Running daily · how I actually run my PM workflow
A system of scheduled AI agents that handle the operational half of my PM job, so my time goes to judgment instead of upkeep.
It started with a simple question: which parts of my day do I actually hate? Then I went and operationalized all of it — ticket updates, to-do lists, the manual tasks I never wanted to do from scratch. If it was busywork, I wanted an agent doing the first draft.
I did not want AI running everything. Every routine is human-in-the-loop by design: it drafts, proposes, or flags, and I approve. Blanket "one-click approve everything" is too much trust to put in a system that isn't deterministic, so the agents do the gathering and the drafting, and the judgment stays with me. It's the same principle I hold at work: AI should augment the human, not replace them.
Ten scheduled agents, each defined as a self-contained skill the agent reads and runs on its own schedule, wired into the tools I already live in (analytics, Slack, Notion, Linear, calendar). A few of them:
A weekly analytics digest that interprets what actually moved and why, not just charts.
A multi-agent PRD reviewer. Three agents — a skeptic, a customer advocate, and an engineer — poke holes in a PRD in parallel. Human review is still better and I still want it; I just don't want to wait for it. This gives me the perspectives I don't always have on tap — an engineer's eye, someone deeper in the codebase, a gut-check on whether I'm thinking about the users right — the moment I need them, so the PRD is already sharper before anyone else sees it.
A Friday weekly-update agent that crawls my week and drafts my update for me to edit and post.
Daily task scans that pull open asks out of Slack and flag which of tomorrow's meetings need prep.
Running daily. It's how I actually work now.