Dylan Field | Sourcery Summary

Podcast: Sourcery | Guest: Dylan Field (Co-Founder & CEO, Figma) | Host: Molly O'Shea

1. Key Themes

The AI Taste Gap: Differentiation Through Design in a Templated World

The central thesis of the conversation is that AI has commoditized execution, making taste, point-of-view, and creative risk-taking the new competitive moats. Dylan warns that defaulting to AI outputs without human steering produces generic, averaged results that will be invisible in an increasingly crowded attention economy.

"If you just use an LLM for everything and have it literally tell you how to go build a thing, how to go design a thing, you're going to get something that it wants to give you. And it's kind of diverged that point from whatever's in your head to however it processed it... people that are just like, yeah, design me the thing — you're going to get the average and it's not going to stand out." 00:16:06

Design Is Expanding, Not Dying

Despite the "design is dead" narrative, Dylan argues design is evolving into a broader discipline that now encompasses user empathy, interaction paradigms, system thinking, and even code. Engineers are entering design workflows through tools like Figma, and the scope of what designers are asked to do is actually growing.

"Designers are giving us the feedback we used to get from developers now. And so I think that if we can kind of do both of those and find ways to help people create their own tools on the platform, share them with their team and get value out of that, that'll be really awesome." 00:41:01

Human Judgment Cannot Be Replicated by Raw Model Intelligence

Dylan makes a sharp distinction between verifiable formal domains (math, code) where scaling intelligence works, and judgment/design domains where adding IQ points to a model doesn't translate to capability. This has profound implications for which jobs and workflows are actually at risk.

"On formal domains, things that are verifiable, you can get to amazing outcomes with enough IQ points. And yet judgment is somehow not purely correlated with just adding IQ points to the model. Design certainly does not seem to be correlated with adding IQ points to the model." 00:27:22

The Platform Flywheel: Data Liquidity as Competitive Moat

Dylan outlines his framework for which enterprise software will survive the AI wave: network effects, distribution breadth, and crucially, data liquidity flywheels. He argues that boring, non-obvious verticals with embedded data context will be the most defensible.

"The data liquidity that you can establish and how that can create context, how that context can create capability in a system that is its own flywheel. And if you get that flywheel right, I think it's very value accruing to the customer in a way that can be extremely useful." 00:39:43

"I would say that boring software is not to be undercounted... there are just so many things that you can find where the folks who are living the problem every day don't have that default impulse to go build software for it. And yet it's a huge problem." 00:39:12

The Interface Revolution Is Still Ahead

Despite five years of LLMs, Dylan believes we're still in a primitive "prompting age" and that the interaction paradigms for AI are largely unexplored. He draws a direct line from HCI pioneers like Alan Kay to the design opportunity ahead.

"I'm still shocked that we're in a prompting age. Like it's been, what, five years since GPT-3 now?... There are other ways to explore latent space. I just think that the interfaces for how you interact with models will evolve so much." 00:21:49

"The last big one that people will reference if you just got to look for the consensus within is pull to refresh. Like we can go further." 00:45:54

Trust = Consistency + Creating More Value Than You Capture

Dylan cites Tim O'Reilly's foundational maxim and applies it to AI companies in the current moment, arguing that platforms that demonstrably create more ecosystem value than they extract will win the trust battle.

"Tim O'Reilly would always talk about create more value than you capture. And I think that is extremely important... The definition of a platform is that you're creating even more value outside of the platform than you are from the platform itself." 00:23:54

The Jailbreak Problem Is Understudied and Underdefined

Dylan flags AI jailbreaking as a critically important but poorly framed safety challenge, calling for more structured adversarial testing programs with better incentives — analogous to HackerOne for cybersecurity.

"Jailbreaking is something that is understudied. And I actually think that there needs to be a kind of different posture towards trying to get people to jailbreak models in order to battle test them the same way that we do with HackerOne. There are programs that are being run by the labs, of course, but I think that there needs to be way more incentive out there to do that." 00:29:27

Figma's Multi-Product Convergence Strategy

Dylan hints at Figma's roadmap direction: bringing design, code, and AI generation into one unified surface, eliminating the "where do I start" friction that currently fragments creative workflows.

