Mass AI can answer almost any specialist question. That is precisely what makes it so easy to mistake for a specialist. The Decision-Maker asks a particular question — about a collection, a structure, a lineage, a foundation — and a credible-sounding response arrives. The answer is fluent. It uses the right vocabulary. It cites the right references. It satisfies the form of the question. On the strength of that fluency, the instrument is taken to be specialist. It is not. It

The AI powered instrument that does not know who you are will always meet you at the level of the abstraction you provide. Most AI is built for a general user. The conversation begins with translation — the Decision-Maker reducing the complexity of what they actually face into terms the instrument can process. The stakes get simplified. The context gets summarised. The things that cannot quite be said directly get left out. The exchange happens, but it happens at a distance f

A general-purpose AI is trained on the world. A Bespoke Confidante is trained on a person. Mass AI is built by drawing on vast quantities of human knowledge — text, conversation, data — processed at scale by the companies that build it. What emerges is an extraordinarily capable instrument that knows about everyone and everything in aggregate. By construction, it knows nothing about any one person in particular. A Bespoke AI Confidante is built on the same kind of underlying

A question is one thing. A relationship is another. Mass AI is built around the exchange. A question goes in, an answer comes out, the exchange completes. The next conversation begins from something close to a clean slate. This is not a flaw — it is the structural reality of a tool serving millions of users at once. Continuity at that scale is not what the tool is for. A Bespoke AI Confidante is built around continuity. It holds context across time. It remembers what was said

Every AI has someone it is built for. The mass AI tools most of us use every day are built for everyone, which is to say they are built by companies, for users at scale, within the constraints of business models and policies that apply to millions of people simultaneously. The AI is loyal to the entity that built it, and to the user insofar as that loyalty does not conflict with the first. This is not hidden. It is simply how the technology works at scale. A Bespoke AI Confid

Most AI is built for everyone. That is its strength, and also its limitation. A tool built for everyone cannot be built for anyone in particular. It answers the question you ask in the form most likely to satisfy the broadest user. It has no memory of who you are. No understanding of what you are actually wrestling with. No loyalty, because loyalty requires knowing the person it is given to. There is another kind of AI emerging quietly, for a different purpose. Built not for

On the ninth of April 2026, in honour of the United States’ semiquincentennial — the 250th anniversary of the Declaration of Independence — Forbes published its ranking of the 250 greatest living self-made Americans. The list was assembled in stages. The magazine first mined its own 109-year archive for what it described as classic tales of entrepreneurial capitalism. Its current beat reporters were asked to suggest candidates. The two together — the institutional record and

The individual whose thinking carries consequence has long worked through questions of consequence in the company of some form of interlocutor. The form has varied across time and place — a counsel, a physician, a confessor, a trusted friend of considered judgement, an advisor whose long familiarity allowed shorthand rather than explanation. What has remained steady is the underlying principle. Thinking of a certain weight is not best conducted alone, and the formation of the