Warp bets on open source development with GPT-5.5

Warp is making a larger push into open-source software development built around AI agents, and it is using OpenAI’s GPT-5.5 as part of that effort. The company says the model helps agents reason over larger coding tasks, prepare work for review, and run more efficiently in long-lived workflows.

Warp began as a modern terminal and became popular with developers for speed, collaboration features, command workflows, and an AI-native interface. As coding agents became more common in day-to-day engineering, Warp said the terminal was a natural place to work with them because commands, context, collaboration, and review already come together there.

This year, Warp open-sourced its terminal client, with OpenAI as the founding sponsor of the repository. At the same time, it introduced what it calls Open Agentic Development, a model where humans define goals and supervise results, while agents plan work, write code, test changes, and open pull requests.

According to Warp, recent improvements in frontier AI models made that setup practical at scale. The company says GPT-5.5 helps agents handle wider problem spaces and produce work that is easier for humans to review. In internal benchmarks, Warp said GPT-5.5 used 30% fewer tokens per agentic coding task than GPT-5.4.

Warp says it now has nearly 1 million developers using the product and that more than 56% of the Fortune 500 use it. Inside Warp’s own engineering team, agents now co-create around 90% of the company’s pull requests, giving the company a direct test bed for what long-running agent workflows need.

Those needs, Warp says, include observability, coordination, memory, and human review. CEO Zach Lloyd said the company believes it can ship Warp faster by working with its community to supervise a fleet of agents, and that OpenAI models help make long-horizon coding work sustainable.

Warp’s broader bet is that software development is shifting from individual developers interacting with assistants to systems that coordinate many persistent agents over time. In that model, developers specify intent, verify outputs, and decide what ships, while the decisions themselves become reusable context for later agent work.

Warp says that if orchestration is good enough, agents can produce more consistent code than a loosely coordinated group of humans. In that framing, open source shifts as well: humans contribute product judgment and shared vision, while agents handle more of the implementation work.

To support that workflow, Warp built Oz, a cloud orchestration platform that acts as a control plane for deploying and coordinating agents across local and cloud environments. Through a web interface, developers can launch agents, choose skills and environments, select models and hosting configurations, and monitor execution centrally as workflows run.

Oz also lets agents keep running remotely while developers inspect live sessions, review artifacts, and move work between cloud and local environments without losing context. The platform supports recurring workflows too, so agents can run like scheduled jobs.

Warp says it uses techniques such as context compaction, persistent memory, and dedicated subagents for code search and file analysis to keep long workflows on track. Within Oz, tasks are routed by type and difficulty, with more complex coding and reasoning work sent to stronger model configurations.

GPT-5.5 is part of the OpenAI model mix Warp uses for demanding agentic coding workflows, according to the company. Warp also uses OpenAI models as LLM-as-a-judge systems in its evaluation pipelines.

The company says OpenAI models have performed strongly in internal tests for long-horizon engineering work involving reasoning, planning, code generation, and code review. Warp says that is why it is building its open-source and orchestration strategy around those models as it tries to shape how agent-driven software development develops next.