Key Takeaways

  • Dylan Field reversed his past public claim from Config, Figma's annual conference, admitting generative AI does not raise the creative ceiling.
  • Foundation models are bound by their training distributions, meaning automated generation naturally trends toward the statistical mean.
  • As software generation commoditizes code and digital production, company differentiation shifts entirely to brand, taste, and craft.
  • Figma enforces an internal workplace rule requiring employees to label any document with "written with AI" if machine models were used.

The Config Reversal

Standing on stage at Figma's annual Config conference a few years ago, Field made a bold promise to thousands of software designers. He told the audience that generative models would raise the creative ceiling for everyone, unlocking work nobody could previously achieve.

Sitting down with Jason Calacanis in Napa following the collapse of Adobe's blocked $20 billion acquisition of Figma, Field took that statement back.

“I used to believe even a few years ago, I said it our annual user conference Config. I said the ceiling will be raised by AI. You'll be able to do greater things than before. That was wrong,” Field said. “The ceiling has stayed the same. AI is not raising the ceiling. Humans are raising the ceiling though.”

Instead of pulling the best creators higher, machine intelligence democratizes basic execution. “It's the case, I think, that we're lowering the floor with AI, making it so that all of the production, whether it's code or it's motion, video, design, whatever you name it, if it's digital, you're lowering the floor to make it happen.”

The Trap of the Statistical Mean

Large models predict outputs based on historical data. By definition, a system trained on existing internet content spits out the median of its training distribution. When you rely solely on text prompts to generate an interface, a marketing campaign, or a codebase, you receive the mathematical average of human output.

Calacanis pointed out the failure mode of treating that average as real work: “If you're not doing something that's never been done before, you're not creating something. You're mimicking something. You're copying something. You're xeroxing something. But to do something new is to create. And newness requires doing things that have never been done before.”

Raising the bar requires human beings who are willing to ignore the median. “The creatives, the designers out there, when they choose to focus, invest their time, and to be bold and try to do something that no one else has ever done before, that's when we see the ceiling raised,” Field explained.

When basic coding and interface mockups take zero effort to produce, software features become commodities. The advantage goes to teams with high personal taste, strong brand identities, and an appetite for aesthetic risk.

The Transparency Rule Inside Figma

To prevent unthinking reliance on automated text, Field introduced a blunt policy for Figma's internal communications. Anyone who relies on a foundation model for written work must state it openly.

“At Figma, we actually made it so that it was just a simple thing, but we said, 'Hey, if you are using AI in your writing of any kind, just put at the top written with AI.' Very simple, very easy to follow.”

This single line changes team behavior. It strips away the illusion of deep effort when someone drafts a strategy memo with automated software, protecting organizational trust. If an employee claims to have spent hours refining a thesis, the disclosure forces them to prove where their original judgment actually altered the draft.

What to Do With This

Institute Figma's disclosure rule across your team tomorrow. Require every internal product brief, customer memo, and strategy document generated with automated tools to state "Written with AI" on line one. If the author cannot point to three specific decisions in the document that bucked the model's suggestions, reject the memo and make them rewrite it by hand.