AI Lab

From a 2017 chatbot to running stores with agents.

I've been curious about AI since I built a chatbot at Accenture in 2017. Today AI agents help me build and run most of what I make. This is where I show how, and what I learn.

To be clear about credit: agents write a lot of the code. I design, direct, review and decide.

  1. 2017Built a Dialogflow chatbot and improved its predictions with SVM.
  2. 2021Moved into analytics engineering with dbt, BigQuery and Looker.
  3. LatelyStarted building apps and running stores with AI coding agents.
  4. NowApps, stores and content run on agents, skills and connectors I set up.

The scorecard

apps shipped with agents
2 + 1
agent sessions
1,500+
hours building with agents
400+
custom skills I built
30+

My stack

  • Claude
  • Codex
  • Cursor
  • Antigravity
  • Gemini
  • ChatGPT
  • OpenRouter
  • Kie.AI
  • Databricks Genie
  • Vercel
  • Expo
  • Xcode
  • Android Studio

How I run it

  1. 01

    A question

    Usually from my own life, like how do I track my HRT.

  2. 02

    Agents plan and build

    Claude Code, Codex and Cursor turn a written plan into working code.

  3. 03

    Skills do the repeat work

    Custom skills and connectors handle listings, images, copy and reports.

  4. 04

    Things ship

    Apps, three Shopify stores, CodeCulture and free blog sites.

  5. 05

    Data says what worked

    Dashboards and daily KPIs, because I am still a data person.

Skills I built

  • Shirt designer

    Turns a topic into a tee concept, slogans and an image prompt.

    Built because: I had hundreds of designs to plan.

  • Reveal video maker

    Turns a screen recording and a mockup into a vertical reveal video.

    Built because: Posting every design by hand took forever.

  • Fulfillment sync

    Matches shipped print orders to the store and adds tracking.

    Built because: Orders were slipping between two systems.

  • Ad copy writer

    Writes ad copy for each stage of customer awareness.

    Built because: Good copy needs a framework, not vibes.

  • Listing finalizer

    Cleans up product titles, tags, categories and variants in bulk.

    Built because: Messy catalogs hurt search and sales.

  • ADHD output mode

    Makes every AI answer lead with the next action, in short numbered steps.

    Built because: My brain needs the first step, not an essay.

Experiments log

First experiment coming soon. Follow along on LinkedIn.

LinkedIn