Warp launches Warp Factories, ready-made infrastructure for AI software factories

The term "software factory" has gained momentum in recent months as a way to reorganize engineering teams for the AI era. The concept is straightforward: an agent loop that wraps the classic development stages — triage, specification, implementation, review and validation — letting automation handle large portions of the process. Companies such as Stripe and Ramp have already built systems like this in-house. Stripe developed a "minions" system for internal automation, while Ramp runs a background agent that monitors code after deployment. The problem is that building that infrastructure from scratch demands resources only large companies possess.

Warp, the AI coding company, is targeting precisely that gap. Chief executive Zach Lloyd told TechCrunch that running agents in the cloud, steering them at runtime, returning work to the local environment, managing cross-agent memory and setting up shared evaluations — all together — constitute a "massive infrastructure project." Warp Factories arrives with the architecture already built and the hard decisions already made. The system sits as an infrastructure layer, providing a simple environment for deploying agents plus a roadmap for how to use them.

There is no lock-in to a specific model: the system works with Codex just as well as with Claude Code, and the user chooses the harness that fits. On the integration side, Warp Factories connects to ticketing systems such as Linear and Jira, and to communication channels including Slack and Teams, aiming to slot into existing workflows without breaking tools. The standard development stages remain, but the agentic approach lets teams automate any stage as needed.

Beyond writing code, the system gives managers tools to track factory performance. Because every agent runs in the same environment, it is easy to compare performance metrics across different configurations and monitor total token spend. There are also self-improvement loops that optimize the entire process — automation of the process itself, not just the code.

Lloyd emphasizes that the goal is not to replace software engineers, but to give them a more convenient way to collaborate with the new agentic workforce. In his experience, plenty of tasks still require a human at the wheel. "We're automating about 30% of tasks, 30 to 35% on a weekly basis," Lloyd said, "and as the models improve, context grows and the harness improves, I think that number will go up over time."