Blog
Original articles by Dr Darryl Carlton on the four subjects of this site: IT project delivery and assurance, expert evidence in failed IT projects, AI governance and delivery, and IT and AI contracting. Each article is published here in full. Writing already published elsewhere is listed separately below and in the publications library.
No article has been published here yet. New pieces will appear on this page and in the feed.
Foundation models are becoming infrastructure; the defensible work lies in engineering machine intelligence into systems that do real work, with the workflow, evaluation and context that a supplier's model cannot absorb.
A reading of a major consultancy's report on agentic shopping through the parable of the blind men and the elephant: each description of the market is accurate and none alone describes the whole.
On a government order restricting access to frontier AI models, and what such a decision reveals about competence at the top of government when the people deciding cannot tell what they are deciding about.
A response to the question of who owns the profile an AI assistant keeps on you: the answer, the author argues, is architectural, and turns on where that profile is stored and who controls it.
Argues that a merchant's terms for AI-agent purchases and Europe's payment rules converge on one requirement, a per-transaction record of what the buyer actually authorised, and that the card networks' new agent protocols verify intent at the payment end rather than where the buyer's evidence has to live.
Examines what Mastercard's Verifiable Intent and Visa's Trusted Agent Protocol actually do, layer by layer, and where each stops short of proving what an agent was instructed to do.
Where the evidence has to live when an AI agent buys the wrong thing: the instruction, the agent's reasoning under it, and the decision, all of which sit upstream of the payment. A practical account of evidentiary design for agent transactions.