How to run an AI-first operation

The old CPG playbook isn't worth automating.

I’m Kat Hutcheson. A decade in global supply chain, most of it at PepsiCo. Data science at the University of Chicago. Now Head of Operations and Strategy at oac snacks, a 4-person CPG startup in Cork. Agents I built run our supply planning, Shopify data stack, and daily ops reporting, against a real P&L.

You can run the same systems: a kit you install yourself, or installed with me. I train ops teams, and coach founders, execs and ops leaders one to one, until AI is a tool they actually use day to day. The old playbook assumed big teams and big budgets. I’m rewriting it for companies with neither.

Work with me

Available now
  • Free fit call, 15 minutes
  • AI Fast Track, 1:1 for founders, execs and ops leaders
  • AI ops audit, credited toward an install
  • AI for Ops training · Speaking
See all ways to work with me

The agents

Two systems I run at oac every day, rebuilt for brands on Shopify and spreadsheets, no ERP. Kits first, or installed with me.

Coming for founders and lean teams

Not ready to hire anyone, including me? Tell me what would actually be useful to you, and I'll build that first.

The builds behind the products: the code, a leverage rating out of 10, and where each one failed. Wondering if I can build this? Start here.

  1. the agent framework I use, and how I built our supply planner

    How I go from a job I do by hand to an agent that does it: skills first, then stacked into one agent, with a written job description, an autonomy matrix, traces on every run and a second agent auditing the whole thing. Worked through with Batch, our supply planner.

    Read the build →

    agentssupply chainframework

    Leverage 8/10
  2. the dispatch console: from rules in our heads to an app anyone can open

    Dispatch rules at oac went through four stages: remembering them, posters on the wall, a sheet that only ran on my laptop, and now a web app anyone on the team can open. This is the build: the rules engine, the hashed PII matching, and the customs rule that breaks every UK freebie.

    Read the build →

    automationopsteam enablement

    Leverage 8/10
  3. oac's brain, part 1: the data layer

    Every agent I run sits on the same data layer: one Google Sheet workbook as the single source of truth, fed by scheduled Python syncs with health checks. The full wiring: source-by-source workarounds, the rules that keep the numbers honest, what failed first, and a where-to-start map for building your own.

    Read the build →

    data foundationsautomationagents

    Leverage 6/10
Browse all builds →