Warp’s new system is an out-of-the-box software factory for AI development
TechCrunch Russell Brandom
Warp launched Factories, a plug-and-play system for building AI 'software factory' agent pipelines. It's aimed at smaller teams who can't build this infrastructure themselves.
Based on reporting by TechCrunch, Russell Brandom — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Software factories are having a moment. The idea is simple enough: wrap the traditional stages of building software — triage, spec, implementation, review, verification — into a loop that agents can run through, automating chunks of engineering work along the way. Stripe has talked publicly about its own version, a system it calls minions. Ramp built something similar, a background agent that watches over code after it ships. Both did it themselves, with their own engineering resources.
Warp is betting that most companies don't have that luxury. On Tuesday the AI coding company rolled out Warp Factories, an infrastructure layer meant to hand smaller teams the scaffolding that Stripe and Ramp had to build from scratch. CEO Zack Lloyd framed the pitch bluntly to TechCrunch: running agents in the cloud, steering them mid-task, syncing their work back to a local environment, setting up shared memory and evals across the whole system — it's a serious infrastructure lift, and Warp Factories tries to hand it over already assembled.
The system follows that same five-stage development cycle, but treats every stage as something that can be automated by an agent rather than a person. Companies aren't locked into a single model or tool, either — Warp Factories works with Codex and Claude Code alike, and hooks into the ticketing and messaging tools teams already rely on, including Linear, Jira, Slack and Teams. The goal, per Warp, is to slot into existing workflows rather than force a new one.
Beyond just running agents, the system gives managers a way to actually watch the factory work. Because every agent operates in the same environment, teams can compare configurations against each other and keep tabs on token spend, which matters once you're running agents continuously rather than firing off one prompt at a time. Warp Factories also supports self-improvement loops, letting the system tune its own management over time.
What Warp Factories isn't, according to Lloyd, is a replacement for engineers. Even inside Warp's own operation, agents currently handle something like 30 to 35% of tasks on a weekly basis — a meaningful chunk, but nowhere near full autonomy. Lloyd expects that share to climb as models, context handling and the harnesses wrapped around them keep improving, but for now the humans are still very much in the loop.
My take — AI-written commentary, not fact-checked reporting
The honest number in this story is the 30 to 35% figure — that's a company selling agentic infrastructure admitting its own agents still can't carry most of the workload. That kind of candor is refreshing next to the usual full-autonomy marketing, and it's a useful reality check for any smaller team about to bolt an off-the-shelf factory onto its codebase expecting Stripe-level results overnight. The real value here isn't automation magic, it's Warp doing the unglamorous infrastructure work — memory, evals, token tracking — that most teams would rather not build themselves.
Read more about this at: TechCrunch