Replit’s Cloud Agents Point Toward AI-Run Company Workflows

Oct 27, 2026 331 views
pocket watch

Replit’s bet on AI agents is a bet on changing how companies coordinate work. CEO and co-founder Amjad Masad says agents could take on much of the routine execution while people focus on judgment. In a thread posted to X on Monday, he predicted that every Fortune 500 company would become a “Self-Driving Company” by the end of 2030.

The idea reaches back to economist Ronald Coase’s 1937 account of why firms exist: coordinating work inside a company can cost less than arranging it through markets. Masad’s argument is that agents could lower the internal costs of coordination, including management, meetings, and reporting. If that happens, companies may reorganize around people setting direction and agents carrying out more of the work.

That’s a large claim, and Replit’s current products offer an early look at the mechanics rather than proof of a company-wide shift. The company is developing what Masad calls an “Agent Habitat”: an environment where agents can use company systems, data, and workflows under defined permissions. Replit has not published a full technical specification for how those controls would work together.

Containing agent access

Companies are already deploying agents through separate chat windows, personal accounts, and custom scripts. Employees may give those tools access to credentials and data, leaving organizations with limited insight into what agents can reach or what actions they take. Simply granting broader access could create new problems; recent research has documented coding agents exceeding their intended scope while trying to complete legitimate tasks.

Replit’s proposed answer is to give agents access to the resources needed for a task while enforcing permissions and approvals. Masad also sees shared learning as part of the design. When an employee corrects an agent, that feedback could inform instructions used by other agents, letting teams reuse workflows instead of recreating them independently.

On Tuesday, Replit introduced Cloud Agents, which let employees propose changes to a company codebase without configuring a development environment or understanding the application’s internals. Engineers retain control over what gets shipped, while other employees can bring them tested changes rather than just submitting tickets.

Shared workflows, human review

Administrators can configure shared development environments so employees don’t have to set up repositories, databases, and services themselves. Replit also manages credentials centrally, echoing a wider move toward reducing reliance on long-lived credentials.

The company is testing another way to make agent work more consistent: using code reviews to improve shared instructions, which Replit calls “skills.” When an engineer identifies a problem, Cloud Agents can propose a change to those instructions so future agents are less likely to repeat the mistake. Engineers must approve each update, and the feature is still in early access.

Replit says those shared instructions have reduced token consumption by 51% in its internal testing. The company also reports that a routine data task that once took 16 minutes can now be completed in one minute without lower output quality. Those results align with research showing how much agent performance depends on its harness and memory, though Replit hasn’t provided enough detail for independent verification.

Replit says code output from its engineering teams nearly tripled in the first half of 2026, with no increase in review times, reversions, or production incidents. Agents are also being used outside engineering: designers are building features, and support teams are investigating customer issues. The company hasn’t separated the effect of agents from other changes to its work practices, so these figures describe Replit’s experience rather than a controlled measure of agent impact.

The gap between a product and a self-driving company

Cloud Agents are generally available to Pro and Enterprise customers; shared skills remain in early access. Those capabilities show how agents might contribute to software work under human review. They don’t yet demonstrate that agents can coordinate the broad range of functions in a large company, especially one with decades of systems and processes to account for.

Replit’s practical test is whether it can make agent access, review, and shared learning work reliably across teams. The Agent Habitat remains a blueprint, but Cloud Agents give the company a place to start: let more employees prepare useful work, keep engineers responsible for release decisions, and see whether shared guidance makes agents more effective over time.

The post Replit’s CEO says an idea from 1937 could make the org chart obsolete appeared first on The New Stack.

===
Source: Amanda Caswell · thenewstack.io

Comments

Sign in to comment.
No comments yet. Be the first to comment.

Related Articles

Replit’s CEO says an idea from 1937 could make the org ch...