Dylan Field states that if AI is used in the writing process for Figma, it must be prefixed with 'written with AI'
Dylan Field, CEO of Figma, stated that the company's policy is: if AI is used in the writing process, it must be prefixed with 'written with AI'.
This statement applies to both internal and external document processes and is not a new product switch. Figma's acceptable use policy prohibits deception or misleading others, including implying that AI-generated content is purely human-generated, and prohibits altering or obscuring metadata, digital signatures, or watermarks that identify generated content. The help center requires users to inform others when using or displaying generated results, and some features, such as sticky note summaries, will be marked by the product itself. The AI terms specify that outputs are unverified, do not represent the company's views, and customers must review and obtain necessary authorizations.
The company is fully expanding generative capabilities in its design tools: Figma Make, agents, MCP, and Weave are included in seat expansion. Paid customers with annual recurring revenue exceeding $10,000 increased by approximately 37% year-on-year, while those exceeding $100,000 increased by about 48%. Starting March 18, 2026, all seats will have an AI credit limit, and customers will begin purchasing additional credits. Field mentioned in the earnings report that code is increasingly commoditized, and design has become a differentiator; the Google team used Figma for the end-to-end source files of the Gemini enterprise agent experience. In 2024, Make Designs was taken offline due to its generative interface resembling the Apple Weather app, and he admitted that changes were too few and paused the feature. Content training is enabled by default for Starter and Professional, and can be turned off by team administrators, while Organization and Enterprise are turned off by default.
The labeling policy changes 'used model' from an optional statement to a mandatory format at the beginning of the document. It covers the writing process, not just the final design draft. In hiring, he publicly favors younger candidates, whom he calls AI natives, and stated that if someone resists this direction, they are only suitable for casual dinner conversations. The product narrative aims to lower design barriers and raise professional ceilings while delegating tedious steps to the model; the prefix label turns the same set of capabilities into a must-see craft description.
In market mechanisms, this is a trust label, not a new subscription tier. Buyers are corporate design teams and collaborators reading Figma documents; sellers reduce the premium controversy of 'human-written' through transparency. Funding still follows seats and credits. Beneficiaries are teams that have integrated generation into their workflows and can comply with a single line of text; those under pressure are authors who treat the model as an invisible ghostwriter while maintaining a purely handcrafted persona externally. Once the label becomes an industry habit, unmarked manuscripts will be defaulted as undisclosed.
Supplementary data: Field co-founded Figma in 2012, and after going public in 2025, continues to explain growth through AI. The acceptable use policy was recently updated on August 19, 2026. The help center states that generated results may be misleading or incorrect and cannot replace expert judgment.
Source: Public information
ABAB AI Insight
Field's disclosure moves from product watermarking to the writing header, acknowledging that models have entered documents, not just the canvas. Prohibiting the pretense of pure human effort is to prevent brands and legal issues from arising together. The weather app plagiarism controversy proves that generative outputs can carry memories of others' interfaces; a line at the beginning cannot erase those memories but can shift responsibility from 'we didn't know' to 'we told you'. The credit limit indicates that generation has become a measurable consumption, and labeling serves as the instruction manual for that consumption. A preference for younger employees and mandatory disclosure are two sides of the same coin: the company wants people who can use models but does not allow hiding models in signatures.
The capital path is seat fees plus credits. Make and agents allow non-designers to access files, while design systems remain the core charge. The disclosure policy reduces the procurement committee's cost of questioning 'who wrote the specifications' and provides a traceable audit trail for legal departments. After the Adobe deal fell through, Figma must prove that AI will not destroy the premium of professional design; the label shifts the premium from mysterious craftsmanship to scrutiny and taste.
Comparable examples include AI usage statements in academic papers and mandatory disclosures for generated images in newsrooms. The difference is that Figma sells generation itself while requiring documents to acknowledge generation, effectively acting as both the pharmaceutical company and the instruction manual. The industry phase is a shift from collaborative whiteboards to generative assembly lines, where the visibility of the craft determines whether prices can be maintained. Unmarked outputs will become a new reputational risk, just like unwatermarked images.
Structural judgments belong to industry self-regulation before regulatory changes. The mechanism is: when the authenticity of outputs is visually indistinguishable, the market can only label the process. Whoever first writes the label into policy defines what constitutes honest generative work. A line at the header is very cheap; the cost of not writing it is a devaluation of the entire document's credibility.
ABAB News · Cognitive Law
When models enter documents, signatures must change format
Unlabeled handcrafted work will be treated as undisclosed ghostwriting
The cheaper the generation, the more honesty must be written in the first line