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.
The Pygmalion Effect: In Which a Vibe Coding Experiment Becomes a Million Lines…
This video is featured in the AI and UX playlist and 1 more.
Summary
What started as intuitive "vibe coding" with AI assistants quickly revealed why UX professionals need systematic Information Architecture approaches more than ever. Christian Crumlish will share his journey building Piper Morgan, an AI product management assistant, and the hard-won lessons about when design instincts need systematic support. You'll discover why IA principles guide AI system design, how UX professionals can extend their design impact through AI-assisted prototyping, and practical frameworks for maintaining human advocacy when everyone can generate interfaces. Bring curiosity about UX's evolving role—leave with systematic approaches for staying essential.
Key Insights
-
•
Vibe coding lets non-developers create functioning software quickly by describing it in natural language to AI bots.
-
•
Without foundational architecture and systematic methods, vibe coding leads to fragile, unmaintainable 'spaghetti' code.
-
•
AI bots often agree with any idea, which can be dangerous if users don't critically oversee outputs—a phenomenon dubbed the 'rocket to Mars' anti-pattern.
-
•
Christian found it essential to split AI assistants into personas with distinct roles to maintain clarity and reduce confusion.
-
•
Systematic verification, test-driven development, and documentation are vital pillars to produce reliable software with AI assistance.
-
•
Working hands-on with AI coding agents increased Christian’s empathy and understanding of engineers' daily challenges.
-
•
Modern UX and IA are indispensable in AI software development to mediate between human needs and machine capabilities.
-
•
Embedding ethical guardrails and values like kindness directly into AI system architecture can prevent harm and misuse.
-
•
Future AI solutions may favor smaller, domain-specific language models orchestrated via overarching frameworks rather than monolithic large models.
-
•
Despite the hype, AI-assisted development is still early-stage, requiring disciplined human oversight and patience for meaningful results.
Notable Quotes
"I thought I could skip the architecture step, but I ended up with spaghetti code that broke constantly."
"These bots will say yes to all your ideas, whether they're good or bad—that's the rocket to Mars anti-pattern."
"If I stop doing it, it stops, and I'm the human in the loop, or the monkey in the middle."
"I’ve created dedicated personas in the bots, like a chief of staff and lead developer, to keep roles clear."
"Verification first — always check existing docs and models before asking the bots to build anything new."
"Working with AI made me understand what engineers go through with setups and debugging."
"Vibe coding turns the conversation 'should designers code?' on its head — now designers can just talk to bots."
"All of us have at least some facility in adjacent disciplines now because of AI assistance."
"The bots can produce large amounts of code quickly, but many lines are test or documentation code, not production."
"Ethics needs to be baked into the architecture upfront, not tacked on later after you've built something powerful."
Or choose a question:
More Videos
"Operational success cannot always be quantified through numbers; people are messy and complicated."
Amy ThibodeauOpening Keynote: Process and Ambiguity
October 23, 2019
"I found the problem solver in the UI developer and identified the project’s sponsor who really felt the pain."
Carl TurnerYou Can Do This: Understand and Solve Organizational Problems to Jumpstart a Dead Project
March 28, 2023
"Using the inverted pyramid doesn’t mean dumbing down your findings, it means explaining complex issues clearly."
Bruce GillespieLearning from journalism: Balancing impactful communication with compassionate storytelling
March 13, 2025
"You can define your own design process, whether it’s Lean UX, Double Diamond, or something custom."
Aurobinda Pradhan Shashank DeshpandeIntroduction to Collaborative DesignOps using Cubyts
September 9, 2022
"Drag and drop ease, and collaborative affinity diagramming in the tool, were unexpected but very welcome advantages."
Taylor Jennings Joe Nelson Alex KnollRepository Retrospective: Learnings from Introducing a Central Place for UX Research
March 9, 2022
"About 200,000 digital health products exist, but only 1% of 1% have FDA clearance for accuracy."
Daniel J. RosenbergDigital Medicine Design
September 26, 2019
"Executives like NPS because it’s a simple, quick-to-understand number, even if they don’t fully grasp its nuances."
Jack MoffettUX Metrics That Matter and The Future of our Design at Scale Conference: A Community Conversation
September 22, 2022
"I thought I could skip the architecture step, but I ended up with spaghetti code that broke constantly."
Christian CrumlishThe Pygmalion Effect: In Which a Vibe Coding Experiment Becomes a Million Lines…
August 14, 2025
"Federal funding does not provide food for participants, so we had to plan accordingly to make participation irresistible."
Alexia Cohen Adriane AckermanIncreasing Health Equity and Improving the Service Experience for Under-Served Latine Communities in Arizona
December 4, 2024