Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp

Summary

Zach Lloyd argues that software engineering is shifting from developers working directly with code toward “factory engineering”: designing and managing systems that use agents to move software work through a repeatable lifecycle. He expects significant projects—and eventually most companies and open-source projects—to build or use such factories, much as CI/CD became standard.

How the software factory works

A factory takes in ideas from sources such as task trackers, chat, terminals or IDEs, and monitoring systems, then routes them through stages:

1. Triage: Agents assess incoming issues and may implement simple, well-defined tasks directly.
2. Specifications: For more complex work, agents produce a product specification describing desired behavior and invariants, along with a technical specification describing the architecture and code changes.
3. Implementation: An agent, potentially running in a cloud environment, writes the code.
4. Review: Agents can perform an initial review; people step in when their judgment or risk assessment is needed.
5. Verification: Automated checks and, for user interfaces, computer-based testing can produce screenshots or videos. CI/CD remains part of the process.
6. Monitoring: Agents observe released software and feed information about failures and usage back into the workflow.

Humans remain involved at important decision points, including specification review, code review, and product evaluation.

Building and improving the factory

A factory needs more than automation. Lloyd highlights the need for context and skills, ways to involve people at the right times, and systems that learn from past work.

He describes a “skill loop”: observer agents examine how other agents perform and where they fail, then help improve their skills. For example, an observer could use a senior engineer’s corrections to a code-review agent’s comments to improve future reviews.

Lloyd also recommends measuring the factory’s output and efficiency, including how much software it delivers and the human and token costs involved. A scaled system may include:

• A control plane to distribute work
• Cloud sandboxes where tasks run
• Choices about which agents and models to use
• A data layer that lets agents retain information and improve over time

Organizations can build parts themselves, but Lloyd advises most to prioritize their core product rather than reinventing all the underlying infrastructure.

Why Warp went open source

Lloyd says Warp opened its source after five years of closed development partly to create a public software factory. Its build.warp.dev site shows issues moving through the system, their status, and which agents or contributors are handling them. Warp’s automations are intended to help manage common open-source burdens such as high volumes of issues, low-quality pull requests, and review workload.

His broader argument is that AI makes software cheaper to build—and easier to clone—so a good product alone may not be enough to sustain a software business. Building openly can help a company develop an ecosystem, community, and brand. He reports that Warp had more than 60,000 GitHub stars, several hundred contributors, and over 800,000 active developers.

Advice and implications

Lloyd expects engineers to write less code but deliver more. He sees factory design and agent-system design as new engineering challenges, while acknowledging that people who primarily enjoy writing code may find the shift difficult.

For new graduates, he recommends adaptability, critical thinking, fast learning, and a solid understanding of systems, architecture, code, and specifications.

On product quality, he emphasizes that human taste and product judgment remain essential: automating a process is not useful if it produces something customers do not want.

Resources mentioned

• An open-source GitHub repository for experimenting with factory agents, including agents for triage and specification writing. Lloyd says it uses Warp’s agent platform but does not require it.
• build.warp.dev, an example of a public factory tracking issues and work in progress.

Main speaker

• Zach Lloyd — Founder of Warp and former Google chief engineer. He discusses agent-driven software development, Warp’s move to open source, and the design of software factories.