Dylan Field on What Becomes Valuable When Software Gets Easier to Make

1. Guest Background

Dylan Field, Figma’s co-founder and CEO, approaches the AI debate from the position of someone building tools for professional designers. He dates the company’s beginning to August 2012 and describes its purpose as helping teams design, create and build software. His experience is relevant to the interview’s central question: what remains valuable in a design product when models can increasingly generate interfaces and code? His answer also deserves to be read with his position in view. He is explaining a market in which his own company competes.

Field says Figma is used for marketing, advertising, slides, whiteboards and diagrams as well as software design. Yet designers remain the reference point in product reviews. That combination—broad use, a specific professional center—helps explain his resistance to treating a general generation capability as equivalent to a finished professional tool.

His definition of design is equally important. He describes it as creativity meeting problem-solving. Designers must connect user needs with brand, marketing, culture and business constraints while moving between levels of abstraction. In this account, producing a screen is one visible outcome of a larger exercise in judgment. The interview examines how AI changes that exercise, alongside the organizational and competitive pressures surrounding it.

2. What the Episode Covers

The All-In conversation connects three questions that are often discussed separately: how a company maintains direction through uncertainty, how an application competes with model developers, and what humans contribute when digital production becomes easier. Figma’s experiences supply the connective tissue. The discussion is an account of Field’s reasoning and decisions, rather than an independent assessment of the company’s competitive position.

The proposed $20 billion Adobe acquisition provides the organizational case. Field recalls 16 months of regulatory review. Once a quick completion looked unlikely, he says, Figma kept building and shipping so that the company would be prepared for either outcome. That response is analytically useful without resolving the regulatory dispute: management could not control the external decision, but it could preserve the organization’s ability to move forward.

According to Field, employees greeted the deal’s conclusion primarily with relief from ambiguity. They had been carrying two possible futures in their heads. The next step was to make the independent company’s expectations explicit. Employees who were exhausted, or who had joined expecting an Adobe future, could leave with several months’ pay. They could reapply after six months and go through another interview. Field calls participation a small percentage but provides no precise figure. His account illustrates a way to clarify commitment; it does not establish a measured formula for successful retention.

The competitive discussion begins with a correction. Figma had an Anthropic partnership and an MCP connector, Field says, but Anthropic announced Claude Design separately. He expresses surprise about the communication surrounding that launch while saying its business impact on Figma was not significant. The interviewer then cites a quarterly report showing 48 percent year-over-year growth. Those statements have different sources and limited evidentiary reach. The exchange does not supply the full reporting basis or show that any particular AI strategy caused growth.

Field’s broader answer is that model improvement and product improvement are distinct. Professional software needs carefully developed workflows, mechanics and feel; he acknowledges that Figma still has work to do. He consequently rejects both blanket SaaS optimism and blanket SaaS collapse. Cursor is his example of an application that focused closely on users and anticipated their needs before having its own model. It supports his argument that application companies can create value, without proving that every specialized product has a durable advantage.

3. Core Views: Reasoning, Examples, and Limits

Field’s phrase “design is the new code” is best understood as a claim about where differentiation moves. He explicitly says Figma still needs engineers. His argument is that improving coding models make implementation more accessible, so the ability to produce software becomes a less complete explanation of why someone should choose it. Design, brand, a marketing perspective and emotional response become more important to the distinction. Within his definition of design, this reaches beyond appearance to the choices that give a product purpose and coherence.

He connects that competitive argument to a revision of his own thinking about creativity. At an earlier Config conference, he had predicted that AI would raise the creative ceiling. In this interview, he says AI has instead lowered barriers to production, while people raise the ceiling through focus, time and bold experimentation. The distinction separates access to making from exceptional achievement. It is a personal assessment, however, rather than a demonstrated ceiling on model capability. The conversation supplies no common test that could establish the maximum creative contribution of either humans or AI.

Field also makes room for ordinary usefulness. An average result may be sufficient, depending on the task. That qualification matters because a demand for distinctiveness can otherwise become a demand to overwork everything. An exploratory artifact and a product seeking sustained attention need not meet the same standard. His position asks the maker to decide when adequacy serves the purpose and when the purpose requires something more deliberate.

His market explanation rests partly on a claim that app supply is increasing rapidly while usage remains flat. The supplied exchange includes no dataset, source or measurement method for that comparison. It should therefore remain a stated premise in his argument, rather than a verified law of software economics. Given that premise, he sees a distinctive point of view as especially valuable. People often hesitate to express non-consensus positions, he observes, while attention is difficult to earn and retain. The useful distinction is between a sustained perspective and an isolated novelty; the interview does not establish that provocation by itself produces value.

