Summary
A 30-minute deep-dive into the building of the Rosenbot. We’ll get both hands-on practical and likely a bit philosophical. What does it take to build a useful AI assistant? What does it mean for a business, strategically? And how do we make sure we are building the future that we want, while doing all this? Take-aways: What does strategy look like in an AI world? What does eval-first mean? What the hell is going on deep inside an LLM? And what does all that mean for the future we want to build?
Key Insights
-
•
Generative AI is a new design material requiring fresh approaches distinct from prior UX methods.
-
•
Rose Bot leverages retrieval augmented generation to semantically search Rosenfeld’s extensive UX content.
-
•
Conversation logic beyond the LLM is essential to route, classify intents, and ensure safe responses.
-
•
Evaluation (eval) of AI is fundamentally challenging due to its non-deterministic input and output.
-
•
Effective evaluation requires combining human expert and end-user feedback alongside automated tools.
-
•
A third of a project’s budget and time should be dedicated solely to rigorous AI evaluation.
-
•
‘Prompt deep dive’ usability testing spends extended time on individual prompts to deeply understand interactions.
-
•
New tooling is emerging specifically for tracing conversations, prompt engineering, and observability in AI.
-
•
UX roles remain vital in AI development by inventing new research techniques and ensuring user-centered design.
-
•
Co-creation between users and technology defines how AI applications evolve and succeed or fail.
Notable Quotes
"When GPT came out, my kids immediately adopted it at school and couldn’t pry it from their dead hands."
"This AI stuff is a new design material, just like the internet or mobile was before."
"The Rose Bot has read everything—every piece of Rosenfeld’s knowledge—to help users access it."
"Building generative AI tools is not scary, it’s just very different from the last 20 years of building products."
"There’s a lot of steps behind the scenes to make conversations useful and safe."
"Eval is the engineering word for evaluation, but also what researchers and UX designers naturally do."
"Without proper evaluation, you’re just building a demo – demos are easy, quality production is hard."
"We invented the ‘prompt deep dive’ technique to spend lots of time on one prompt to deeply understand it."
"Co-creation between users and technology determines if this AI evolves to be useful or harmful."
"We should feel okay to throw away old assumptions and tooling and invent new techniques for this new world."
Or choose a question:
More Videos
"We tried hybrid sessions but quickly learned splitting virtual and in-person was better for experience."
Elizabeth Sklar Jessica ShengCo-creating research enablement with your tech org: a case study
March 10, 2026
"Adoption is the only thing that matters in innovation. If you’re not changing behavior, you haven’t innovated."
Saara Kamppari-Miller Nicole Bergstrom Shashi JainKey Metrics: Comparing Three Letter Acronym Metrics That Include the Word “Key”
November 13, 2024
"Working in pixels, we still have fingerprints. You all have a style. That’s a great thing."
Gretchen AndersonScaling the Human Center
June 8, 2017
"We had to charge not just project work but all the costs including vacation, rent, insurance, and bookkeeping."
Chelsea MauldinLet's Talk About Money
November 17, 2022
"We often talk about our customers, but we forget the other people involved: ourselves and our teams."
Lada GorlenkoTheme 3: Introduction
June 10, 2021
"Resilient design needs to bend without breaking — to survive stresses while still delivering value."
Craig VillamorResilient Enterprise Design
June 8, 2017
"Steve just talked about humans being humans in a world full of humans."
Susan Simon-DanielsWar Stories LIVE! Susan Simon-Daniels
March 30, 2020
"Great research doesn’t automatically have the intended impact because people may not want their minds changed."
Robin BeersBeyond Insights: Researchers as Organizational Change Catalysts
March 25, 2024
"We design with our biases and subjectivities, so we must work with diverse stakeholders to avoid tunnel vision."
Paul Pangaro, PhDSystems Disciplines: Table Stakes for 21st Century Organizations
June 6, 2023
