Notes on making AI pay.
Short essays on what to automate, what to measure and what to keep human. Written from engagements, not from the news.
Designing a consultation workflow with Voiceflow, Make and Airtable
What one nutrition consultation build shows about connecting a conversational interface, calculations, operational automation and reviewable outputs.
Read the essayDesigning a concierge with retrieval, live lookup and fallback
Lessons from a boutique hotel concierge about defining agent paths, grounding recommendations and planning a dependable hand-off when a dependency fails.
From Outlook meeting to reviewable legal documentation
A documented legal workflow for turning meeting context and a recording into drafts a solicitor can review, approve and own.
What to automate first: choosing the workflow that proves the case
How to pick the first workflow to hand to an AI agent: one with a counted cost, sharp edges, a human check and a number that moves within weeks.
Adoption is the outcome: why the last mile decides whether AI pays
A working AI system returns nothing until people stop doing the job the old way. What adoption actually requires, and how to know when it has happened.
The clarity problem: why understanding will always beat capability
Why understanding the problem matters more than raw AI capability, and how clarity drives results that last.
What is worth measuring
Choosing the metrics that actually matter when you bring AI into a business, and avoiding the ones that don't.
What should remain human
Where AI should take work off your plate, and where human judgement should stay firmly in the loop.
