(Part 1):AI-Native Product Management
I've automated ~16 hours/week of my PM work with a system of scheduled AI tasks and self-reinforcing feedback loops.
I built a set of "collection skills" that continuously pull information and update a shared context layer that all of my "process and delivery skills" read from. Those delivery skills handle the actual work: morning briefings, Linear triage, Slack responses etc. Every time the collection skills pull in new information, the delivery skills get smarter, and the feedback loop compounds. The stack is Claude Code scheduled tasks with MCP integrations; the diagram shows the full pipeline.
Building this has clarified for me that the fundamental PM job of deciding what to build and driving those decisions through to fruition isn't changing. But the actual work is shifting from executing individual tasks to designing and managing a system of AI workflows, then spending more time on the decision-making those workflows enable.
Funnily, this is also the closest my work has felt to chemical engineering since graduating chem eng school - designing probabilistic systems with feedback loops, tuning inputs and observing how outputs shift. Process engineering with language models instead of reactors 😅
My next step is to dramatically expand the number of delivery skills - open to ideas from anyone building similar systems!
