01
Where should human approval sit in an AI workflow?
A practical framework for placing human approval before consequential AI actions without slowing every low-risk step.
Notes from applied AI work
Useful AI, explained plainly.
Practical guides for choosing, scoping, and operating AI systems—written from a product and engineering point of view.
RSS feed01 / Latest insights
Answers worth building from.
Each article starts with the decision a founder or team actually needs to make, then connects it to implementation and risk.
Read article
02
How to test a RAG system before employees rely on it
A practical RAG evaluation checklist for testing retrieval, evidence, access controls, failure behaviour, and operational readiness before launch.
03
AI agent or workflow automation? Start with the decision boundary
A practical way to decide whether a business process needs an agent, a deterministic automation, or a careful hybrid of both.
04
When should a business build a RAG system?
Use retrieval-augmented generation when answers must be grounded in changing, organisation-specific material—not simply because you have documents.
05
How to scope an AI automation project without buying a demo
A practical project brief for moving from an impressive prototype to an automation that survives real inputs, human review, and failure.