💻 What the Research Actually Says About AI at Work and What It Doesn't

Most teams adopt AI tools the way they adopt resolutions: enthusiastically, and with no way to tell if they're working. Drawing on my own peer-reviewed research in NLP, evaluation, and human–AI interaction and putting it into practice at Microsoft, this session shows how data scientists design AI workflows that survive contact with measurement. We'll move past demos to the harder questions: where AI genuinely belongs in a workflow versus where it quietly manufactures rework, how to instrument a task so you can prove (or disprove) a productivity gain rather than assume one, and how to design human–AI handoffs that catch errors instead of compounding them. You'll leave with a concrete evaluation framework, a set of recurring failure patterns to watch for, and a checklist you can run against any AI workflow your team is weighing — starting Monday morning.

Delivered from the author of research that revolves around driving people's work streams with AI.