A proposal for AI products to target tasks with 50–70% completion rates

A user questions whether meeting the old 99.9% reliability bar too early limits an AI product’s ambition. They suggest building for capabilities today’s models can complete 50–70% of the time. In their proposed cycle, better models push completion rates toward 90% or higher after three to six months, prompting teams to expand what the product attempts until rates return to around 70%.