Mar 13, 2025 · 37m · y-combinator
Figma's Dylan Field: Exploring the idea maze, vibe coding, and the power of “locking in” · Y Combinator
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this Y Combinator interview hosted by Garry Tan, Figma co-founder and CEO Dylan Field discusses how artificial intelligence is transforming software development and elevating the strategic importance of human design. Field also shares Figma's founding journey, key internal product practices, and actionable advice for scaling startups in an AI-accelerated landscape.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The partners as informed peer 5.0 Guest teaching 3.0 Guest disagreement 1.1 The partners pushing back 1.2
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 12:19 Skepticism of dynamic per-user custom interfaces
Dylan openly rejects Gary's hypothesis of AI generating custom unique vertical UI per user, arguing learned shared interfaces like Snapchat prove the necessity of consistent UI paradigms.
Hardest push from the partners ▶ 5:18 Gary challenges the design readiness of prompt-generated appsGary pushes back on AI hype by emphasizing that prompt-based app generation remains in an uncanny valley where outputs fail basic functional and design standards.
Biggest teaching moment ▶ 7:05 Defining design as art applied to problem solvingDylan educates on why modern AI fails at design by explaining how diffusion handles art while LLMs handle logic, but models lack holistic context, user research, and cultural nuance.
The partners hold their own ▶ 6:09 Gary articulates architectural split in AI generation modelsGary demonstrates deep domain expertise by outlining the technological bifurcation between image diffusion models, CodeGen LLMs, and multimodal models unable to render vector SVG.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome Dylan Field: Contextualizing the AI Revolution | 4 | 3 | 1 | 1 | Gary Tan kicks off by setting up the broad context of AI's impact on software design. Dylan Field explains Figma's historic user base metrics and frames AI as an enabling tool that raises both the floor and ceiling rather than replacing human design judgment. | |
| Vibe Coding, Flow States, and Codebase Untangling | 5 | 3 | 2 | 2 | Gary Tan brings up 'vibe coding' and the uncanny valley of generated UI, while Dylan Field breaks down why fast feedback loops generate flow states despite leaving tangled codebases and unrefined designs. | |
| Evaluating AI Model Architectures: Art Versus Problem Solving | 6 | 4 | 1 | 1 | Gary displays technical depth contrasting diffusion models with CodeGen and multimodality. Dylan reframes design as 'art applied to problem solving,' explaining why separating generative images from reasoning models still falls short of cohesive UX. | |
| The Human Empathy Deficit in AI and Future Developer-Designer Roles | 6 | 3 | 1 | 1 | Gary shares AI researchers' characterization of LLMs as 'hyper-intelligent toasters' lacking agency. Dylan builds on this by arguing product judgment is an entirely distinct cognitive muscle from pure CS problem solving. | |
| Interface Evolution: Beyond Prompting to Spatial and Dynamic UIs | 5 | 4 | 3 | 2 | When Gary suggests CodeGen will lead to dynamically custom-generated vertical software per user, Dylan gently pushes back, citing Snapchat and social software learning curves to argue consistency is essential for user learnability. | |
| The Early Days: Exploring Drones, WebGL, and Meme Generators | 4 | 3 | 1 | 1 | Dylan narrates Figma's origin story from evaluating drones and WebGL to building a failed meme generator, noting how early canvas text rendering technical breakthroughs survived the pivot. | |
| Pivot to Cloud Design: Fireworks, Photo Editors, and Software Trends | 6 | 3 | 2 | 2 | Gary notes how conventional VC market-sizing would have dismissed Figma early on. Dylan reflects honestly on their unfocused seed deck and how bottom-up user behavior eventually pulled them into FigJam and Slides. | |
| Minimum Feature Sets, Landing Coda, and Field Emergency Fixes | 5 | 2 | 0 | 1 | Dylan recounts landing early customers like Coda and literally turning the car around on highway 280 to patch broken local font rendering in person before losing the account. | |
| Multiplayer Engine Innovation, Skepticism, and the Viral Design Party | 6 | 3 | 1 | 1 | Gary highlights operational transforms in multi-browser sync. Dylan describes the initial design community backlash against multiplayer design and how an impromptu viral 'design party' crashed their servers but validated the paradigm. | |
| Recognizing Product-Market Fit Signals and Hiring Hindsight | 4 | 4 | 1 | 1 | Dylan shares a story of drinking wine through a painfully slow user study at Coursera and receiving a 12-page spec doc, reflecting on his regret of not hiring faster once product-market pull emerged. | |
| Internal Dogfooding: Eliminating Friction to Prove Product Value | 5 | 2 | 0 | 1 | Both host and guest emphasize internal dogfooding as essential to software quality, with Dylan explaining how dogfooding revealed the pain of file locks and forced reloads in early browser prototypes. | |
| Crafting Design Culture: Balancing Power, Simplicity, and Maker Week | 4 | 3 | 1 | 1 | Dylan details how Figma creates its design culture by balancing power with approachability and running company-wide Maker Weeks that incubated core products like Figma Slides. | |
| Founder Scaling: The Self-Awareness and Delegation Loop | 5 | 3 | 0 | 1 | Dylan shares an operational loop framework for founder scaling: identifying current primary time sinks and hiring or automating replacements to avoid remaining in reactive fire-fighting mode. | |
| Speed of Execution: De-scoping Roadmaps and Rapid Delivery | 5 | 2 | 2 | 1 | Dylan argues forcefully against multi-year stealth builds in the AI era, stressing that founders must aggressively de-scope nine-month roadmaps to ship and learn rapidly. |
Statements from this episode (20)
Field: Design offers the opportunity to differentiate software in the AI era
“My point of view is that design has the opportunity to differentiate software at this age of AI, where we'll see whole new workflows emerge, where designers, engineers, product will work together in different ways than maybe they're doing today, or at least di…”
Field: Over 80% of Figma users are outside the United States
“So even from the start, over 80% of our weekly active users were outside the United States.”
