I build systems and write about what I learn doing it.
My work spans software architecture, enterprise platforms and AI-assisted engineering. I am interested in where probabilistic reasoning helps, where conventional software should take over, and what it takes to trust the result.
Current thinking
LLMs are useful reasoning tools. Reliable engineering still needs systems around them.
The interesting question is not whether AI replaces traditional software. It is where probabilistic reasoning helps, where deterministic controls are better, and how we know the result is actually correct.
- ReasonUse models for ambiguity, exploration, synthesis and candidate solutions.
- ControlUse deterministic mechanisms when the rule is known and repeatability matters.
- VerifyJudge outcomes using evidence rather than confidence.
Latest writing
Why LLMs Can Seem So Dumb to Experienced Developers
Experienced developers often judge an answer using years of context they never supplied to the model. That does not excuse bad output, but it changes how we diagnose the failure.
Read the articleSelected work
The writing comes from building.
The useful ideas usually come from trying to make something work, finding where it breaks, and understanding why.
See the work- Enterprise architectureDesigning systems, standards and delivery patterns for complex organisations.
- AI-assisted engineeringExploring where models add leverage and where conventional software should take over.
- Software assuranceMaking important checks reproducible, observable and harder to hand-wave.