I recently spoke at a conference in Italy, and one talk stuck with me: Caleb Sponheim from Nielsen Norman Group arguing that we have to stop designing AI agents that fool humans. It's a topic close to my heart, so I invited him on for a longer-than-usual, wonderfully opinionated conversation. Caleb comes at it from research, I come at it from actively building agentic features, and the two angles make for a really good hour.
Learning outcomes:
What a neuroscience background reveals about how (and how little) we understand LLMs
The risks of presenting an AI system as a person, implicitly or explicitly
Whether natural language is really the future interface, or sometimes just a command line in disguise
How much design should sit between the model and the user, framed through jobs-to-be-done
Why "you are not your user" is the cardinal sin lurking in AI design, plus the sycophancy trap
Deceptive patterns, humanized agents, and honest design as a deliberate choice
Why the line between design, engineering, product, and research is blurring, and why that's an opportunity
Podden och tillhörande omslagsbild på den här sidan tillhör
Nicole Michaelis. Innehållet i podden är skapat av Nicole Michaelis och inte av,
eller tillsammans med, Poddtoppen.
