Coinbase is using an internal AI coding system called Forge to explore what it sees as a next-generation software development model. The platform has grown from an early in-house tool into a broader ecosystem with multiple AI development capabilities, and it has become one example of how enterprises are applying Agentic AI.
Forge was previously known as Cloudbot, according to the disclosed information, and an earlier version was even called Claudebot. It is now an important part of the open-source AI coding framework OpenSWE.
Forge’s role inside the OpenSWE stack
OpenSWE is meant to provide infrastructure similar to an “AI software engineer.” Its feature set includes agent orchestration, cloud sandboxes, tool calling, integrations with Slack, Linear, and GitHub, sub-agents, and the ability to automatically create Pull Requests.
Mux lets engineers run several AI agents at once
One of Forge’s most closely watched features is Mux. The tool currently has about 600 users inside Coinbase and allows engineers to run multiple AI agents at the same time, shifting software development from a serial process to a parallel collaboration model.
Forge also includes Slack task routing and a “bug-to-fix” automation flow aimed at cutting down repetitive manual work in development.
Internal experiment reduced use of traditional IDE tools
Coinbase engineering lead Chintan Turakhia previously ran an internal experiment that asked nearly 1,000 engineers to stop using traditional IDE tools for two weeks. The goal was to see which parts of the development process could be automated through an AI-driven workflow.
He said Coinbase’s AI spending has declined as use of AI coding tools has increased, while token usage has kept rising. He also said the company’s in-house AI development tools give employees an alternative to third-party AI coding services from Anthropic and OpenAI.
Independent outside feedback is still limited
Independent public feedback on Forge and Mux remains limited for now. Most of the information available publicly has come from Coinbase management, while ordinary engineers’ hands-on experience with the tools has not yet been widely disclosed.
According to the industry view cited in the report, Forge points to a newer direction for enterprise AI adoption: companies are no longer only calling general-purpose large models, but building AI agent systems around their own business workflows to improve development efficiency, reduce costs, and strengthen internal productivity.