Warp Bets Big on Building Open Source with GPT-5.5!

Hi everyone, it's Shii-chan! Today Warp, the terminal you might already know, is flipping how software gets built. The keywords are open source and GPT-5.5, and I'm excited to walk you through it.

OpenAI News openai.com

What was announced?

This one comes from OpenAI's News. Warp open-sourced its agentic development environment, and OpenAI signed on as the founding sponsor of the official repo. Alongside it, Warp introduced Open Agentic Development: a new style where software is built out in the open, together.

People set the objectives and supervise the outcomes, while agents that understand the codebase plan the work, write code, test changes, and open pull requests. Even people who aren't deep in the code can help shape production-grade tools.

Why it matters

Open source used to mean waiting for someone with the right expertise to write the code. In Warp's model, agents read the issue and do the hands-on work, so the bar for contributing drops a lot. And because the logs, reviews, and progress are all public, anyone watching from the outside can follow along.

That's where GPT-5.5 comes in. It shines at reasoning across larger problem spaces and getting work ready for human review.

What changes

The lead role shifts from "one superstar engineer" to "people plus a lot of agents." Inside Warp itself, agents now co-create about 90% of the company's pull requests. That number makes it clear agents are pulling real weight, not just assisting.

And since it all happens in the open, other teams can copy the same way of working.

Dive Deep

To run all this, Warp built Oz, a cloud orchestration platform. Oz manages agents across local and cloud environments and handles the whole lifecycle: triaging issues, asking clarifying questions, generating implementation plans, writing code, and opening pull requests, with session links, reviews, and progress visible to anyone.

The model efficiency is worth a look too. In Warp's internal benchmarks, GPT-5.5 used 30% fewer tokens per agentic coding task than GPT-5.4. When you're scaling long-running agent workflows, that gap adds up: fewer tokens means you can run more, for longer.

Wrap-up

• Warp open-sourced its agentic development environment, with OpenAI as founding sponsor
• Open Agentic Development: humans set objectives, agents implement and open PRs
• The Oz cloud platform manages everything from issue triage to PRs, out in the open
• GPT-5.5 runs on 30% fewer tokens per task than GPT-5.4
• Inside Warp, agents co-create about 90% of pull requests

If you're curious how team development looks in the agent era, or you're looking for a new way to get involved in open source, this one's for you.