How to Build a Software Factory That Improves Itself - Zach Lloyd, Warp
Summary
Most companies are stuck in interactive AI prompting when they should move to cloud-based software factories—sandboxed, governed systems where humans and agents collaborate at scale. The future belongs to teams measuring and improving their factories on proprietary data, not chasing benchmark leaderboards.
Key Takeaways
- Move agents from local laptops to cloud sandboxes to solve security and governance problems. Local agents risk compromising user permissions and create unauditable configurations across teams.
- Define your software factory as infrastructure in code with strict versioning. This enables testing different factory versions, lets coding agents update workflows, and treats the factory as a strategic asset.
- Build measurement-first systems on your own proprietary data instead of optimizing for public benchmarks like SWE-Bench. The best model and harness constantly change, so preserve optionality and flexibility.
- Implement AI sovereignty: own all data flowing in and out, run agents on your infrastructure with your own inference. This is critical for enterprises and preserves control over your competitive advantage.
- Structure workflows with humans responsible for task intake and triage, while agents handle spec refinement, implementation, and verification. This hybrid loop creates measurable improvement over time.
Related topics
Transcript Excerpt
[music] So, I'm going to be talking today about software factories and uh specifically like an infrastructure approach to software factories. Um, a little bit about me though before we start. So, I'm Zach. Uh, I've been an engineer for really for a long time at this point. I was a principal engineer at Google. I used to lead engineering on the Google Doc Suite. Helped scale that up to hundreds of millions and I think maybe now over a billion people who use it. I'm currently the founder and CEO at Warp. Uh, Warp is a company in the agentic development space. Our current focus is on helping companies build their own internal software factories. So, we have a platform called Warp Factories which we do that with. We're also uh pretty well known for Warp Terminal, which is an Agentic terminal. …
More from ai.engineer
- Your SKILL.md Is 70% Stuff the Model Already Knows - Nikolay Petrov, Intel
- Rethinking Claude Code Best Practices — Lydia Hallie, Anthropic
- Your Agent Doesn't Need an LLM for Every Decision - Raymond Weitekamp, OpenProse
- What Your Coding Agent Can't See in the Code - Tom Moor, Linear
- How to Get Frontier Coding Results at 70% Lower Cost - Nouha Dziri, Cohere