The Workflow Edge

• Factory argues competitive advantage comes from workflow structure, not just model intelligence; for instance, code review using GPT 5.5 in their harness costs approximately $1.70, compared to $5 to $6 for Opus 4.8 without that optimization.22:22
• Despite heavy automation, the process remains strictly human-in-the-loop; an engineer named Alvin focuses almost entirely on building lint and type-check rules to prevent agents from committing increasingly nuanced architectural errors.0:42
• The product suite includes 'Auto Wiki,' which creates structured knowledge and 4-to-6-minute video walkthroughs of codebases for human onboarding, and 'Missions,' a verification-driven orchestration layer that forces agents to complete specific hypothesis-checking hurdles before merging changes.46:55

Economics and Strategy

• Enterprise model routing is presented as a high-stakes economic problem, with some companies wasting hundreds of millions of dollars; Factory allows admins to use 'savings sliders' to tune quality-cost tradeoffs in real-time.0:01
• The company treats benchmarks as transient assets, noting a 3-to-6-month shelf life before they saturate or distort development, and currently prioritizes 'agent readiness' reports—which ensure codebases are deterministic enough to test—over raw agent output.55:35
• While Factory avoids building its own models, it acknowledges that frontier labs might eventually use server-side state or proprietary memory to create lock-in, though it views this as a high-risk strategic move that would contradict the providers' platform businesses.