Core content
Brian Armstrong said that internal platforms should disclose utilization, latency, and cost data, as if they were competing with external customers.
He noted that AI agents are already becoming the largest users of many APIs.
He proposed that if agents find missing functionality or bugs, they should submit structured feedback to a dedicated endpoint.
He also argued that internal teams operating real products should be required to charge other internal teams for costs.
This article is an extension of his earlier proposal to introduce a standardized feedback endpoint for all backend services.
Brian Armstrong, CEO of Coinbase, publicly urged on Sunday (local time) that internal engineering platforms should apply the same level of accountability and competitive principles as products aimed at external customers. He pointed to the rapid rise of AI agents as the catalyst for change.
Armstrong raised the issue that while AI agents have become the main drivers of maximum traffic on most APIs, the internal culture and operating system that supports this is still stuck in a people-centered environment of the past.
He argued that on the social platform X, internal platforms should disclose uptime, latency, and cost metrics just like external services, and that feature requests should be collected in public forums rather than via private channels.
Armstrong’s three proposals
Armstrong’s argument can be summarized in three points. First, internal platforms should transparently disclose performance data just like external commercial products. Second, he emphasized that requests for feature improvements and requirements should be collected not through an internal ticket system that other teams can’t see, but through open channels that anyone can view. Third, he suggested that internal teams operating real “products” be able to charge other internal teams that use them, so that cost signals can be used like real market mechanisms.
He wrote, “If internal platforms have to ‘acquire’ internal customers in the same way they deal with external customers, most companies would move much faster.”
These concerns are directly tied to AI agents. Armstrong described AI agents as “users who don’t sleep, knock on APIs at high frequency, and reveal missing functions and bugs at a speed far faster than human engineers.”
In another post he shared later that evening, he also proposed a more specific mechanism. He suggested standardizing dedicated feedback endpoints for all backend services so that agents can send structured reports immediately when they discover functional gaps or bugs. He explained that this feedback could even act as a trigger for other agents to automatically generate pull requests (PRs).
In other words, it’s an architecture that can fundamentally change how software maintenance works inside large organizations—where one agent reports a defect and another agent writes the corrective patch.
Related article: Bitcoin surges 6%…Iran signals a plan to resume Hormuz operations for 7 days
Why crypto infrastructure matters
Coinbase is one of the largest crypto trading platforms in the U.S. for both retail and institutional customers. Their internal infrastructure handles massive volumes of order routing per second, wallet management, regulatory compliance checks, and market data processing. As agent-based trading tools, compliance bots, and automated market maker (AMM) systems grow as well, Coinbase’s internal APIs are effectively facing pressure similar to that on an external developer platform.
The core of Armstrong’s point is that the internal teams building and operating these APIs have, until now, faced less of the “competitive discipline” that external product organizations have to endure. If a team serving ordinary consumers releases slow infrastructure, the cost is immediate—users churn and revenue drops. But if an internal platform team provides a slow and unstable system, all that comes back is complaints in the form of private tickets, not market signals like actual customer churn.
His claim is that the situation changes if internal performance metrics are made visible across the enterprise, if feature requirements and improvement proposals are handled openly, and even if cost charges are introduced. Other teams can evaluate services based on numbers, raise requirements in public, and “vote with wallets” through internal budget allocation.
How did agents become the “primary users” of APIs?
Armstrong has repeatedly spoken publicly about the role AI agents play in reshaping how software companies operate. Earlier this year, he said that inside Coinbase, AI agents emerged as a “major consumer” of developer tools, and that the trend accelerated as it expanded API offerings aimed at external builders through the Base network.
Discussions about agent-based infrastructure are spreading throughout the industry. Anthropic, OpenAI, and others are also releasing guidelines for developers who design multi-agent systems on top of commercial APIs.
Armstrong’s comments came exactly one week before Coinbase was set to present at a major developer conference. However, there has not yet been an official product announcement directly connected to his remarks about internal platforms.
Related article: Ethereum exchange sees a sharp surge in net outflows…ETH recovers $2,700
What to watch next
Among Armstrong’s proposals, the most concrete part is the “feedback endpoint” standard. If Coinbase actually adopts this structure internally and publishes the specifications externally, other companies building agent infrastructure could converge on the same pattern. It means that a de facto industry standard could emerge without the need for a separate standards body.
Meanwhile, the internal cost-charging model is far more controversial. In the past, several companies tried internal chargeback models, but many assessments say they were difficult to sustain due to administrative complexity and organizational politics.
Next read: Bitcoin rises 12% in September…Risk assets bounce back as worries on Wall Street ease