Warp launches Factories platform for agent-driven software delivery
Warp has launched Warp Factories, an infrastructure platform intended to help companies build and operate AI software factories. The product provides an environment for deploying and managing agents across standard development stages, while Warp CEO Zack Lloyd says the company currently automates about 30% to 35% of its weekly tasks.
The system is aimed particularly at smaller companies that may lack the resources to assemble the underlying infrastructure themselves. Warp presents Factories as an out-of-the-box architecture for running agents in the cloud, steering their work, bringing work into local environments, maintaining memory across agents and applying evaluations across them.
Agent workflows across the software lifecycle
Warp Factories follows a familiar sequence: triage, specification, implementation, review and verification. The agent-based model means that any of those phases can be automated, rather than treating AI solely as a code-completion tool at the implementation stage.
Users can select their own coding models and harnesses. Warp says the platform works with both Codex and Claude Code. It also connects to workflow tools including Linear and Jira, as well as Slack and Teams, so teams can fit the system into existing ticketing and communication processes.
The software-factory approach is already being tested inside larger technology organizations. Stripe has described a “minions” system for automating development in its own codebase, while Ramp has developed a background agent to monitor code after deployment. The wider enterprise competition also includes Microsoft competes with OpenAI and Anthropic, where Microsoft’s positioning against OpenAI and Anthropic highlights the importance of AI tools in corporate technology stacks.
Measurement and human oversight remain central
Warp Factories is designed to provide managers with visibility into how an agent operation performs. Because agents run in the same environment, teams can compare performance metrics across configurations and monitor aggregate token spending. The platform also supports self-improvement loops intended to optimize the process that manages the agents.
Lloyd does not describe the product as a replacement for software engineers. He says many tasks still require a human operator, although he expects automation levels to rise as models, context and harnesses improve.
Business implication
For engineering leaders, the immediate decision is not whether to remove people from delivery, but whether a shared platform can make agent work measurable and governable. A practical rollout would begin with clearly bounded stages, then track output quality, human review needs and token costs before broadening automation.