The lab
What I'm actually working on right now. No polished case studies, no portfolio gloss. This is the bench.
01 / agents you can trust
Agentic harnesses that check their work
Everyone has seen the demo where an AI agent does something magical. Far fewer have watched one survive contact with real work. The difference is the harness: the structure around the agent that verifies output, catches drift, and refuses to ship a guess.
I build and run these every day across my own products. Multiple models reviewing each other's work, independent verification gates before anything ships, and clear rules about what a machine decides versus what a person decides.
status: in daily use, always hardening
02 / answers with receipts
AI that quotes your records, not its imagination
Ask a chatbot a question and it will answer, right or wrong. For a business, that's poison. The fix isn't a smarter chatbot. It's wiring the assistant into your actual data systems, so every answer comes from a record you can point to, and "I don't know" is an acceptable response.
This is how my products work in production today: the assistant reads the database, the database holds the truth, and every claim is checkable. Accurate, verified, and auditable.
status: live in production
03 / people, still first
Teaching new coders to think in systems
AI can write the code now. What it can't do is decide what to build, how the pieces fit, or when to say no. So that's what I teach: one-on-one training that's less about memorising code and more about how people, systems, and workflows fit together.
Yes, you need to be able to read code. The real skill is translating what's happening back into plain human language: what problem are we solving, who is it for, what has to happen next, and where might it break?
Vibe coding helps you move faster, but it doesn't remove the need to think clearly. You still need structure, testing, edge cases, process, and handover. Good coding is not just technical. It's practical, human, and systems-aware.
status: one-on-one, limited spots
04 / where this comes from
Background, briefly
Ex-Microsoft engineer. Years of cross-Tasman software and government projects where privacy and compliance were never optional extras. Along the way I've led dev teams, admin and systems teams, run my own small business, and worked deep inside the not-for-profit world, where everyone wears at least three hats and the budget never stretches as far as the mission.
I've also judged barista championships across Australia, New Zealand, and Singapore. It sounds unrelated. It isn't: judging is calm assessment under pressure, against a standard, with someone's best work on the line. Every hat I've worn ends up in this one.
The result is range. I can sit with a sole trader or a corporate board, talk plain English or architecture, and understand the pressure of a tight budget because I've carried one. These days I build and run my own AI products, which keeps the thinking honest: if something on this page doesn't work, I'm the first person it bites.