I would spend at least six months on the problem before writing a line of code. Talking to potential customers, not to validate an idea I already had, but to understand what actually goes wrong in their week. We started building too early, and it cost us: several changes of course later on, each one more expensive than the research would have been. Six months of listening feels unbearably slow when you are impatient to build. But a change of course years later costs far more than six months — it just doesn’t feel like it.