Coinbase CEO: AI helps us run the company, and we integrate every piece of data

On Wednesday, Coinbase CEO Brian Armstrong revealed that he has deployed an internal AI system to act as the company's "CEO," which integrates all Slack messages and Google documents, helping him manage the company and provide strategic advice. This practice indicates that the prospect of artificial intelligence replacing executive functions may become a reality faster than expected.

Media reports have previously suggested that AI would replace entry-level jobs and even eliminate some middle management positions. However, Armstrong's approach shows that the possibility of AI replacing CEOs or some executive functions is accelerating. This system has evolved from basic productivity tools to strategic advisors, widely used across teams such as legal and finance.

In an interview with "All-In" co-host Jason Calacanis at Davos, Armstrong stated that he has shifted from using AI for routine tasks to deeper strategic consulting, allowing AI to proactively point out blind spots he might overlook. This "reverse prompting" model is changing the way executives make decisions.

The application of this system has surpassed traditional automation tools, beginning to involve core executive functions such as strategic judgment and time management, providing practical cases for AI's application at the highest levels of business.

AI System Integrates Company Data

Armstrong introduced that Coinbase has deployed a self-hosted AI model over the past year, connecting all company data sources. He said:

Every Slack message, every Google document, all Salesforce data, Confluence content, is now integrated, and you can ask these AI agents questions; every team is using it.

Calacanis likened this system to "Coinbase's oracle," to which Armstrong agreed. The system creates a unified information repository by aggregating all internal communications and documents, allowing AI to have a comprehensive understanding of the company's operational status.

This comprehensive data integration enables AI to identify patterns and issues that human managers might overlook, providing unprecedented informational depth for executive decision-making.

From Execution Tool to Strategic Advisor

Armstrong stated that his use of AI has fundamentally changed. "I'm not just prompting it to 'help me write this memo' anymore," he said, "as CEO, I ask these AI agents, 'What is something in the company that I should know but might not?' It will tell me, 'Did you know that this team actually has disagreements about strategy?' and I genuinely didn't know because it can read every Slack message and every Google document."

This application method transforms AI from a passive execution tool into an active strategic advisor. Armstrong can discover potential issues within the organization through AI, which might be filtered or delayed in reporting through traditional management hierarchies.

According to Armstrong, Coinbase board member and Shopify CEO Tobi Lütke introduced him to the concept of "reverse prompting." “ **He said it’s like reverse prompting "You don't tell the AI agent what you want to do, but rather ask it what you should think more about," Armstrong said.

AI Supervision of Executive Time Allocation

Armstrong further demonstrated the application of AI in executive performance management. "Like a mentor, a coach," Calacanis commented. Armstrong responded, "For example, 'What can make me a better CEO?' It would say, I looked at your time allocation from last quarter, this is how you said you wanted to allocate your time, but in reality, you spent 32% of your time on this instead of 20%."

He also asked the AI other strategic questions, such as "What have I changed my mind about the most in the past year?" This application has gone beyond traditional management tools, involving core executive functions such as self-reflection and strategic adjustment.

Armstrong concluded, "It will proactively prompt you with information you should be thinking about, rather than the other way around." This marks a fundamental shift in the human-machine interaction model — in the future, it's not humans prompting AI, but AI prompting humans