Here’s a small story that landed differently than expected. In June this year, Ramp’s Economic Lab published spending data showing Cursor had overtaken GitHub Copilot among its business customers. Copilot had seemed like a textbook PMF success — developer love, enterprise adoption, frictionless workflow fit. And then, fairly quietly, a faster and more developer-native competitor rewrote the market in months.
Not years. Months.
The definition that held for twenty years
Marc Andreessen’s canonical PMF framing — customers buying or using your product as fast as you can supply it — was a threshold. A destination. Something you crossed and then, largely, defended.
The implicit assumption baked into that definition is that the problem you solve stays roughly the same shape. That the market doesn’t fundamentally restructure around you while you’re executing. For a long time, that assumption held well enough.
The lens worth applying now is different. AI hasn’t just accelerated the race to PMF. It’s made the destination itself unstable.
When the floor moves
The pattern emerging across markets follows a similar shape. A product finds genuine fit. Retention is strong. Users are happy. And then a capability shift — often AI-native, often arriving faster than expected — restructures what users want, or makes something they previously paid for feel suddenly unnecessary.
Chegg is the case study nobody in product wants to sit with too long. Its core homework help business lost PMF within months of ChatGPT going mainstream. The problem space itself became obsolete faster than the company could adapt. A product that had done everything right found itself solving a problem users no longer needed help with.
The uncomfortable name for this is the AI PMF Paradox. AI makes fit simultaneously easier and harder. Easier because the speed of prototyping has collapsed. Harder because every interaction users have with increasingly capable AI tools recalibrates what they expect from every product. The baseline rises constantly, and without warning.
Four fits, not one
One of the more genuinely useful re-framings here is Reforge’s Brian Balfour and his Four Fits framework — extending PMF into a multi-dimensional view of what sustainable growth actually requires.
The four: Product-Market Fit (does a meaningful segment desperately need this?), Product-Channel Fit (is the product built for how customers discover things?), Channel-Model Fit (do the economics work with the chosen channels?), and Model-Market Fit (does the monetisation model align with how the market wants to buy?).
The key insight is that these dimensions are interdependent. When AI shifts one, the others often need revisiting too. Intercom’s Fin AI charging per resolved ticket — rather than per seat — only worked because AI restructured the value delivery entirely. A seat-based version of that same product would have struggled. The model unlocked the market.
PMF as a practice, not a certificate
The Lean Loop still works — as explored here before, build-measure-learn hasn’t expired, it just needs new questions. The same applies to PMF. Treating it as a diagnostic signal rather than a one-time certificate is exactly what the current environment demands.
The signal worth chasing, as Bessemer Venture Partners notes in their State of AI research, is second-bite usage: does the user return? Do they build habits? Does the product become part of a workflow, or does it remain an interesting experiment? Early adoption numbers can be deceptive when experimentation budgets, not durable demand, drive the initial traction.
The conversation worth having in product rooms right now isn’t “have we hit PMF?” It’s something more like: “Which of our four dimensions is quietly under pressure — and from which direction?”
Looking at a product you know well — which fit dimension feels most exposed to an AI-native competitor arriving in the next twelve months?
Let’s keep learning — together.
harish.balasubramanian 
Share your thoughts