Case studies
What AI looks like in a real business.
Four places I've put AI to work, and what we built in each.
By Seth Shoultes
Local service company
A+ Garage Doors: three years of AI, working in the background.
The problem. A garage door company needs to show up in every neighborhood it serves, and its staff need answers fast.
What we built. Research-backed pages for each neighborhood, and a report for every one showing where it ranks in Google, so we can see what's working. Apps for the warehouse and the sales team. A company knowledge base the team asks questions of, instead of hunting through documents.
3 years working together
How it works day to day. I've worked alongside the company for three years, so the tools grow with the business instead of being handed over once.
Software product
MemberPress.ai: an assistant that runs a membership site for you.
The problem. Running a membership site means a long list of small jobs: members, access, payments, content.
What we built. An AI assistant that does those jobs when you ask, plus the content, videos and launches around the product.
AI agent team
My own fleet of agents, and what I've learned running it.
The idea. Instead of one chatbot, a team of AI agents that each own a job: writing, checking, publishing, reporting back to me at every step.
Why it matters to you. On the Embedded AI Team plan, this is the team that works for your business. I run it for my own sites first, so I've already hit the problems.
I write about what goes wrong and what works in the essays.
Agent pipeline
Shipyard AI: five weeks of PRD in, production out.
The idea. Write down what you want as a product requirements document, and a team of AI agents debates it, plans it, builds it, tests it, reviews it and ships it. Each step is run by a named agent, and the pipeline checks the build before anything ships.
What we built. An autonomous pipeline of twelve steps, from a PRD check at the front to the ship at the end, that makes EmDash sites and plugins. A clean run takes about 35 to 45 minutes.
The numbers. The ship log lists 85 shipped PRDs, from April 6 to May 12, 2026. It also keeps a record of 12 failed ones, with the cause of each. In the EmDash plugin suite, EventDash works today and the others are in progress.
Search and AI answers
Garage Door Science: guides built to be found by search and by AI.
The problem. Homeowners search before they call. A guide site only helps if people can find it, and if AI assistants quote it when asked.
What we built. Plain, accurate garage door guides for homeowners, plus team apps for garage door companies.
What I'm watching. I'm tracking how the site shows up in Google and in AI answers, and I'll post what I find.
Want this for your business?
Tell me what you'd like AI to do, and I'll tell you where I'd start.