Rob Collie, a former Microsoft engineering leader and founder of a data and AI consulting firm, argues that most companies are approaching AI incorrectly. Based on his 13 years at Microsoft, where he was a founding engineer behind Power BI, Collie says the future is not simply buying more AI-powered software or rushing to build custom large language models. Instead, the real opportunity is teaching today's AI models how a business works and empowering the right people internally to do so. He notes that the first wave of enterprise AI adoption, which assumed companies could buy AI software and call it transformation, is not delivering on its promise. Collie agrees with Satya Nadella that every company will eventually build its own AI, but he says that is the "deep end of the pool" and not where most organizations should begin. The more practical "shallow end" involves treating commercial LLMs like new employees that need context, rather than expecting them to know a business automatically. Companies should build systems around existing models, connecting them to data, workflows, and software, using commercial LLMs as "Lego bricks." Collie's book about organizational AI strategy, "Fair Game," will be published on August 11.