Field: Two-thirds of Figma users are non-designers
“From the start, it's been like one-third designers consistently, and the other two-thirds are non-designers.”
Field: AI expands ideation breadth, but product exploration requires human depth
“My point of view is you get more breadth now in terms of the things that you can try, but it still requires a lot of depth to fully explore it.”
Tan: Prompt-to-app generation does not yet produce well-designed software
“We're sort of in this uncanny valley still where, you know, you can do prompt app, but prompt app doesn't give you an app that is well designed yet.”
Field: Rapid AI prototyping generates spaghetti code that requires untangling
“Like you can go and get locked in for three hours and make something cool. Will it be the design you want? Probably not, but also it will take some time to untangle the spaghetti mess of your code base. As it currently stands when you're going from start to pr…”
Field: AI struggles with design because diffusion and LLMs remain unmerged
“It's, I think maybe the reason why these models are not great design yet, it's like on the art side you've got diffusion, on the problem solving side you've got LMs, and it's not clear that people have figured out how to, you know, marry techniques together ye…”
Field: Software developers will soon call themselves designers
“I think people that call themselves developers today, and they call themselves designers, you know, in the near future. That doesn't mean they're not writing code. It just means that they're gonna be thinking more holistically about the product experience alon…”
Tan: AI researchers say scaling laws have not produced human-like agency
“I think we were hanging out with some AI researchers recently, and they were sort of disappointed that the models are not, you know, but they don't demonstrate as much human-like agency you know, even after all these innovations with, you know, climbing the sc…”
Field: Prompting alone represents the Telnet era of AI
“I think, like, prompting alone is sort of like we're in the telnet days of AI.”
Field: Figma's first text rendering came from a meme generator
“Our first text rendering in Figma came from the moon generator.”
Field: As software gets easier to build, design becomes increasingly important
“And then we, you know, saw fireworks get killed, and went, wait a second, maybe there's an opportunity there, and started to form the thesis of, as software gets easier to build, design becomes more important.”
Tan: Founders shouldn't use VC market sizing to pick startup ideas
“You shouldn't use like VC thinking to decide what to work on. Cause a lot of people Might have looked at how many designers are there in the country, and then multiplied that out by, like, whatever they thought they could spend. And it's like, oh, that, you kn…”
Field: Figma's original seed pitch was murky and unfocused
“If you looked at the seed pitch for Figma in June it was all over the place. Like, we were saying we're gonna do, like, a million things. None of it made sense. It was all very murky. So, yeah, I'm pretty impressed that Danny at Index was like, yes, let's go. …”
Field: Figma Built FigJam and Slides From Observed User Behaviors
“You know, FigJam, brainstorming. It was a behavior people were doing in Figma. We saw it, especially as the pandemic hit. It's like, okay, let's pull it out into this new product. And slides. Like, people are creating all these slides in Figma design. Okay, gr…”
Field: Figma's earliest adopters chose the tool because it lacked features
“When I look back at the first people that used Figma for real design work, they were minimalist, because they liked how few features we had. So it's kind of interesting that there is, like, this group of designers that are minimalist out there that, like a min…”
Field: Figma's first customers were Notion and Coda
“You know, like our first customers were Notion and Coda. Coda was called Krypton at the time.”
Field: Designers fiercely rejected Figma's 2016 multiplayer launch
“I remember when we finally launched the GA in October of 2016, I mean, the comments we got were like, if this is the future of design, I'm changing careers. You know, that person commented about it and said, a camel is a horse designed by committee. So it's li…”
Field: Founders should continuously hire to replace their biggest time-sinks
“Sort of the loop you're always in is be self-aware. What are you doing the most of right now? Then go replace yourself with that task if you've got the resources. If you don't have the resources, like, figure out how to get them either by being profitable, by …”
Field: Reject nine-month roadmaps and aggressively de-scope to ship faster
“When my team comes to me and says, hey, we've got like a nine-month roadmap on this feature, I'm like, what the fuck? Like, what can we do to de-scope or figure out how to get us out faster to start talking to people and showing to people”