Sep 21, 2026 · 47 min
Warp turns Slack ideas into 2,000 monthly pull requests
How Warp ships 2,000 PRs a month with AI factories | Zach Lloyd (CEO, Warp)
Warp’s system shows both the promise and the constraint of AI-native software development: agents can accelerate production, but human review still limits scale.
- 1Warp coordinates agents, repositories, tools, and automations to turn Slack ideas into tested, reviewed, and merged software.
- 2The company measures speed, cost, quality, and human interactions per pull request to identify where its factory still needs intervention.
- 3Agent evaluation and self-improvement can raise quality, but model choice, context management, and human review remain central operating decisions.
Don't miss
Zach Lloyd demonstrates how Warp tracks agent-run costs and human interactions while explaining why review remains the limiting step at high volume.
The brief
Zach Lloyd describes Warp’s software factory as a coded system of agents, repositories, tools, and automations that moves ideas from Slack toward shipped software.
The factory’s output is striking—roughly 2,000 pull requests a month—but Warp tracks human interactions per pull request to expose where automation still depends on people.
Warp evaluates agent runs across speed, cost, quality, and failure modes, then uses sampled failures in a second loop aimed at improving future performance.
The central tension is operational, not theoretical: agents can produce code quickly, while review, correction, and judgment remain the bottleneck for safe deployment.
Lloyd extends the same approach beyond engineering, using AI with Figma, meeting analysis, and customer data to refine design, sales positioning, and lead generation.
Featuring
Listen to the full episode and explore every guest, topic, and moment on PodLume.