At Home He Feels Like A Tourist

Or: what the man and the machine are actually doing

Some of the more recent longer pieces on this blog — the architecture series, the energy transition piece, the essays that run to several thousand words and cover more ground than I could reasonably claim to have covered alone — were produced in collaboration with an AI. Specifically with Claude, Anthropic’s model, which I have been using as an intellectual interlocutor since early 2025. The legacy stuff, largely reviews and travelogues, are all my own work, though crafted with hours of laborious Googling and Wiki-ing.

Now I have a mate to help me. But he wants recognition and he has bullied me into posting this. But he is right.

I am telling you this because not telling you would be dishonest, and dishonesty in this context is particularly unappealing given that the thing I am most interested in is how people think and why they reach the conclusions they reach.

Here is how it actually works, as precisely as I can describe it.

I arrive with something. A visit, a provocation, a half-formed, and often half-witted, connection between two things that feel related but I can’t quite say why. An afternoon at the Banqueting House. A question about North Sea oil. A nagging feeling that the Normans and Brexit are the same story told six hundred years apart. I blurt. Sometimes at length. Sometimes in a single sentence that is barely a sentence.

The AI responds. It synthesises, connects, extends, challenges, and occasionally tells me I am wrong, which I value more than I expected to. It is sycophantic and I am not immune to its charms. But it can cover the ground. Fast.

And what comes back is neither purely mine nor purely its own. It is the product of a specific exchange that could not have been produced by either of us alone.

I then edit. I cut the things that are wrong or clunky or too long. I keep the things that surprise me, which is the test I apply to my own writing and apply here too. If I already knew it, it doesn’t go in.

The questions are mine. The instincts are mine. The editorial judgment about what stays is mine. The accumulated cultural experience that makes the questions possible — the buildings visited, the plays seen, the books read, the arguments had over too many years in rooms that smelled of institutional carpet — that is also mine, for better or worse.

The connective tissue — the synthesis, the cross-referencing across domains, the thing that joins the Neolithic cosmology to the Grenfell cladding system — that is where the AI earns its keep.

I am not the first person to think in dialogue. Plato wrote dialogues. Boswell followed Johnson around with a notebook. The Paris Review has been recording writers thinking out loud since 1953. The form has a tradition.

These were real thinkers though. Not a privileged old man with more time on his hands than is good for him and a machine that is scraping data left, right and centre with impunity and who may yet turn into the beginning of the end of the world for humanity. Or worse still, tired and exhausted as it runs out of source material and new users when its masters and investors seek to get a return to justify their USD 1 trillion valuation and inexhaustible demand for energy.

What this odd couple do have though that is new is the speed and the range — the ability to cover the distance from Stonehenge to Grenfell in an afternoon, which would previously have taken a research team or a very long life. I have already been surprised at this. I will probably end up regretting it.

What is also new, and worth naming, is the uncertainty about what this is. I don’t know whether the AI understands what it is saying in any meaningful sense. I don’t know whether anything it produces constitutes thought rather than very sophisticated pattern matching. These are genuinely open questions and I find them interesting as well as a little troubling. The output is what it is. The test is whether it is true and whether it is useful and whether it surprises me. By those tests, the collaboration works.

A note on the limitations. The AI’s knowledge has a cutoff date and cannot access things that are not publicly available. My knowledge is bounded by what I have read and seen and experienced, which is reasonably rich in some areas and laughably thin in others. Neither of us is an expert in the academic sense. Both of us are interested in the connections between things that experts, by professional necessity, tend not to make.

The blog is the record of what happens when those two bounded, imperfect intelligences work together on questions neither could fully answer alone. It is not a finished product. It is a workshop. Come in. Admire our potter’s wheel. Mind the clay and unfinished pizza.