Key Takeaways

  • When Anthropic launched Claude Design, critics declared Figma obsolete, yet Figma reported 48% year-over-year revenue growth soon after.
  • Dylan Field argues observers confuse model progress with product progress, overlooking the years required to dial in mechanical precision for daily user workflows.
  • Jason Calacanis compares professional creative software to a musical instrument, where micro-feel and mechanics matter far more than raw generative capability.
  • Cursor captured developers by obsessing over specific text editing interactions long before foundation model labs attempted to clone them.

Claude Design Did Not Slow Figma Down

When regulators killed Adobe's $20 billion bid to acquire Figma, industry skeptics warned that standalone design software was living on borrowed time. Anthropic dropped Claude Design shortly after. The tech internet immediately predicted Figma's demise. Prompts would replace vectors. Canvas tools would vanish overnight.

Then Figma posted 48% year-over-year revenue growth.

The predicted SaaS apocalypse did not arrive. As Field points out, a working demo is not the same thing as a production tool. Field observed: “I think it's easy to mistake model progress for product progress, right? The reality is a tool like Figma has taken a very long time to build and a very long time to make sure that we get the workflows right for our users.”

Generating an isolated screen from an English prompt looks magical in a twenty-second video clip. Managing three hundred screens across twelve designers, four product managers, and twenty engineers across five operating systems is an entirely different discipline. Generative engines output assets. Product teams need shared, durable environments.

Models Are Not Products

Every foundation lab faces an identical identity crisis. They run multi-billion-dollar compute clusters, yet their consumer products often resemble thin text boxes.

Field framed the core problem facing every AI company today: “The hard part about model companies and product is like what is the product? Is it the model? Is it the UI on top? Is it some hybrid of both? If it's a hybrid, how is it weighted?”

Calacanis pinpointed why generalized model labs struggle to wipe out specialized software: “The feel of it is like an instrument. You have to get the feel right. You have to get the mechanics right.” A musician does not switch guitars simply because a synthesizer can generate a guitar riff faster. Professional users live inside their software for six to ten hours every day. Latency, undo states, keyboard shortcuts, and tactile feedback decide whether someone stays or leaves.

Cursor Beat the Labs on Workflows

Look at Cursor. The team entered an arena populated by the best funded tech giants in human history. OpenAI, Microsoft, and Google all owned the underlying models, yet Cursor dominated developer mindshare.

Field noted: “Cursor by all means should not have grown the way they did. You've got all these frontier labs that are chasing them with their own models before Cursor had a model of any kind. But yet, Cursor really obsessed over the user.”

Cursor did not win because it had a proprietary trillion-parameter model. It won because it solved tab-completion ergonomics, multi-file context, and diff reviews inside the code editor. It treated the interaction model as the primary product.

Software defensibility is not binary. As Field put it, “There are more modes for software than are implied by the SaaS apocalypse. And it's not like software as a category is binary, good or bad.” If raw intelligence is cheap and widely distributed, the advantage shifts directly to taste, distribution, and workflow mechanics.

What to Do With This

Stop trying to build a foundation model wrapper that does five generalized tasks poorly. Audit the single tool your users spend the most manual hours operating this week. Map every click, shortcut, and handoff in that workflow, then design your product so that completing the job feels like playing a finely tuned instrument.