A Team of One, Running Two Teams

Short summary

A solo developer built Omnist, an open-source Python library, in two weeks by running two distinct AI roles: a Research Assistant for conceptualization (allowed to disagree) and a DevOps team for materialization (never verified its own work). The key discipline was explicitly assigning roles upfront and requiring the AI to declare which hat it wore before acting. Mismatches between request and role surfaced immediately, catching poorly framed questions faster than any other method.

• •Two-role AI framework: Research Assistant (conceptualization, allowed to disagree) and DevOps (materialization, separate verification agent)
• •Explicit role declaration upfront caught misframed questions faster than any other signal
• •Research Assistant needed continuous human interaction; DevOps needed almost none once decisions were settled

Generated with AI, which can make mistakes.