Most of the forecasting I hear arrives in the same shape — a confident direction and no way to check it. Rates will do this. AI will do that. Boards will have to do the other. None of it comes with a test. If you cannot say what you would expect to see, you have not made a prediction; you have expressed a mood.
So my business outlook on capital, governance and AI is deliberately written as questions rather than calls. For each of the three shifts below I have tried to name the observation that would tell me the shift is structural, and the observation that would tell me it was a cycle wearing a costume. Both halves matter. A thesis that cannot fail is not useful to an investor or a director.
A prediction without a falsifier is just a mood
The discipline I am trying to hold is small and annoying: before forming a view, write down what would change it. It is the same discipline that makes diligence honest, and it is just as unpopular in a board room as it is in an investment committee, because it forces you to specify in advance the thing you would rather reinterpret later.
There is a second reason to work this way. Shifts that are real show up in artefacts — in operating plans, in board packs, in how a company prices work and who it hires. Shifts that are narrative show up in language only. If you know which artefact to look at, you can usually tell the two apart within a few quarters, and you can tell without access to anyone's proprietary data.
Capital: is the story the price of money, or who is supplying it?
Almost every conversation about investing trends collapses into the cost of money. That is the least interesting part. The cost of money moves and then moves back. The more consequential question is whether the composition of who provides capital, and on what terms, has shifted in a way that outlasts the next easing.
Here is what I actually want to know. When capital gets cheaper again, do sponsors do different deals, or do they do the same deals at higher prices? That distinction is the whole thing. If cheaper money simply re-inflates the assets everyone already wanted, nothing structural happened and the underwriting discipline of the tight years will evaporate on schedule. If cheaper money flows toward asset types or holding structures that were uninvestable before, something did change.
The second question follows from the first: are returns being underwritten to operations or to exit? Long holds force an honest answer. When the sale is far away and uncertain, the model has to earn its return from cash the business produces, not from a multiple someone else pays. I find that healthy. It is also checkable — you can read a plan and see whether the value creation is in the operating lines or in the terminal assumption, and whether anyone has underwritten the downside first or merely sensitised the upside.
The signal I am watching, then, is not a rate or a spread. It is whether the documents change. If investment memos in two years still carry the same exit-multiple logic with a new cover page, the shift was weather. If they carry cash-yield logic and longer operating roadmaps, it was climate.
Governance: boards are being asked to supervise what they cannot inspect
The remit of a board has widened faster than its capacity. Directors are now expected to oversee data handling, model use, security posture, supply concentration and reputation — domains where no non-executive can personally verify the state of things. The honest position is that the board is dependent on management for the evidence it uses to judge management.
Most of the governance trends I see are responses to that discomfort, and most of them are cosmetic. A new committee gets formed. A new standing item appears. The question I ask is simpler: has anything reached the room that a director could actually test? Assurance is a sentence. Evidence is something with a date, an owner and a number that could have come out wrong.
This is why I ask to observe a board meeting before I accept a seat. You learn more in ninety minutes of watching than in any amount of conversation about culture. You see whether disagreement is possible, whether the pack arrives early enough to be read, and whether decisions get made or deferred into impressions. A meeting that ends in impressions did not happen, and no amount of new committee structure fixes a room that works that way.
Chairing Berkeley Florida has sharpened this for me, because in a volunteer organisation you have no coercive power at all — the only governance that works is the kind people can see the logic of. The same framing applies to a company board facing an unfamiliar risk: ask how concentrated the exposure is, how reversible the decision is, and how quickly you would detect it going wrong. Those three questions travel across domains a director does not know well, which is precisely the situation boards are increasingly in.
So my test for whether governance is really changing is this: does the board start declining things? Widening remit without widening time means something is being done badly. A board that adds oversight of a new domain and removes nothing from the agenda has not taken the new domain seriously.
AI: production got cheap, verification did not
The cost of producing a draft, a design, a model or a first-pass analysis has fallen dramatically. The cost of confirming that any of it is correct has barely moved. That asymmetry is the most important fact in current AI business trends, and it predicts where the pain lands: not in creation, in review.
The question I put to management teams is deliberately narrow. Has anything in the unit economics changed? Not whether people use the tools — everyone uses the tools. Whether gross margin improved, or cycle time shortened, or a customer noticed. If a year of enthusiastic adoption leaves the margin structure identical, what was bought was personal productivity, which is pleasant and does not compound into enterprise value. I have written before about where AI actually earns its keep, and the honest answer is in a small number of places, repeatedly, rather than everywhere, thinly.
The second question is about the shape of the team. If output per person rises but the organisation now needs more senior reviewers to catch errors the junior layer used to catch while learning, you have shifted cost rather than removed it — and you may have broken the apprenticeship that produces the reviewers. That is a slow-acting risk, invisible for a while and expensive when it arrives.
The signal I am watching here is whether firms start measuring verified output rather than output. Anyone can show you volume. The interesting company is the one that can tell you what fraction of machine-assisted work shipped without rework, and whether that fraction is improving. Until a business can answer that, its AI story is an input story.
The same three questions, applied to all of it
Across capital, governance and AI I am really running one procedure. What would I expect to observe if this shift is real? What would I expect to observe if it is a cycle? And what is the cheapest observation that distinguishes the two, given that I will not get access to the data that would settle it conclusively?
The third question is the one most people skip. Plenty of theses are falsifiable in principle and untestable in practice, which makes them no better than opinions. The useful signals are usually mundane and close at hand: the structure of an operating plan, what fell off a board agenda, whether a margin moved. None of those require privileged information. All of them are harder to fake than a narrative.
Being early is fine; being unaccountable is not
I would rather hold a view early and be wrong on a schedule than hold one loudly and never settle it. Writing down the signal in advance costs almost nothing at the time and saves you from the slow drift where a thesis quietly rewrites itself to match whatever happened.
So I am not going to tell you where capital, governance or AI will be in two years. I will tell you what I intend to look at, and that I expect at least one of these three to disappoint me. The work for those of us who allocate capital and sit in board rooms is not to be right about the direction. It is to notice, earlier than comfort allows, when we are not.