Log in or create a free Rosenverse account to watch this video.
Log in Create free account100s of community videos are available to free members. Conference talks are generally available to Gold members.
Hands on AI #3: Claude Code for UX people
This video is featured in the Evals + Claude playlist.
Summary
If you’re a product manager, UX researcher, or any kind of designer, you know everyone is talking about agents. And perhaps the best agent out there right now is Claude Code. In this final talk in the series, we are not going to write code, but we will dig into Claude Code, and do a deep analysis of why it has taken off the way it has as an agent. What are the properties and the UX decisions that went into building it that make it such a success? What are some of the things we as UX people can learn from it. You will go away with a much better intuition around AI agents, how you can set them up yourself or build them for your customers in a more successful way, how to think about adoption and agent UX properties. It’s still very early in this agent journey - get ahead of the pack right now.
Key Insights
-
•
Claude Code is a minimalist AI agent that uses tools in a loop, enabled by post-training models for reasoning and tool use.
-
•
Despite a terminal UI being poor UX by traditional standards, Claude Code creates user feelings of respect, control, and trust through transparency and interaction design.
-
•
Agents like Claude Code operate by the model writing output tokens to its own prompt window, effectively chatting with itself while calling tools to perform actions.
-
•
Context and memory for agents are managed by simple markdown files that describe projects, workflows, or skills, allowing models to have relevant information without built-in memory.
-
•
Claude Code supports multimodal inputs, enabling tasks like analyzing images as well as text or code.
-
•
The 'decade of agents' implies ongoing rapid evolution with agents becoming central to product workflows over years rather than an instantaneous shift.
-
•
Users, rather than designers or developers, should be empowered to define workflows and prompts, shifting traditional roles and responsibilities.
-
•
Model memory is externalized via saved context text files or 'skills' that function like onboarding materials loaded on each session start.
-
•
There is no fixed best practice yet for agent workflows; experimentation and simplicity outperform over-engineering.
-
•
The evolving AI landscape challenges rigid job titles like 'designer' and 'developer,' encouraging more collaborative, role-fluid approaches.
Notable Quotes
"Cloud code wasn’t programmed to do this; it’s a very lightweight system where the model does most of the work."
"Agents are literally models using tools in a loop—they respond to themselves and the tools in chat-like conversations."
"Despite the worst user interface ever, cloud code is the most popular and successful agent that exists today."
"Trust the user and trust the model are the two lessons I’m getting over time."
"Your short term memory is the context window; too big, and the model has trouble accessing everything."
"Skills are like onboarding material for the model; you write them as simple markdown files that describe workflows or tools."
"If you build for the next model, even if it can hardly do something now, it will probably do it well in the future."
"Don’t overcomplicate workflows; the simplest thing that could possibly work is often best."
"We need to dramatically change the hats, the walls, and the workflow of how we work together in design and development."
"You can’t just be a designer adapting UX workflows assuming everybody else’s workflow stays the same; everything is changing with AI."
Or choose a question:
More Videos
"Without these rules and conventions, we wouldn’t be able to improvise."
Jim KalbachJazz Improvisation as a Model for Team Collaboration
November 6, 2017
"Evergreen content needs a strong perspective on why the topic matters along with practical examples on how to apply it."
Louis RosenfeldCoffee with Lou: Should You Write a (UX) Book?
March 7, 2024
"You don’t have to become a manager or director to advance; becoming a principal IC role is equally influential."
Catt Small Micah Bennett Brian Carr Jessica HarlleeWhat's Next for ICs: Exploring Staff and Principal Designer Roles
February 22, 2024
"We can’t be there for every behavior after the nudge, nor do we necessarily want to be."
Marieke McCloskeyUser Science: Product Analytics & User Research
March 11, 2021
"If you can’t see a human in a video, how do you know the entire conversation wasn’t fabricated?"
Llewyn Paine[Demo] Deploying AI doppelgangers to de-identify user research recordings
June 5, 2024
"Correlation doesn’t imply causation, but sometimes two things happening together do imply a causal mechanism."
Joshua NobleCasual Inference
October 6, 2023
"Choosing the right output depends on the desired outcome—from quick iterations to gaining senior leadership buy-in."
Sara LogelYour Colleagues are Your Users Too
March 29, 2023
"Sponsor sessions may require registration by email to prevent Zoom bombings and protect session integrity."
Bria Alexander Louis RosenfeldWelcome
January 8, 2024
"We often say don’t make me think. When that’s not possible, reuse and recycle those learnings."
Sam ProulxOnline Shopping: Designing an Accessible Experience
June 7, 2023