Experts say the high failure rate in AI adoption isn't a bug, but a feature (fortune.com)

At Fortune’s Most Powerful Women Summit, leaders from Microsoft, Bloomberg Beta and AI startup Sola pushed back on headlines about AI’s high enterprise failure rates, arguing those rates are a feature of a nascent, experimental technology rather than proof AI won’t deliver. Panelists refuted the doom narrative around an MIT stat that’s been interpreted as “95% of AI pilots fail,” clarifying the study found only about 5% of tested tools reach production — a result not far from historical success rates for major IT projects. They framed current outcomes as the inevitable product of massive, rapid experimentation: far more tools are being tried, many models will “hallucinate,” and only iterative testing uncovers useful patterns and production-ready workflows. The discussion highlighted practical implications for AI adoption: prioritize AI fluency and cross-functional pairing of engineers with domain experts, encourage hands-on “vibe coding” for nontraditional builders, and run small, safe pilots (including in non-sensitive contexts) to learn quickly. Technical takeaways included the rise of “agentic process automation” for back-office tasks, the need for human oversight to manage model hallucinations, and the importance of governance, monitoring and feedback loops to move pilots into production. Ultimately, the panel argued that organizational culture — leadership support for experimentation, tolerance for messy failure, and bottom-up engagement — matters more than any single model for turning AI experiments into sustained value.