Imagine an open source application that allows you to merely suggest a new feature, or flag a bug you found, and a team of well-managed agents will swarm together to write the necessary code.
This is the thinking at Warp, the company (aka Denver Technologies) that is testing how well an army of agents could manage – and improve – an open source repository. OpenAI is providing the tokens to test out this idea.
The company just open sourced its flagship Rust-based terminal emulator. In doing so, Warp the company rethought how an open source repository could operate in the age of AI.
“Open-sourcing with an agent-powered repo is our vision of how software will be built in the future,” wrote Zach Lloyd, Warp founder and CEO, in a blog post.
For Lloyd, the future of production-grade software development will be based on managing agents “at scale.” In this new world of agentic development, devs don’t code but instead craft specifications to be interpreted by agents.
Warp recently rebranded its agentic software engineering-focused terminal emulator as an “agentic development environment” (ADE). Among its features, it can run autonomous agents on Oz, Warp’s proprietary cloud-based orchestration engine, a version of which, Oz for OSS, the company also commercially offers for third parties to manage their own repos.
How Open Sourcing Warp Is Different
Open sourcing Warp wasn’t the usual GitHub code dump.
Warp’s own agents will be the primary maintainers of the repo.
The Warp core team built over 6,700 agents in the past month to manage the software as well as 20 automations, run through cron, that clean up the code base daily. A user can suggest a feature or submit a bug report, and a Warp agent will do the lowly work of writing the needed Rust code.
In a GitHub comment, a contributor can ask “@Oz” to draft a specification for a new feature. It will create a pull request with two files describing what should be created and how. The user reviews the work to ensure it corresponds to their request, editing the files in markdown if necessary. When finished, they submit it to the core Warp development team for final approval. (Users can also write the code by hand or use their own agents for the task).
At last count by the company, about 60% of merged pull requests are managed by Oz.
This move could greatly expand the number of contributors to Warp, as anyone with a good idea could potentially add to the application. “Anyone contributing should have a high chance of success coding a feature correctly,” Lloyd wrote. “We now have a lot of confidence in code that is generated by Oz with our rules, context and verification.”
In turn, having Oz manage this repository will help the company refine its own practices in agentic engineering. “There’s a virtuous loop here,” Lloyd wrote.
How OpenAI Supports the Warp Repo
With over 700,000 active users, Warp is one of the most popular agent-focused command line interfaces (CLIs) for creating agents, alongside Wave Terminal and the Microsoft AI shell for Copilot. However, managing an open source repository agentically requires significant resources for API tokens.
To address this, OpenAI is supporting Warp’s efforts with token credits, allowing the company to refine its agentic engineering practices despite the high costs.
Warp is part of a broader trend of OpenAI supporting agentic experimentation; the AI lab recently launched a $1 million initiative, called the Codex Open Source Fund, to provide $25,000 grants in API credits to open source projects that integrate AI into developer workflows.
