What I Would Be Careful Buying in 2026
Not a list of doomed categories. A list of patterns where the usual diligence answers stop being sufficient, together with the conditions under which each is perfectly investable.
This is a framework, not advice, and certainly not a forecast about specific companies. It sets out patterns that raise the burden of proof — cases where a conventional diligence can come back clean and still leave the important question unanswered.
Each pattern comes with the counter-case, because none of them is disqualifying. The point is to know which additional question has to be answered before a price is agreed.
1. Products whose value is mostly interface
Software whose contribution is presenting data pleasantly, generating documents, or providing a workflow around a simple underlying operation. This work is precisely what generative tools do well, and the customer-visible portion is largely reproducible.
2. Businesses selling billable hours for reproducible work
Services businesses whose revenue is priced by effort for work that is standardised: implementation, migration, documentation, first-line analysis, routine reporting. When effort falls, revenue priced on effort falls with it, regardless of how good the delivery is.
3. Categories a platform can reach without effort
Horizontal products, sold to technology buyers, sitting adjacent to a workflow a large platform already hosts, where “good enough” is a low bar. The competitor is a bundled feature, not another vendor.
4. Theses that depend on engineering savings becoming margin
A plan in which a substantial part of the return comes from AI-driven engineering productivity. The saving is usually real. Whether it survives competition is a separate question that the plan generally does not ask.
5. Data assets that are large rather than irreproducible
Volume described as a moat. Large datasets that are essentially a record of public or purchasable events, held by a company that has not built anything only it could build from them.
6. Companies that have declared victory on AI
A specific and slightly counter-intuitive pattern: businesses whose AI narrative is strong, whose adoption is genuine, and whose management is therefore confident that the disruption question is settled. Adoption is an operational fact. It says nothing about whether the product remains scarce.
What this does not mean
None of these patterns identifies a bad business, and treating them as exclusions would be a mistake in both directions: it would rule out good assets and it would provide false comfort about the ones that pass.
The more common error, in fact, is the opposite one. Overestimating disruption destroys durable advantages faster than any competitor does — a company that panics out of a defensible position has done the damage itself. The purpose of examining exposure carefully is to know precisely which parts are not exposed, and to keep investing in those with confidence.
Caution is not a view on a sector. It is a view on which questions have to be answered before the price is agreed.