One-third of developers who worked on AI projects over the last 12 months never got a single one into production, and only 16% shipped all of them, according to a survey of 1,585 developers and tech workers by Udacity, which sells AI training courses. Most stalled projects made it fairly far, with 40% stopping at a working demo and 33% at a limited internal pilot. The most common reasons were business problems rather than technical ones, led by budget or resources at 37% and unclear return on investment at 35%, while only 22% blamed unreliable outputs and 14% blamed a lack of skills. Respondents ranked evaluation and testing, at 45%, and safety and compliance guardrails, at 44%, as the most important skills for moving AI into production.