"Both there's the impulse and the quest of how do you take these surfaces and bring them more together. When you are leaning more into AI and generation, there's a huge opportunity in doing that in one place and not having to think about where do I start." 00:40:34

2. Contrarian Perspectives

The "Software Is Dead" Narrative Is Wrong — and Creates Opportunity

The prevailing market narrative is that AI will kill traditional software companies. Dylan flatly disagrees, arguing many software businesses will do fine or even excel, and that the narrative snap creates a buying opportunity.

"Markets snap to narratives. Right now, the narrative seems to be like software is, you know, kaput... But I think that a lot of software will do just fine or actually even amazing." 00:38:30

Most People Underestimate the "Boring" Software Opportunity

Against the consensus that the most exciting AI opportunities are in flashy, consumer-facing products, Dylan argues the biggest untapped value is in unglamorous domains where domain experts haven't thought to build software.

"I would say that boring software is not to be undercounted... there are just so many things that you can find where the folks who are living the problem every day don't have that default impulse to go build software for it. And yet it's a huge problem." 00:39:12

Design Empathy Cannot Be Trained into Models Through Scale Alone

Contrary to the dominant belief that better models will solve everything, Dylan argues that true user empathy — the core of great design — requires capabilities that don't emerge from simply scaling intelligence.

"How do you teach a model true empathy for a user? How do you make it so that they can not just read a transcript of a user interview, but understand the relationships that user has, the inflections in their voice when they're talking about the problems they have... and actually have that empathy for their worldview and where they're coming from to build for them. I think it's really hard." 00:28:21

The Interaction Paradigm We're In Is a Dead End, Not the Destination

Most of the industry is doubling down on prompt-based interfaces as if they're the final form of human-AI interaction. Dylan believes this is an embarrassingly primitive moment we'll look back on.

"I'm still shocked that we're in a prompting age... There are other ways to explore latent space. I just think that the interfaces for how you interact with models will evolve so much. And to the degree that models are part of software, we will see a ton of exploration there." 00:21:49

Having More Agents Than Employees Is a Valid North Star

Against the common anxiety that AI agents displace workers in a damaging way, Dylan frames the ratio of agents-to-employees as a legitimate organizational goal that enables humans to operate on offense rather than defense.

"The new North Star is to have more agents than employees because they'll help equip your employees and you'll run faster. You'll be able to do more as an organization. You can be more on the offensive side." 00:25:52 (Host framing, endorsed by Dylan)

3. Companies Identified

Figma

Design and collaboration software platform. The central company of discussion — Dylan is co-founder and CEO. Highlighted for its evolution from design tool to full platform spanning design, code, and AI-assisted generation; strong earnings; 10,000-person Config conference; and its strategic product "Figma Weave" for steering AI outputs through human-intent workflows.

"We have something called Figma Weave, for example, where you can take model outputs and you can pass them through a workflow and really steer it along the way. And that can get you closer to, I think, non-deterministic generative workflows that get closer to intent and are systematic." 00:16:35

SpaceX

Aerospace and satellite company. Cited as an exemplar of missionary talent culture relentlessly pursuing challenge after challenge, with Dylan suggesting the market underestimates how many new categories SpaceX will enter.

"SpaceX is a great example of a company where you've got amazing talent who are, I think, really missionary. And that talent goes and quests and conquers after challenge, after challenge, after challenge... I believe it could be a very, very important company for the century or the next." 00:41:38

Anthropic

AI safety and model company. Referenced in the context of model trust, the "jailbreak" policy conversation with the U.S. government, and as a company whose models users have formed genuine affection for.

"You've got folks that are literally in the case of, you know, a 4.0 or Claude, like in love with the model... like they have affection for the model. I would say that's trust." 00:22:47

O'Reilly Media

Technology publisher. Dylan's first professional internship; he credits Tim O'Reilly's "create more value than you capture" maxim as foundational to his philosophy of platform building.

