AI that does real work
I design, build, and ship AI products, and I train the leaders and teams who have to live with them. Twenty-five years of software, one product of my own in production, no deck between you and me.
Five things I now believe
Formed building KaizenShots on my own, April to July 2026. Argue with any of them on the call, that is what it is for.
The tool is fast, not senior
Agentic tooling wrote most of KaizenShots. It decided none of it. Architecture, data model, and product calls stayed with me. Treating the tool as senior is how projects rot.
Left alone, AI over-engineers
Abstractions nobody asked for, options nobody will use. Holding it back was a daily job, and it is the part nobody mentions when they show you the demo.
A prompt is code and needs a test
On KaizenShots every prompt is recorded in a manifest and checked in CI. Change one without declaring it and the build fails. Most AI products have no equivalent, and it shows in month three.
Two weeks beats six months of discovery
Anything that cannot be decided with working software in front of you was not the real question. The first clickable slice settles more arguments than any workshop.
Boring stacks ship AI products
Next.js, Postgres, a job queue, a payment provider. Someone has to own the system after me, and the model is the least interesting part of what they inherit.
The long version, with what broke, is in the KaizenShots case study.
What I do
I build AI products and I train the people who will run them. The Build Sprint below is where most people start.
AI Product Engineering
Agents, LLM apps, internal tools, and the full-stack around them. From prototype to production, working software every week.
See how I build →TrainingAI Product Development Training
Hands-on training in building with AI, for CXOs and the teams who ship. Where it helps, where it breaks, and how to take a product from idea to production.
See the program →The AI Build Sprint
Two weeks. One problem worth solving. A working prototype your team can click, break, and judge. Fixed scope, fixed price, agreed before I start.
Frame the problem, map your data, ship the first working slice.
Harden it, wire it to your stack, demo it with your team.
The prototype, the code, and an honest read on what a full build takes.
From problem to owned
Five steps, the same whether I am building your product or training your team. Outcome first. Tooling second.
Frame
I start with the business problem and the decision it serves. No tooling talk yet.
Map
Where AI actually helps, what to build first, and what to leave alone.
Build
Working software, every week. You shape it as it takes shape.
Ship
To production, with your team, measured against the goal we set.
Own
Your team runs the system. I stay close enough to help, far enough to let it be yours.
I ship what I sell
I am not reselling the idea of AI. I built my own product with it and I run my own delivery on it. What is on this page is what I use.
PRODUCTI build my own
KaizenShots, an AI mentor for photographers. Designed, built, and taken to production in twelve weeks with agentic tooling. Open to everyone since July 2026.
Read the case study →IN PRODUCTIONPrompts under test
On KaizenShots, every prompt is recorded in a manifest and checked in CI. Change one without declaring it and the build fails. That guardrail ships with every build I do.
How it works →Have a real problem worth solving?
Tell me where you are and what you are trying to ship. I reply within one business day.