John K. Thompson has spent 38 years in AI, data and analytics at IBM, Dell and EY, where he led AI and built what EY called the world's largest private, secure gen AI environment. He grew up in Marlette, in Michigan's Thumb, has written five books on AI and teaches at the University of Michigan.
Ryan Dolley and Aaron Wilkerson talk with him about AI from the Midwest, where companies make trucks, coats and chemicals, not software. A little more reality, a little less hype.
We cover:
Why everybody feels behind on AI, and why that means nobody is
Context engineering as nested prompts: company, department, process and personal
What an agent harness actually is. It's control and governance, inner and outer loops you've probably already built
The five families of AI – generative, predictive, symbolic, causal and physical – and why stringing them together beats one giant agent
What gen AI is genuinely good at, and where people misuse it
Why AGI and recursive self-improvement aren't two weeks away
One practical money-saver: drop back a model version or two and cut your token bill
John's advice for students: be AI fluent, not just AI literate
