Everything in my previous prose assumes a fairly stable answer to one question: how hard is it for someone to copy what we do? Entry barriers, switching costs, the threat of substitutes; Porter's whole machine runs on the friction of replication. AI quietly drags that friction toward zero. When a discrete, well scoped use case can be vibe coded out of existence over a weekend, the "go narrow, go deep" playbook I'd normally reach for starts to wobble.
I've had to relearn a few of my own defaults because of this, and they all come back to the same realisation: AI doesn't break the competition frameworks. It changes the inputs to them. Five of those reversals have stuck with me.
The first instinct under an AI threat shouldn't be a shiny new product line; it should be an audit of whether your existing EVC (WTP − Cost) still holds when a substitute can mimic your core in a chat window. Typeform's opening move was defensive: bring the conversational interface inside the core forms product before someone outside made that core redundant.
The logic is brutal and correct. If my differentiation can be replicated in a prompt box, my threat of substitutes has just spiked, and any offence I build is sitting on a leaking foundation. Harden the core conversational experience inside the product before someone outside makes it redundant.
Is our core still defensible against substitution or are we funding expansion on top of a moat that's already draining?
I keep seeing the same pattern: users who'd finish a job, cancel without a word, and reappear three months later not angry, just done. For a while I read that as dissatisfaction and reached for the usual levers: onboarding, pricing, support. None of it moved anything, because none of it was the problem.
What that quiet, no-complaint churn was actually telling me was something about recurring WTP the product had solved a one time job, not a recurring one. In competition terms, episodic need + low switching cost = structurally fragile retention, however slick the engagement mechanics get. You can't onboard your way out of a one-off use case.
Before we touch onboarding does this product solve a one off job or a recurring one? Where's the depth that earns a second, third, fourth return?
Every time you have had to narrow a set of vertical bets, the deciding factor is never TAM. It is shared infrastructure. Two bets that drew on the same plumbing are worth far more together than a bet that looks larger on paper but forces you to build a second, divergent stack alongside the first.
This is exactly the strategic-resources question from the internal-strengths side of the house. Overlapping infrastructure is a multiplier; diverging infrastructure is a tax you pay in compounding complexity. Map the feature requirements before you model the opportunity, the revenue number is seductive and the worst place to start.
Which of these bets share the resources we're already building? Where does the second bet subsidise the first, and where does it just split our focus?
Here's the reversal I least expected. Classic strategy states that the cost/uniqueness table I drew earlier pushes you toward narrow / focused differentiation: own one thing completely. I no longer think that's the safe default.
When any single, discrete feature can be cloned cheaply, the moat migrates to workflow coverage capture + enrichment + scoring + automated follow-up + downstream integration. Every additional surface I own raises the replication cost for a new entrant and lowers the threat of any one substitute. Breadth stops reading as unfocused and starts behaving like an entry barrier.
I won't pretend this kills depth; depth still manufactures WTP, and a product that does ten things badly loses to one that nails the job. But the shift is real, and it's a genuine inversion of the niche-differentiation default I used to take for granted: end-to-end workflow ownership can now be more defensible than single-feature excellence.
In our market, what's actually more replicable right now our depth in one feature, or our coverage across the whole workflow? Which barrier is really holding the line?