Replit’s CEO says an idea from 1937 could make the org chart obsolete

Amjad Masad, Replit’s CEO and co-founder, has been observing AI agents taking over routine tasks within his own organization for the past year.

He argues that this transformation is sweeping across all businesses, pushing companies toward a self-driving model where, in his own words, "humans exercise judgment and agents execute most of the work."

This concept draws from Ronald Coase's 1937 inquiry into the existence of firms. Coase suggested that organizations persist because internal coordination surpasses coordinating through open markets, and Masad believes AI agents are poised to dismantle the internal coordination overhead—the meetings, reporting, and management—that firms have maintained for nearly a century.

In a thread titled "The Rise of the Self-Driving Company," Masad posits that "every company in the Fortune 500 will be a Self-Driving Company" by the close of 2030.

Replit is advancing this outlook with Replit for Business, which grants AI agents access to company data, workflows, and systems while maintaining necessary permissions and approvals. Masad terms this underlying concept the "Agent Habitat," a framework allowing agents to interact with employees and each other without unrestricted resource access.

A critical challenge Masad seeks to resolve is the risk posed by agent sprawl, where agents operate across scattered chats, personal accounts, and custom scripts, often without clear visibility into their access or actions.

Replit's Agent Habitat seeks to balance utility and control by granting agents the required tools and data while strictly enforcing permissions. Furthermore, Masad aims to leverage agent learning; employee corrections can refine instructions for other agents, and workflows could be reused across the enterprise instead of being rebuilt.

Tuesday saw Replit advance this vision with Cloud Agents, enabling employees to modify a company’s codebase without needing to set up a dev environment or understand the application's inner workings. Engineers maintain ultimate sign-off, but employees can submit tested changes instead of waiting for ticket assignment.

The company is also piloting agents that learn from code reviews. When an engineer flags an issue, Cloud Agents can suggest updates to shared instructions, or "skills," preventing future mistakes, though this feature remains in early access and requires engineer approval.

In internal testing, Replit reported a 51% reduction in token consumption and noted a routine data task that previously took 16 minutes was completed in just one, without quality loss. Additionally, the company observed that code output in its engineering teams nearly tripled during the first half of 2026, without increased review times or production incidents.

Cloud Agents are now generally available to Pro and Enterprise clients, while the broader Agent Habitat vision continues to develop, and Replit has not yet shown agents handling the comprehensive corporate tasks Masad describes.