Human-AI Interaction | Kate Blair (Director, IBM Research)

Listen to full episode:

Summary: This episode brings IBM Research's Kate Blair together with hosts Anant and Ed to unpack how human-AI collaboration needs to evolve as agents get more capable. They start by cutting through AI hype before digging into the real meat: why "prompt engineering" is giving way to "context engineering," whether natural language is even the right interface for AI going forward, and how systems can dangerously "launder" weak information through confident-sounding output. The conversation closes on a forward-looking note — a future where AI might prompt humans for input rather than the other way around — while Kate makes the case that judgment, empathy, and team leadership stay firmly human, even as AI accelerates the work around them.

Chapters:

0:00 — The episode topic: human-computer interaction in the age of agents

2:42 — Debate on AI hype and the MIT pilot study

14:01 — Kate Blair introduces her work on agentic systems

21:52 — The importance of context engineering

27:09 — Avoiding the “nightmare bicycle” in AI design

39:34 — Why interfaces should surface sources and uncertainty

48:30 — What AI won’t replace in management and people leadership

57:39 — The future of small models and specialized agents

Sound Bites:

“Natural language is not the interface for everything. It is a wonderful entry point, but it can’t be the destination.”