Key Takeaways

  • After regulators blocked Adobe's $20 billion acquisition, Figma stayed independent and focused on building node-based creative interfaces rather than single-prompt text boxes.
  • When Adobe introduced Photoshop layers and Kai's Power Tools, photographers argued the craft was dead; instead, those tools created an entirely new generation of digital artists.
  • Figma developed Weave, a node-based tool that links model outputs to other inputs so designers can direct and reshape generations like raw clay.
  • Every major model release triggers what Field calls "AI psychosis," a brief period where viral feeds boost the best generations before users get tired of the model's obvious tropes.
  • Model capabilities stack rapidly in formal domains like math and cybersecurity, while progress in open-ended creative domains remains slow and incremental.

Photoshop Layers Caused the Same Panic Thirty Years Ago

Every time software automates visual production, existing practitioners assume their craft has reached its expiration date. Calacanis pointed out that photographers had this exact reaction when Adobe added layers and filters to early digital workflows. In the 1990s, when Kai Krause released Kai's Power Tools, the design community claimed algorithmic gradients and texture generators would destroy creative intent. Field brought up another forgotten relic from that transition: “The resistance to change is real and there's a great thing you can Google actually for your point. Look up the mailing list for when layers was introduced and it's like people going this is the worst thing ever.”

Layers did not eliminate the artist. They gave artists more granular control over complex compositions. Instead of baking every pixel permanently into a single flat file, designers could isolate components, adjust opacities, and test variations without starting over. The current panic around generative AI treats image synthesis like an end state. Field sees it as an early layer that still lacks real tactile control.

Figma Built Weave to Treat Model Outputs Like Clay

If you drop a generative model into an empty prompt box, you produce static assets that the user cannot edit without rewriting the prompt. That is an awkward match for professional designers who need precise steering. Figma took a different approach by treating AI outputs as raw material inside an active canvas, using shaders and a node-based workflow called Weave.

“We have a product called Weave,” Field said. “And Weave is basically a node-based system. You can connect model outputs into other inputs and such. And the whole idea is it's not bad to generate with AI, but you need a way to craft that clay. You need a way to mold it.”

In a node setup, an output from one diffusion step or text generation acts as the parameter for the next step. A designer can adjust color data, alter vector coordinates, or pipe an image through custom shaders without running another black-box generation. You move away from prompt roulette and get back to mechanical manipulation.

Why AI Psychosis Fades When Model Tropes Appear

Builders often mistake early social media hype for lasting user adoption. Field noted a recurring dynamic with every new model release: “When a new model drops there's this AI psychosis moment that people are in for a few days. I think also aided by algorithmic feeds, the best results rise to the top. Everyone says this model is everything. And then a few days later it's okay, I know the main tropes of the model.”

That plateau happens because creative work does not scale the same way formal logic scales. Foundation models improve rapidly on benchmarks that have clear true-or-false answers. They solve complex math proofs, spot vulnerabilities in codebases, and crush standardized tests. Creative taste has no objective test score. “Models keep getting so good when it comes to these formal domains, math, cybersecurity,” Field explained. “The IQ points just stack and stack and stack, and then for creativity and for design it's very incremental progress overall.”

Because creative models make only small improvements from version to version, raw generation loses its novelty quickly. The lasting software companies in this cycle will not be the teams wrapping another raw text prompt. They will be the teams that build tools to refine, branch, and inspect the output.

What to Do With This

Open your product roadmap tomorrow and audit your AI features. If you are shipping a text box where a user types a prompt and receives an uneditable final asset, replace it with a chained step. Give the user at least two discrete control points where they can inspect, tweak, or pipe intermediate parameters into another action before the final output renders.