It’s never been easier to build - and never been harder to build something that holds.
As AI capabilities evolve, engineering velocity accelerates and is no longer the primary constraint. But beneath that acceleration, something more fundamental is happening: the problem space itself won’t sit still.
Expectations are shifting in real time. Interfaces are collapsing into natural language. Users continuously re-trade accuracy, control, cost, and independence. Even the model providers aren’t operating against a stable plan - discovering the product as they go. Which means we’re all building on moving ground.
Drawing from a year of building an AI-powered platform for code understanding, this talk argues that many products fail not because they can’t keep up with the technology, but because they’re built on assumptions that expire within months.
Through hard-earned lessons: what worked and what didn’t, we’ll explore how to build for that instability, and rethink what it means to create sustainable value in an environment defined by constant change.