This article addresses the risks of deploying AI within Learning and Development (L&D) functions before establishing reliable behavioural baselines. The author, Ross Dickie, argues that AI accelerates existing L&D strategies without interrogating their validity, meaning organisations with flawed data foundations will produce personalised solutions to the wrong problems at greater speed. The article identifies core weaknesses in common diagnostic tools — manager feedback distorted by office politics, self-assessments driven by confidence rather than capability, and engagement surveys that oversimplify complex dynamics. Dickie contends that a robust behavioural baseline requires triangulation across quantitative and qualitative methods, including structured interviews, behavioural surveys, focus groups, observational analytics, and control groups. The article concludes that AI's primary value lies in filtering and accelerating analysis of qualitative data, freeing practitioners for human-centred work such as observation and evaluation. The broader implication is that AI in L&D is a multiplier of existing strategy quality, not a corrective mechanism, and that human judgment remains essential for diagnosing whether training is the appropriate intervention at all. Key insights: AI in L&D amplifies existing strategy quality — flawed behavioural assumptions fed into AI systems will generate faster, more elaborate iterations of the wrong interventions. Common diagnostic inputs such as manager feedback, self-assessments, and engagement surveys carry inherent biases that are frequently treated as objective truth, distorting programme design. The concept of 'desirable difficulty' in learning design suggests that removing friction from the design process via AI can undermine the critical thinking necessary to identify the right problems. Practical takeaways: A multi-method behavioural baseline — combining structured interviews, observational analytics, focus groups, and where feasible control groups — produces more reliable diagnostic data than any single source. AI's immediate value in L&D is most defensible when applied to data filtering and thematic analysis of qualitative inputs, reinvesting saved time into direct employee observation and programme evaluation.