← back
The Agent Behind the Curtain: Building the Oz Cloud Agent Platform — Safia Abdalla, Warp
Takeaway
A cloud-agent platform should absorb infrastructure complexity while exposing composable primitives and preserving developer choice.
Summary
- Warp's Oz platform moves long-running coding work into cloud sandboxes while preserving developers' established workflows.
- Managed and self-hosted execution support differing infrastructure and security needs; multiple harnesses share consistent conversation-state and artifact handling.
- Agents can divide research, implementation, and validation across different models and harnesses through prompt-based orchestration.
- APIs expose agent creation, parent-child relationships, environments, compute, and artifacts so teams can build their own interfaces and workflows.
cloud-agentswarpdeveloper-tools
Original description
Warp open sourced about three months ago and went from roughly 20,000 GitHub stars to over 60,000, with thousands of pull requests and hundreds of contributors arriving at once. Rather than let agents fire off code, Safia Abdalla's team put them inside the repository's process. File an issue and an agent triages it, researches the codebase, and asks you questions when the request is too abstract to act on. Every pull request then goes through an agent managed review, and nobody on the team gets pinged until an agent has approved it. The humans only ever see the high signal ones. That is one payoff of a principle she applies all the way down the cloud agent platform, that a platform should absorb complexity before it reaches the user. Agents need somewhere to run, so sandboxes come both managed and self hosted, because teams doing serious work already run their own infrastructure. Developers are attached to their harness, so the platform accepts any of them while keeping conversation state and artifacts consistent across all of them. Real engineering rarely fits in one prompt, so agents orchestrate sub agents, either by prompt or through an API that exposes every primitive in the stack. People outside engineering at Warp used that API to build their own Slack tooling for triaging social mentions. Abdalla closes by pushing back on the phrase software factory, which she thinks loses the people in it, and offers a potter's workshop instead: stations, sourcing, verification, and an owner who keeps refining the system around the work. Speaker info: https://github.com/captainsafia https://x.com/captainsafia https://captainsafia.com Timestamps: 0:00 - Eight years of building developer tooling 2:43 - Why the work moved to the cloud 3:35 - Absorb complexity before it reaches the user 5:18 - Multi harness support without fragmentation 6:59 - One agent is rarely enough 7:53 - Orchestration by prompt, and by API 9:33 - What non engineers built on the SDK 11:16 - Agents inside the open source repo 14:44 - Why the term software factory falls short 17:14 - The workshop as a serious system