Latent Space: The AI Engineer Podcast · Nov 3, 2026 · 35m · 26m saved
With Zach Lloyd, CEO and Co-Founder of Warp
Photo via Latent Space on YouTube
Warp cut what it spends on AI coding agents from about $80 per merged pull request to $15 to $20, while its pull-request throughput rose 40% in a month. Zach Lloyd, CEO and co-founder of Warp, told Shawn Wang, who hosts Latent Space as swyx, that most companies claiming to run a "software factory" have only strung together a few automations. Warp's version is a system defined in code, measured on every run and tuned like any other product.
"You should define your factory in code... everything is like a concrete thing that you should be able to run at some point in time and measure how it's doing." — Zach Lloyd
Zach's evidence comes from Warp's own engineering team, which runs on the same factories the company sells to others.
The full interview is covered here so you can skip it. 35 minutes of audio, 9 minutes of reading.
Here are the 10 insights that matter.
Key Takeaways
A software factory should be a file of code, not a loose set of automations
Warp cut its cost per pull request from about $80 to $15–20
Throughput rose 40% in a month and the revert rate did not rise
Agents grading agents catch waste, such as tests nobody needs
A company's own tasks are a better benchmark than public leaderboards
One model was 95% correct at about $6 a task, another 93% at about $2
Teams of a few people should not build a factory at all, Zach says
Factory engineering is becoming its own job, separate from product engineering
1. A Factory Defined in Code
Warp defines each factory in a YAML file, a plain configuration format, that lists the code repositories, the tools agents can reach, the integrations and the workflow steps: triage, spec, implementation and review. Each factory maps to one product, such as a mobile app or a server app.
Zach Lloyd says a factory has two kinds of users
"There's actually sort of two personas. So there's someone who is building the product... And for those people, the way they use the factory is through their existing tools." — Zach Lloyd
Product engineers send work in through Slack, Linear, Jira, GitHub or an API, with no new interface to learn. Factory engineers build and tune the system itself.
