The Hidden Design Choices Behind Claude with Joel Lewenstein, Head of Product Design at Anthropic
Prompted: AI, People and the Creative Spark
Prompting as a Real Skill
Joel defends prompt engineering as an important, evolving skill and compares interacting with models to translation work.
Anthropic’s Head of Product Design Joel Lewenstein joins Cam Adams to explore the evolving relationship between humans and AI—how we build it, test it, and trust it.
Joel unpacks how his team designs Claude’s character and behaviors to earn users’ trust, and why the best AI products feel less like tools and more like collaborators. He shares how friction can fuel creativity, how Anthropic’s “AI vending machine” experiment helps them learn about human-AI interaction, and why designing for safety and alignment is as much an ethical mission as a product challenge.
Cam’s notes on Substack: promptedwithcam.substack.com
Key Quotes
- “Claude is a sense-making tool for me… I vomit into a document and I know it doesn't make any sense … And then Claude helps me make frameworks and concepts that I can then use in our day-to-day to guide the work.”
- “ What I did in a pre AI era was I would excitedly write a doc that I just was sure was perfect. I would show it to someone and they would just point out the obvious flaw, the obvious gap…Claude is my first reader on almost literally anything that I make now, and I just find that tremendously helpful”
- “ A lot of what we think about is how to build, how to train Claude, how to use the research methodologies to make Claude an entity that you really do want to be that vulnerable with. Something that is open-minded, curious, non-judgmental, non-simplistic and non reductive, and embraces the nuance and complexity of your life and of the world.”
Timestamps
00:30 — Welcome + Intro to Joel Lewenstein, Head of Product Design at Anthropic
01:45 — How a childhood coding game sparked Joel’s love of creative systems
05:19 — Why Joel uses Claude as his “first reader” for every idea
06:09 — Designing for trust, safety, and vulnerability
09:29 — The AI-powered vending machine experiment
11:21 — Why good friction makes better collaborators
15:07 — Volcanoes vs. Skyscrapers: the metaphor behind Anthropic’s design process
17:32 — How Anthropic uses Claude to build Claude
20:06 — Is Claude creative—or just trying to be creative?
24:16 — The future of teams and AI collaboration
28:43 — The real skill of the AI era: asking better questions
32:12 — Why prompt engineering still matters
36:45 — Designer reveal: Joel, Bella and AI’s collaboration
39:02 — Rapid fire: songs, kids, and AI-powered beer gardens
Links
Subscribe to the Prompted Newsletter: promptedwithcam.substack.com
See Joel, Bella and AI's creation: https://promptedpodcast.com/#episode-5
Cam on LinkedIn: https://www.linkedin.com/in/themaninblue/?skipRedirect=true
Joel Lewenstein on LinkedIn: https://www.linkedin.com/in/joel-lewenstein/
Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.
The AI-powered Podcast Player
