In this special Warp engineering roundtable, Ben Holmes sits down with Aloke Desai (Founding engineer at Warp), Ian Hodge (Lead Engineer, Oz), and Lili Wilson (Engineer, Oz) to unpack how Oz — Warp’s cloud agent platform — was designed and built.

The conversation covers the evolution from Warpy, an early Slack bot MVP, to a fully programmable system capable of running parallel agents in secure cloud environments.

Topics include:

• What Oz is and why Warp built it
• Why traditional backlog planning doesn’t map cleanly to agent systems
• Planning mode and context engineering
• GitHub review workflows for agents
• “Skills” as a reusable abstraction
• Scheduled agent jobs and feature flag automation
• Fine-grained environment security
• Testing agentic systems
• Terminal integration
• Prompt best practices
• Running agents in parallel
• Cloud mode in Warp
• UI automation and computer use
• Artifacts and session transcripts
• What’s next for agent platforms

This episode is for engineers building AI-native tooling, platform teams experimenting with agents, and developers curious about how agent systems work beyond the demo.

About Oz
Oz is Warp’s programmable agent platform for running and orchestrating coding agents in secure cloud environments. Developers can run agents in parallel, automate recurring work, and build on top of agents using APIs, SDKs, and the Warp terminal.

Learn more at: https://warp.dev/oz