Warp’s New AI System Could Automate Software Development

AI coding company Warp is moving beyond tools that help individual developers write code. Its newest product, Warp Factories, is designed to let entire engineering organizations run fleets of AI coding agents across the software development lifecycle.

Launched on August 18, Warp Factories acts as an infrastructure layer where companies can automate tasks including issue triage, specification, implementation, code review and verification. Instead of asking developers to individually configure AI coding assistants on their laptops, Warp wants companies to run those agents through a centralized cloud system.

The concept is similar to a physical factory: work enters through one side, passes through specialized stages and eventually produces a finished output. In Warp's version, the workers are AI agents and the output is software.

From Coding Assistants to Coding Systems

Most AI coding products currently revolve around an individual developer interacting with an agent. A programmer gives Claude Code, Codex, Cursor or another tool a task and supervises what it produces.

Warp believes this approach becomes increasingly difficult to manage as companies deploy more agents.

Its new system instead creates an automated development pipeline in which different agents can handle different parts of a task. A triage agent can decide what needs to happen, another can prepare specifications, an implementation agent can make the changes and a review agent can inspect the resulting code.

The company describes the process roughly as:

Triage → Specification → Implementation → Review → Verification

Humans can still intervene at important checkpoints rather than allowing the entire process to operate without supervision.

Warp says companies can establish a basic factory capable of working on their own codebase in less than five minutes.

Companies Are Already Building Their Own Factories

Warp did not invent the idea of autonomous development pipelines.

Stripe has developed an internal system known as Minions to automate software-development work, while fintech company Ramp has experimented with background agents that continue working on code after deployment. These systems demonstrate how larger engineering teams are moving beyond simple code completion toward persistent agents that can independently handle defined tasks.

The problem is that building such infrastructure internally requires considerable engineering work.

Warp CEO Zach Lloyd told TechCrunch that companies need infrastructure for running agents in the cloud, steering active agents, maintaining shared memory and evaluating performance across different agents. Warp Factories is intended to provide those components without requiring each company to construct them from scratch.

That potentially makes the model more accessible to companies that don't have Stripe-sized engineering resources.

Warp Doesn't Want to Lock Companies Into One AI Model

One notable part of Warp's approach is that Factories isn't tied exclusively to Warp's own coding agent.

Companies can select different models and agent harnesses depending on the task. Warp says the infrastructure can work with systems including Codex and Claude Code, while its broader factory architecture is designed around swapping models as their cost, speed and performance change.

Factories can also connect with tools engineering teams already use, including GitHub, GitLab, Jira, Linear, Slack and Microsoft Teams.

Warp is treating the factory itself as code. Configurations are version-controlled, meaning companies can test changes, roll back configurations or experiment with different agents and models in much the same way they manage infrastructure through tools such as Terraform.

Warp Says About 30% of Its Own Work Is Already Automated

Perhaps the most interesting number from the launch comes from Warp itself.

The company says roughly 30% of its tasks are currently being automated through its own software factories. Lloyd separately told TechCrunch that the figure varies around 30% to 35% on a weekly basis.

That also shows the current limitation of the technology.

Warp isn't claiming that AI agents can replace an engineering organization. Complex or ambiguous development tasks still frequently require human involvement. Warp's earlier technical guidance estimated that around 20% to 30% of pull requests could currently be automated internally, with developers taking over when agents become stuck or a task needs hands-on testing.

The goal is for that percentage to rise as models and agent infrastructure improve.

Software Development Could Start Looking More Like Infrastructure

Warp Factories may be more significant for what it suggests about the direction of AI coding than for the product itself.

The first generation of coding AI focused largely on autocomplete. The next introduced conversational coding assistants. Products such as Claude Code and Codex then gave agents greater ability to work across entire repositories.

Software factories represent another step: multiple agents operating continuously across an organization's development workflow rather than waiting for a developer to prompt them individually.

Warp is also adding metrics around agent throughput, quality, token spending and ROI, allowing companies to compare different combinations of models and agents.

If this model works, engineering teams may eventually spend less time deciding whether to use an AI coding assistant and more time managing the automated systems that decide which agent should handle which development task, at what cost and with how much human oversight.

Warp Factories is currently in closed beta, with the company onboarding a limited number of organizations. Warp says selected customers can receive $10,000 in factory usage to begin testing the platform.