Key Takeaways

  • Pitch decks and memos drafted entirely by language models often backfire by signaling that the author avoided the hard work of clear reasoning.
  • Figma launched an updated design agent focused on speed, file context, and canvas comprehension rather than replacing human designers.
  • Teaching foundation models to operate inside a specific design environment requires training them on canvas mechanics first before expecting quality output.
  • Figma enforces Dylan Field's 'Written with AI' Transparency Rule across internal communications to protect trust between teams.

Dylan Field's 'Written with AI' Transparency Rule

When employees outsource their writing to language models, colleagues end up reading generic text that nobody actually thought through. To stop this from degrading team communication, Field established a clear internal standard at Figma:

  • Requirement: Add 'with AI' at the top of any internal message or document containing AI-generated or AI-assisted writing.
  • Threshold: Applies whether the entire document was drafted by AI or only specific paragraphs, while basic spellchecking and proofreading do not require mandatory labeling.
  • Core Objective: Prevent degradation of trust across teams by ensuring readers know the author actually evaluated, understood, and thought through the ideas they are asking others to read.

As Field explained, “Because of that, I came up with this sort of rule for Figma, which is let's just have it be that you put in with AI at the top of messages, where there's AI writing of any kind.”

The purpose is accountability. “I think that in general, people should not give up on their own thinking,” Field noted. “They cannot give that to the model. And I see so many people that are essentially looking for the shortcut.”

Building Figma's own agent required a similar philosophy. Instead of trying to automate creative taste, the product team focused on canvas context. “Today's launch you'll see that the Figma agent's faster, higher quality. It has better ability to deal with context across your files and more overall agent capability,” Field said. “You have to teach the models Figma first and make it so that they can understand how to go and work with Figma as well as do good design work.”

When This Works (and When It Doesn't)

This rule works best in high-trust engineering, product, and design organizations where written memos drive major roadmap decisions. When a team lead reads a technical proposal or a product spec, they need to know the author wrestled with the trade-offs rather than letting a prompt generator paste smooth paragraphs.

It breaks down in high-volume, external-facing administrative tasks. Customer support macros, automated release notes, and basic status digests do not require a disclaimer because nobody assumes they represent deep personal synthesis. Applying the label there creates clutter without protecting creative rigor.

What to Do With This

Pick the most important product spec or strategy memo your team is reviewing this week. If you used an LLM to generate the outline or draft paragraphs, add a clear tag at the very top: Drafted with AI assistance; arguments and metrics verified by [Your Name].

Then test your own knowledge before sending it. Close the document and explain its three core trade-offs out loud in sixty seconds. If you cannot explain the logic without looking at the generated text, delete the draft and write the core arguments by hand.