AI progress isn't just about better models – data access determines which applications thrive. Coding agents like Cursor improved rapidly because developers could adopt them instantly, creating feedback loops that fueled model training. This data flywheel outpaces theoretical advantages.

Enterprise tools face different dynamics. Support and IT automation require organizational adoption, slowing deployment but building deeper customer-specific insights. Startups like Sierra and Decagon gain traction not through viral growth, but by embedding into workflows, learning each client’s unique processes over time.

Products that are easy to adopt *and* solve face intense competition from big labs like OpenAI and Google. These firms gather massive data sets and can subsidize costs, making it hard for startups to compete. As office productivity tools improve, expect platform consolidation unless customization creates real switching costs.