Warp's CEO says to start a software factory with chores, not with the backlog

Lloyd defines a software factory as "agentic loops in the cloud that are measurable and improveable over time." He recommends starting with low-risk chores the team already dislikes. Warp's own factory fixes server crashes reported by Sentry, triages issues found while dogfooding, writes the weekly changelog and updates public docs. He advises against starting with agents working through the backlog, because unreviewed PRs would make the team conclude "this thing is just a slop cannon."

He also argues for letting engineers send work to the factory from Slack, Teams, Jira, Linear, GitHub or GitLab rather than a new web portal, and for local agents such as Claude Code or Codex to submit work to the factory through an MCP server. Adoption should be encouraged, not required. The first metric he suggests is the percentage of merged PRs that went through the factory, followed by throughput, cycle time and cost per PR.

Lloyd reports that Warp's cost per PR fell from $71 to $20 over six weeks, with a lower revert rate and 40% higher PR throughput than the previous month. These are Warp's figures about its own team and product, with no method or independent check given, and the article closes with a pitch for Warp Factories.

Why it matters: If you are deciding how to introduce a shared agent setup to a team, this is a usable order of operations: low-risk chores with a named owner for every output, then entry points in tools people already use, then metrics. Treat the cost and throughput numbers as one vendor's report, not a benchmark.