Twenty years on both sides of the table — underwriting industrial companies, then building one.

Running strategy, or running the company.

There is a particular seat that uses all of it at once: strategy and execution held by the same person — leading strategy, or leading the business — inside a company, or from the investing side of one. The shape of that seat matters more than the size of the organisation: close enough to the decision that direction and delivery are not separable, and far enough into the horizon that the answer takes years to arrive. It asks for three things. Two of them are seats, usually held by two different people. The third is a habit, and it is not usually anybody's job.

The preference is a permanent one. A time-boxed version of the same work — interim, operating partner, a board seat — is the same job with a duration attached, and equally of interest.

Long-term strategy, and the relationships that carry it.

Investment banking, corporate banking and building a company from zero look like three different jobs. The unit of work is the same in all three: a relationship measured in years, and a strategy that has to be financed before it pays.

Covering a large corporate is a decade-long position rather than a transaction — you inherit what your predecessors did and you hand on what you did. A credit relationship is renewed annually and remembered permanently; the file either holds or it does not, and no amount of narrative rounds it up. An enterprise sale into aerospace, automotive or infrastructure equipment runs for years before it pays, and the second contract depends entirely on how the first was handled. Partnerships with far larger counterparties get negotiated from the smaller side of the table, where the only real leverage is being the one who is right about the customer.

What these share is a horizon. The strategies that survive are the ones somebody wants to keep paying for, and relationships are what carry them across the years — because long-term value creation is the area under the curve, and both parties recognise it. That is the work I keep returning to, and the reason capital, industry and software have never felt to me like three separate careers.

The machine that builds the machine. Manufacturing for atoms, AI for bits.

Capital, risk, manufacturing, hardware and software — five things I spent twenty years learning separately — have become the same problem. The world still runs on the oldest frame there is, Land, Labour and Capital, but the mode of the inputs has changed: agency now also belongs to the machines, which makes them both an input to the machine that builds the machine and an output from it. The machine that builds the hardware the intelligence runs on — and everything else made of atoms — is manufacturing: harder than it looks, and under-rated as a source of both wealth creation and value destruction.

Risk is being re-priced along the way. A bank prices the downside and says no in committee; a startup treats risk as oxygen and lets portfolio mathematics carry the failures. When the cost of building collapses, decisions that used to be startup-grade bets become banker-grade experiments — and for the first time it is possible to underwrite like a credit officer and move like a founder at once. The risk does not disappear; it migrates, from "can we build it?" to "can we trust what we built?" — and beneath that sits the sharper question: what deserves to be built at all? When building is cheap, focus becomes expensive.

The full argument, at essay length →

The inputs of a flourishing society — and the Nordic hand of cards

Societies flourish on a short list of inputs: affordable energy, basic materials, industrial capacity — and now, increasingly, compute. Europe has concluded it must rebuild its position in all four. The Nordics hold an unusual hand for exactly that: a low-cost, low-carbon grid; genuine abundance in raw materials; relative political stability; ice-free ports; a corporate history built on engineering; and a domestic market so small that thinking in global market access has always been second nature rather than strategy.

It is also where the companies I find most interesting sit — industrial businesses with real assets, real order books and a decade of repositioning ahead of them. The essay makes the argument at length; the December 2025 op-ed is the same argument made in public, seven months before this site existed.

The through-line is curiosity

I have been a lifelong learner by default, and it is the habit underneath everything on this site. It filled the toolbox: fire-control mathematics, credit committees, factory floors, term sheets, board rooms, five countries' ways of doing business. I had circled AI for years — the Norna board seat was no coincidence — but as a non-coder I had always needed someone else's labour and capital to build anything. On 10 February 2026, the day Claude's collaborative coding tools were released, that stopped being true, and I have been building since.

The point was never to become a software company, and it is not what I want to do next. Interstice has never been a full-time occupation — it is how I engage with different kinds of company on different kinds of strategic problem, and it follows whatever the pivotal question of the moment happens to be. In 2026 that question is what AI actually does to a business, so that is where the last twelve months went. Understanding first-hand what these systems can and cannot do — the possibilities and, more usefully, the constraints — is becoming table stakes for allocating capital and attention inside an organisation. The building was the tuition, and I put in the hours. What the tuition bought was not speed but connection: most of what I now do well comes from carrying a discipline out of one world and into another where it is rare — credit-file provenance into software, fire-mission sequencing into agent orchestration, factory quality gates into AI governance. Orthogonal thinking is the dividend curiosity pays.

Having seen both eras from close range also settles a distinction most rooms still blur. The AI of the late 2010s — models, data, prediction — and agentic AI are different animals: different economics, different failure modes, different things that go quietly wrong. They are bought differently, governed differently and break differently. A board that treats them as one subject will mis-price both, and there is a whole industry with an interest in keeping the two words interchangeable.

— Oscar

Information, thoughts and opinions — a place to think out loud about the professional territory I find interesting, and a fuller answer than a CV can give to the question of what I am for. If something here is useful to your own thinking, take it. If you think I have it wrong somewhere, write to me — that is the conversation I enjoy most.