Zach Lloyd: Engineers Will Build Software Factories

Why Zach Lloyd Says Software Engineering Is Becoming Factory Engineering

Zach Lloyd is the founder of Warp, an open source agentic development environment with over 800,000 active developers, and a former Google chief engineer who led Google Docs. He says he hasn't written a line of code in six months. His AI Engineer talk argues that the job is moving from operating agents interactively to building the system that produces the product.

The cloud software factory: Lloyd sees development moving from chat and autocomplete, to interactive agents, to automation. "Every project of significant size will have something similar. This loop is called the cloud software factory." Ideas arrive from a tracker, Slack, the terminal or monitoring. Agents triage, write specs for complex work, implement in cloud sandboxes, and review. Humans step in at a few defined points: approving specs, reviewing code, checking the product.

The parts of a factory: Lloyd lists the stages: input, triage, spec (a product spec for invariants and a technical spec for architecture), implementation, agent-first review, computer-use verification with screenshots and video, CI/CD, and monitoring that feeds back into the top. Under that sit a control plane for routing work, cloud sandboxes, model selection, and a data plane so agents remember and improve. Warp's public version is build.warp.dev, which shows every issue and the agent or contributor handling it.

Self-improvement loops: The factory is only as good as its feedback. "You don't just create a product. You create what creates the product." His example is a skill loop: a code-review agent leaves comments, a senior engineer corrects them, and an observer agent uses those corrections to improve the reviewer for the next run. Teams should also measure output, human time and token cost per unit shipped.

Open source as a moat when code is cheap: "It's very difficult to build a software business if software development is free." Cloning gets trivial, so a great product is no longer enough. Lloyd points to distribution, ecosystem, brand, data and capital, and argues building in the open builds an ecosystem. Warp open-sourced only after building a factory that handles the issue floods and low-quality PRs that used to make open source painful.

Build or buy: Asked whether this contradicts "focus on your product," Lloyd concedes the tension. A simple factory is easy to build, but one that scales needs control plane, sandboxes and data plane. Most organizations should focus on their core product, though tuning the factory to their domain still matters.

5 Takeaways from Zach Lloyd on Factory Engineering

• The unit of work is the pipeline - Engineers design triage, spec, implementation, review, verification and monitoring stages rather than writing the code inside them.
• Humans move to touch points - Spec approval, risk-based code review and product review stay human. Lloyd says human taste is what stops a factory from churning out things nobody needs.
• Loops make it compound - Observer agents that watch skills fail and improve them turn a static automation into a system that gets better.
• Measure like a factory - Track output, cost in human time and token spend, and improve both.
• Hiring is up, not down - Warp is hiring more than ever, looking for adaptable, product-focused problem solvers who can read specs and reason about architecture.

What the Software Factory Means for AI-Powered Organizations

Lloyd's factory is a template for any function, not only engineering: intake channels, triage, specs, execution agents, human checkpoints, verification, monitoring, and feedback loops that improve the agents. The scarce skill becomes designing and tuning that process, and knowing what is worth building. That is the same shift teams see wherever agents take over execution and people move to judgment.