The AI labs are coming for your creative tools

Mike Krieger resigned from Figma's board on a Tuesday. By Friday, his employer had launched the product that sent Figma's stock down 7%.

Anthropic's Claude Design generates prototypes, slide decks, and marketing collateral from text prompts. It reads your codebase and design files, auto-applies brand guidelines, and exports to Canva, PDF, and PPTX. The target user isn't a designer. It's the founder or product manager who used to ask a designer. Figma, Adobe, and Wix all dropped on the news, and the sell-off wasn't irrational.

The conventional reading is that AI is coming for creative work. The more precise reading is that the AI labs are coming for creative tools. When Anthropic ships a design product, it isn't replacing the designer the way image generation replaces stock photography. It's replacing the application the designer used. The model provider looked at the application layer and decided to build it.

Adobe saw this coming. Its Firefly AI Assistant, announced at Summit this week, is an agentic creative tool that orchestrates multi-step workflows across Photoshop, Premiere, and Illustrator from natural language. Adobe confirmed it's building Firefly integrations for Claude and other third-party models. The strategy: if the labs want to be the interface, Adobe wants to be the infrastructure underneath. