How Figma was built: design, AI, and the long path to product-market fit

Executive overview

Design is becoming more critical as software gets easier to build — not less. AI lowers the floor for who can participate in design, but hasn't closed the gap between fast prototyping and finished, well-designed product.

Dylan Field built Figma through years of failed pivots before locking onto a thesis: more software means more designers. The multiplayer bet was initially rejected by the design community — then became Figma's defining feature.

The designer's role is not at risk from AI; it is becoming the most important role in building software.

AI and the future of design

• AI is a tool category, not a replacement — designers and non-designers alike use it to explore more, not to exit the process.
• The idea maze now has more breadth: AI lets you try more paths, but depth still requires human judgment.
• Prompt-to-app exists; well-designed prompt-to-app does not yet.
• "Vibe coding" and "locking in" describe the flow state that fast feedback loops produce — Figma's goal is to extend that feeling to more people.
• The gap is not just in design: getting from prototype to extensible, production-ready code remains unsolved.
• Text-only prompting is the telnet era of AI; interfaces will diversify significantly from here.

Why models don't yet understand design

• Design = art applied to problem solving; diffusion models handle art, LLMs handle problem solving — the marriage hasn't happened.
• Designers bring context far beyond a two-line brief: user research, brand, cultural moment, emotional state.
• Models lack the empathy and judgment piece; this may change, but hasn't yet.
• Engineers and mathematicians often lack product judgment — it may be a distinct muscle that doesn't transfer through math-heavy training data.
• Developers may increasingly call themselves designers; that doesn't mean they stop writing code.

How Figma started: the idea maze in practice

• Dylan Field and Evan (co-founder) started in late 2011 by asking "why now?" — landed on drones and WebGL; drones vetoed quickly.
• Explored gaming vs. tools — chose tools; explored photo editing — abandoned it as too crowded and phone-native.
• Spent a week building a meme generator before recognising it wasn't the answer; the text rendering engine built for it became Figma V1's text layer.
• Saw Fireworks get killed by Adobe; formed the thesis that growing software complexity would grow designer headcount.
• The seed pitch in June 2013 was "all over the place" — the thesis clarified over time, not upfront.

Getting the first customers

• Figma launched closed beta in December 2015 with no multiplayer — browser-based, high quality, minimal features.
• First real customers were Notion and Coda (then called Krypton); Coda's CEO didn't know he was Figma's first customer until years later.
• Fonts broke 20 minutes after the first customer meeting; Dylan and Evan turned the car around, went back, and fixed it on-site without saying it was a crisis.
• Early adopters were minimalists who liked the small feature set — not a broad market signal, but enough to keep going.
• Multiplayer launched at GA in October 2016; initial reaction from designers was hostility ("a camel is a horse designed by committee").
• A user threw a public "design party" in a shared Figma file, intending to mock it — the servers nearly broke, but it demonstrated the product's core value better than any marketing could.

Scaling from zero to one and beyond

• The self-awareness loop: identify what you're spending most time on, then replace yourself in that role if you have resources.
• Danger zone: becoming so reactive that the organisation never self-improves.
• Field's regret: not hiring faster once product-market pull was clear — a 12-page feedback doc from an early user was a signal he underweighted.
• FigJam and Figma Slides both came from user behaviour Figma observed rather than top-down planning; users pulled the products out.
• Maker Week (internal hack week open to the whole company) has produced several of Figma's best features, including Slides.

Advice for founders

• Move as fast as possible; current AI tools make speed more achievable than ever.
• When a team proposes a nine-month roadmap, ask what can be de-scoped to ship and learn faster.
• Two-year build timelines are rarely necessary; the exceptions are genuine hard-tech companies.
• Don't use VC market-size math to decide what to work on — look at where users and the market are going.
• Early signal of fit can appear before the product is ready; don't wait for a perfect product to read the signs.
• Dog-fooding matters: if the builders don't use the product daily, quality degrades invisibly.