"My first internship professionally was at O'Reilly Media and I loved it there. Tim O'Reilly would always talk about create more value than you capture." [00:23.54]

Flipboard

News aggregation app. Dylan interned there before the Thiel Fellowship, noting the isolation of being a 20-year-old among colleagues in their 30s — which made the Thiel Fellowship community so valuable by contrast.

"I was doing an internship at Flipboard before the Thiel Fellowship. And at Flipboard, it was awesome. My co-workers were great. I hung out with them. But they're like in their 30s. And I was 20." 00:35:27

Brex

Intelligent finance platform (cards, expenses, banking, agentic finance). Sorcery's sponsor; named alongside OpenAI, Anthropic, Vercel, Granola, and Deepgram as exemplary companies using the platform.

"The companies building what's next from Vercel, OpenAI, Anthropic, Granola, and Deepgram all made the same call. They all run on Brex." 00:18:25

Figure AI

Humanoid robotics company. Briefly cited as an example of the agent/employee ratio trend — they have more humanoid robots at their office than employees.

"You could also see this with Figure AI. Like, they now have more humanoid robots at their office than employees." 00:25:52

Microsoft

Enterprise software and platform company. Referenced for their historical definition of platform success — measuring the GDP of the ecosystem built on Windows as greater than Windows revenue itself.

"What's the GDP of everything created on Windows back when Windows was the top priority of Microsoft? They wanted that sort of GDP of stuff built on Windows to be greater than the revenue of their Windows line." 00:24:45

Tesla

Electric vehicle company. Used as an illustration that truly distinctive products have brand identity baked into their form — logo becomes secondary to product identity.

"Can you take the Tesla logo off of a Tesla car? Sure. But like, you probably still know it's a Tesla." 00:42:52

4. People Identified

Tim O'Reilly

Founder, O'Reilly Media. Mentioned as Dylan's early mentor whose "create more value than you capture" philosophy became a foundational principle for how Dylan thinks about platform building.

"Tim O'Reilly would always talk about create more value than you capture. And I think that is extremely important." 00:23:54

Alan Kay

Computer scientist and HCI pioneer. Singled out by Dylan as a thinker whose work on fundamental human-computer interaction needs to be revisited as AI reshapes interface design.

"Alan Kay is awesome. And there's many other HCI and sort of design thinkers that I also think are incredible looking back. But also I think that we need to get back to the sort of work that was done back by, you know, sort of 60s through 80s timeframe in terms of just what are the fundamental ways to think about using a computer in the first place?" 00:45:25

Elad Gil

Investor and author. Provided a guest question for the interview; noted as a mentor to Dylan, implying a close advisory relationship.

"With great mentorship from Elad Gil." 00:37:46

Holly Herndon

Artist and musician. Cited as a visionary who was years ahead on AI voice open-sourcing and novel uses of blockchain/NFT-adjacent technology as early as 2020–2021.

"She was into very specific, quirky ways of using NFTs... she open sourced her voice with AI early on. You know, this is back in like 2021 or something. 2020. I mean, it's just like the timescale that she's operating on is different than the rest of us." 00:08:24

Grant Sanderson (3Blue1Brown)

Math educator and YouTuber. Praised for using design and graphics to achieve craft-forward, deeply empathetic explanations of complex mathematics — cited as a model for how design amplifies teaching.

"He does these explainer videos for math that are just incredible. And really uses graphics and design to be really craft forward in his explanations. And he puts so much care into them. And just has been a teacher for so many." 00:08:36

Mark Pincus

Founder, Zynga. Referenced as a product-focused thinker who pays close attention to the app release vs. usage gap data that Dylan shared, and was discussed in a separate Sorcery interview.

"I talked about that with Mark Pincus. He's a super product guy. He definitely pays attention to that." 00:19:43

Daniel Strachman

Director, Thiel Fellowship. Interviewed by Sorcery approximately a year prior; described Dylan as "hyperfluent" and cited him multiple times in her own interview. Described as "wonderful."

