Replit lets non-engineers send tested code changes to company codebases

From plain-language request to pull request

Replit says its coding agent, the first generally available one when it launched in 2024, made it possible for almost anyone to build an app. Changing a company's core product, however, has stayed out of reach for people outside engineering. Designers, support leads and other experts often know what should change, but their ideas usually end up in a ticket that waits for a sprint.

With Cloud Agents, a contributor describes the change in plain language. The agent finds the relevant code and makes the change, while a live preview of the app updates next to the chat. The contributor can react to what they see and ask for adjustments before anything reaches engineering.

When the result looks right, the agent opens a pull request in GitHub or a merge request in GitLab that follows the team's skills and contribution guidelines. If a human or an agent reviewer asks for changes, the agent makes them and updates the request. Replit calls this part of what it names the Self-Driving Company.

Testing in sandboxes, with video

Replit describes review as a common bottleneck: non-technical contributors cannot run the app themselves, so they depend on engineers for screenshots and fixes. Cloud Agents try to remove that loop by checking the change in isolated cloud sandboxes before the pull request reaches anyone.

The agent opens the running app in a browser and clicks through the change the way a customer would. If something fails, it fixes the code and tests again. It then attaches screenshots and video of the final version to the pull request. Engineers therefore receive pull requests they do not have to run themselves to understand.

Replit also says anyone in a company can ask questions about the code and get answers grounded in what is actually built. Its examples include a support lead investigating a recurring customer error and a product marketer checking that a feature works as the launch copy says. According to Replit, its agent on the company codebase was rolled out internally at the start of March 2026.

Shared environments and workspace memory

Setting up an app to run properly takes the right repos, runtime versions, databases, background services and credentials. Replit says an admin can configure a shared environment once. A Cloud Agent explores the codebase and sets up an initial virtual machine with repos, runtimes, skills, tools and dependencies. It updates automatically from the main branch, so each change starts from current code.

For engineers, Replit says the environments cover dozens of programming languages, microservice setups with several repos, Docker images, databases, queues and caches. Credentials come from external secrets managers through a credential broker.

A separate feature, self-improving workspace memory, is in early access. When review feedback is worth keeping, the agent proposes an update to the team's skills, the written instructions agents follow in the codebase, or drafts a new one. Engineers approve each update before it takes effect. Replit says that internally it used workspace memory to reduce token usage by 51% and cut the time for a common data task from 16 minutes to 1 minute, without reducing output quality.

Slack, mobile and availability

Requests can also start where teams already talk. Users can mention or message @Replit in Slack to investigate an issue or turn a proposal into a pull request, follow progress in the thread, and continue in Replit later. Microsoft Teams support is coming soon. Pull requests can also be started from the Replit mobile app.

Cloud Agents for existing repositories are available to all Pro and Enterprise customers, who can connect a repository at replit.com/code-repos or talk to the sales team about an organization-wide rollout. The Slack integration is available to all customers and is set up through Workspace Settings. Replit held a webinar on getting started with both on November 5.

For teams, the pitch is that engineering still decides what ships, while the people closest to a problem can submit a tested pull request instead of a ticket.