Warp Factories Puts Coding Agents on an Assembly Line
Warp opened a closed beta for cloud infrastructure that runs fleets of coding agents through a fixed five-stage pipeline — triage, spec, implement, review, verify — with a different model allowed at every step.
Warp launched Warp Factories in closed beta on Tuesday, cloud infrastructure that runs fleets of coding agents through a repeatable pipeline instead of leaving them to be babysat one terminal at a time. The pitch is aimed squarely at companies that want the “software factory” model without building the plumbing for it themselves.
The pipeline has five stages — triage, spec, implement, review and verify — and any of them can be handed to an agent. A ticket arrives, an agent triages it, another writes a spec, another implements, another reviews the diff, another verifies the result, and the run is monitored end to end. Warp’s argument is that most teams already have agents doing one or two of those steps in isolation; the hard part is the connective tissue between them.
Factories is model-agnostic, and that is the most interesting design decision in it. Each stage picks its own model and its own harness, so a team can run Claude Code for implementation, Codex for review, Cursor somewhere else, and an open-weight model on the cheap mechanical stages. It plugs into Linear and Jira for tickets and Slack or Teams for notifications, and adds the layers agent fleets tend to lack in practice: shared memory across agents, evaluation harnesses, token-spend analytics and self-improvement loops that feed results back into the pipeline.
CEO Zach Lloyd was notably unbreathless about how much of the work this actually takes over. By his account agents currently handle something like 30 to 35% of weekly tasks, with room to grow as models and context handling improve — a number that reads as a description of the present rather than a forecast. He also framed the product as a response to how much engineering the category demands, calling it “a huge infrastructure undertaking to do this right.”
The target customer is the mid-sized team, not the frontier lab. Large engineering organizations have spent the past year assembling internal versions of this; smaller companies have the same agents available and no platform team to wire them together. Warp is selling the wiring. Pricing is per agent run, with rates unpublished, and qualified organizations joining the closed beta get $10,000 of factory usage to start.
Whether an assembly line is the right metaphor for software is the open question. It works when tickets are well-specified and verification is cheap, which describes a real and growing share of maintenance work — and describes almost none of the ambiguous, design-heavy work that eats senior engineers’ weeks. Warp Factories is a bet that the first category is large enough to industrialize.
Want AI news before everyone else?
The morning's most important AI stories, straight to your inbox. No fluff.