"Feature or a business?" is the venture test of whether a product's value is a capability an incumbent platform can absorb, or something structural the platform will never own.
The question is the standard challenge applied to any company built adjacent to a fast-moving platform: will the platform owner simply ship this capability and erase the company? A product is a feature when its worth reduces to a function the platform can replicate within a release cycle. It is a business when its worth lives in something the platform structurally will not hold — an accumulated data asset, a relationship, a switching cost, or a layer of trust.
The test bites hardest for tools built on frontier AI labs, where the release cadence of Anthropic, OpenAI, and Google can obsolete a thin wrapper overnight. Surviving the question means locating defensibility outside the model layer — in proprietary data, in workflow lock-in, or in a use case the labs have no incentive to own. A company that cannot name what the platform will never replicate is a feature waiting to be absorbed.
I've never tried to win 📝MythOS on technology — it's deliberately simple, and has been since 2017. That's exactly why it isn't a feature: what Anthropic or Google would have to replicate isn't the code, it's the user's own library, their sovereignty over that data, and a collaboration layer addictive enough that they don't want to leave. The labs ship capabilities; they don't ship someone a home for their thinking that they own. Features get absorbed — a compounding, self-owned body of knowledge does not.