A small creative example makes this argument more tangible. Field describes a web page that became slightly more degraded each time a visitor left and returned. He reads the behavior as a comment on attention and also praises the letter it contained. Here the interaction is part of the meaning. The work interests him because its mechanism expresses a thought, rather than merely demonstrating a technical effect. This is an example of his aesthetic judgment, not evidence of improved engagement or commercial performance.

His preferred working method keeps that judgment active. AI can provide something to react to and help a maker escape the blank canvas. The next move is to identify what is wrong and change it. When someone already has a clear vision, Field warns, delegating its implementation entirely to a model passes that vision through the model’s own distribution and fit bias. His proposed answer is rapid movement between generation and direct manipulation, with code treated as material on the canvas. Weave supplies a related example: model outputs connect to subsequent inputs in a node-based system, allowing the creator to shape generated material. These are descriptions of a product direction, not comparative evidence that a particular interface always produces better work.

Field’s observations about model launches add a further reason to retain control. Initial amazement, amplified by feeds showing strong examples, can give way to recognition of recurring tropes. He speculates that the absence of easily verifiable standards in design and creativity helps explain a different pace of progress from formal domains such as mathematics. He also allows that this could change. The explanation is a hypothesis, and his account of repetition should not be converted into a permanent technical limit.

Finally, the photography analogy complicates any simple forecast. The interviewer recalls layers and filters opening new creative possibilities. Field adds that the camera phone was a larger disruption to photographic behavior and declines to label the overall change simply better or worse. Applied to AI, the lesson is to examine access, habits and control together. A new generation feature alone cannot settle where professional practice will ultimately reside.

4. Learning and Application

For a product team, the most actionable starting point is to evaluate an entire job rather than its first generated artifact. Choose a real task and follow it through drafting, revision, detailed adjustment and delivery. Note where users lose control or need workarounds. This translates Field’s distinction between model and product into an assessment method: an impressive first result is useful evidence, but incomplete evidence of professional suitability. Existing tools should face the same test. Years of development do not automatically make a product the best fit for the task now being performed.

For an individual maker, write down the intended audience, central idea and a few essential experience requirements before generating. Afterward, examine where the result has replaced those requirements with familiar model choices. Direct editing may be preferable when a local change needs precise control; further generation may help when the overall direction remains uncertain. A node-based approach, as illustrated by Field’s description of Weave, suggests preserving intermediate outputs and connections so that a creator can identify which stage to change. The benefit is greater visibility into the making process; the tradeoff is additional effort managing that process.

Decide how much refinement the purpose warrants. A draft meant to start a discussion may only need an adequate result. A product whose appeal depends on a recognizable perspective needs time reserved for that perspective, its interactions and its consistency. Field’s degrading web page offers a useful question: does the mechanism express something, or does it merely attract momentary curiosity? The example cannot substitute for user feedback, and it is no reason to add experimental interactions to every product.

Internal AI writing presents a different responsibility. Field says Figma asks people to label AI-assisted writing “written with AI,” against a background of concerns about decks whose authors may not know their contents. A team could adopt a disclosure rule and separately require the submitting author to explain the argument, check numbers and identify the decision the document supports. Those additional steps are practical recommendations derived from the case. The interview does not establish that labeling itself improves quality, and a label should not be treated as evidence that review occurred.

The acquisition story also supports a bounded management practice: preserve delivery while an external outcome remains unresolved, then explain the demands of the chosen direction and make room for differing preferences. Figma’s severance and reapplication terms were specific choices, not a package every organization can afford or should copy. A low reported departure rate would not, by itself, establish a healthy employee experience.

For personal learning, Field’s distinction is sharper than a simple productivity metric. Ask whether AI helped you understand something, make a judgment and attempt an activity you previously avoided. After completing a task, try explaining the important decisions without relying on the generated prose, identify what remains unknown and choose the next action. This retains cognitive work, which is part of the point. The closing discussion of an alleged OpenAI mathematics release involving more than 700 papers provides no verification and should not anchor career predictions. Field’s mixed excitement and sadness, alongside his acknowledgment of unknown challenges, supports a more provisional practice: expand what you can do while continuing to revise what you believe.

Source

• Original episode: Dylan Field: Is AI Replacing Designers?, Design is the New Code & Democratizing Creativity