Replit CEO Amjad Masad declared Monday that every Fortune 500 company will become a "Self-Driving Company" by 2030, with AI agents handling most execution while people provide judgment, according to a thread posted to X. Masad argues that artificial intelligence agents are poised to eliminate the internal coordination expenses — management layers, meetings, status reports — that organizations have accepted as unavoidable for almost a hundred years. His reasoning draws on economist Ronald Coase's 1937 work questioning why companies exist at all, with Coase concluding that firms persist because organizing work within them proves cheaper than coordinating through open markets.
Replit's own engineering teams saw code output nearly triple during the first half of 2026 without longer review times or an uptick in reversions or production incidents, the company reported. The firm's internal testing showed a 51% reduction in token consumption, with a standard data task that previously required 16 minutes completed in just one minute without compromising output quality. Agents are now deployed beyond software development at Replit, with design teams building features and support staff investigating customer issues. The company on Tuesday released Cloud Agents, allowing workers to modify a company's codebase without configuring a development environment or understanding how the application functions internally.
Masad is attempting to address a growing challenge as organizations roll out more AI agents, according to the post. Many currently function through dispersed chat interfaces, individual accounts, and bespoke scripts, often with whatever permissions and information employees provide them, leaving companies with minimal visibility into what those agents can reach or have executed. Replit's Agent Habitat concept aims to provide agents with access to necessary tools and information while maintaining permission structures, though the company hasn't yet released a complete technical specification detailing how those safeguards would function together. Masad also envisions companies capturing what their agents discover, so that when an employee corrects an agent, that input could be incorporated into guidelines that other agents reference.
The shift toward self-driving companies rests on collapsing the coordination costs that have justified corporate structures since the 1930s, according to Masad's framework. Replit is advancing toward that model with Replit for Business, which grants AI agents entry to company systems, information, and processes while preserving approvals and controls. The platform lets administrators establish shared development environments so employees don't need to set up repositories, databases, and services themselves, with credentials managed centrally rather than scattered across teams. Cloud Agents are now available to Pro and Enterprise customers, while the skills feature — which allows agents to learn from code reviews and suggest updates to shared instructions — remains in early access. The broader Agent Habitat vision is still forming, and the company cautions that Replit hasn't demonstrated agents can manage the company-wide work Masad describes, with gains at an AI-native firm potentially much harder to achieve at a large enterprise with decades of legacy systems. What functions inside Replit's fast-moving culture may require fundamentally different governance when the stakes involve regulated industries or entrenched organizational hierarchies.