Approach
AI is a material, not a feature.
AI Product Architecture is product thinking applied to a new material: what the model sees, what it may decide, what gets checked, and where the result lands.
- 01InputA real task, inside a real workflow
- 02ContextOnly the data this decision needs
- 03RulesWhat must always be true
- 04ModelJudgement where it is cheap and useful
- 05ValidateChecked against the rules, again
- 06OutputBack into the flow, not a chat window
Three patterns from things I’ve built
- P1
AI inside the flow
Decision support lives where the decision happens: in the work order, not in a separate tool or a chat tab.
Seen in Schmitz Cargobull · In production - P2
Rules first, model second
When a mistake is not acceptable, validated rules set the boundaries and the model personalises inside them.
Seen in Infant nutrition - P3
Real use over demos
Prototype quickly with LLMs, then keep only what has measurable impact in the real world.
Seen in Decision automation · In development
How I think
What I believe.
- 01
Outcomes over output.
The best teams obsess over impact, not launch speed.
- 02
Clarity is a superpower.
A well-defined problem unlocks faster decisions across the whole organisation.
- 03
Users over opinions.
Five real conversations beat any internal debate.
- 04
Small bets, big learning.
Test cheaply to earn the right to bet big.
- 05
Teams build products.
Psychological safety and shared ownership come first.
- 06
Data informs. Vision leads.
Great products need judgement, taste and the nerve to go where the data hasn’t arrived yet.
“The best product isn’t the most complex one. It’s the one that solves the right problem in the simplest way.”
Have a problem like this? Let’s talk