards and open infrastructure turn programmable labor into a functioning economy.

Companies face a math problem with AI

While businesses race to adopt machine learning, processing costs remain a significant barrier. Treating AI like a computer operating system offers a practical solution.

Companies gain a distinct advantage by directing queries to the appropriate software, storing answers for later use, and reserving premium programs only for critical moments.

This strategy shifts software procurement from back-office accounting to strategic AI planning, where organizations can manage rising compute demands by implementing clear triage protocols:

• Classify prompts by difficulty: Separate everyday support, writing, search, and office tasks from risky reasoning or decisions involving customers.
• Route routine work downward: Send easier questions to free software that costs less to run.
• Protect premium models for hard tasks: Reserve the best software for complex jobs where accuracy or risk control alters the business outcome.
• Cache repeated queries: Store common answers so the company avoids paying repeatedly for the exact same calculation.
• Set sensible defaults: Make the cheap, safe path automatic so employees do not need to choose software manually.
• Measure output with quality: Using machine learning allows developers to produce twice as much computer code each year with fewer errors.

Processing demands outpace budgets
Managing surging computing expenses requires directing easy questions to older, cheaper software.

The engineering team expects routine tasks to run on these efficient programs, reserving the newest models for complex problem-solving. This shift is necessary because “the token budget was going up exponentially,” Armstrong explains.

Traditional finance leaves billions without options

Lowering these operating costs frees up capital to build faster, more accessible financial products. To reach a global audience, the company aims to convert corporate shares into digital tokens, solving a systemic exclusion problem that blocks individuals with cash savings from making reliable international investments.

Following the law provides an advantage

Computer programs require a new way to pay

Base provides a testing ground for computer transactions