"I had an interview with Daniel Strachman from Thiel Fellowship, like, maybe a year ago now... she named you, I think, a couple of times in the interview, but you were an early Thiel Fellow. She said that you were hyperfluent." 00:31:59

Palmer (last name not stated)

Participant in the Mafia game show produced by Solana. Highlighted for exceptional composure and poker-faced gameplay — Dylan called him the actual star of the show.

"I think the star is actually Palmer... he kind of smirks when he's almost won. That was the first time I was going, oh, I would have figured it out. The other two I would have figured it out. Because they had tells. But Palmer, there was like nothing. He was just the same energy throughout." 00:13:31

5. Operating Insights

Memefy Your Vision Internally to Achieve Alignment on Novel Ideas

Dylan identifies a specific operational challenge unique to building genuinely novel products: you can't just communicate the vision — you have to compress it into something that can live in people's heads across a large team. He frames internal memefication as a real strategy for maintaining design and product coherence at scale.

"You kind of have to memefy it inside your company sometimes. But if you can do that and you can come up with more novel ways for people to interact with systems, that's where I think it'll really be amazing." 00:21:19

Your User Base's Feedback Quality Is a Strategic Asset to Design Around

Dylan points out that Figma has an unusual structural advantage: its users are designers, who are professionally trained to give structured, actionable feedback. He treats this as a compounding advantage and explicitly says he never takes it for granted. Operators should audit whether their user base has above-average feedback quality and invest accordingly to exploit it.

"We are very lucky as a design company to have designers as users and customers because they are trained to give good feedback. So the feedback we get is so much better than like your average feedback that a company gets. And I never take that for granted. And I always count as a blessing." 00:44:47

Run Evals on AI Outputs as a Core Product Workflow, Not an Afterthought

Dylan describes running evaluations on AI-generated design outputs as an ongoing, active internal practice at Figma — including evaluating competitor outputs and their own. He distinguishes between verifiable auto-eval and the inherently human-judged non-verifiable domains, suggesting operators need hybrid eval infrastructure.

"The evals we're running on stuff we're doing, stuff others are doing — a lot of it is inherently non-verifiable. It's got to be human judged at the end of the day. And you can set up more automatic methods for that. But you still have that human intent." 00:09:42

6. Overlooked Insights

"Figma Weave" Is a Quietly Significant Product That Could Define the Next Generation of AI-Assisted Creative Workflows

Dylan mentions "Figma Weave" almost in passing as an example of steering AI outputs — but what he's describing is a structured, human-intent-guided generative workflow layer sitting on top of models. This is architecturally distinct from raw prompting and addresses the core problem of the "permanent underclass of zero taste": that direct LLM output regresses to the mean. If Figma Weave succeeds, it positions Figma not just as a design tool but as the orchestration layer for AI-generated creative production, making it a platform that is nearly impossible to displace mid-workflow. No one in the conversation pauses to note the significance of this product name or its strategic implications.

"We have something called Figma Weave, for example, where you can take model outputs and you can pass them through a workflow and really steer it along the way. And that can get you closer to, I think, non-deterministic generative workflows that get closer to intent and are systematic." 00:16:35

The Emerging Investment Thesis in "Non-Obvious Vertical Software" May Be the Most Contrarian and Highest-Return Opportunity in Enterprise AI Right Now

Dylan throws out a single-sentence observation — that the best software opportunities are in domains where domain experts don't think to build software — and immediately moves on. But this is a precise, actionable investment thesis: find industries where pain is high, expertise is concentrated, and the incumbent practitioners have never had a "build software" instinct. These markets are both underserved and likely to have defensible data flywheels once a product gains traction. This is the opposite of where most AI venture dollars are flowing today (which cluster around developer tools, legal, and finance).

"I would say that boring software is not to be undercounted... there are just so many things that you can find where the folks who are living the problem every day don't have that default impulse to go build software for it. And yet it's a huge problem." 00:39:12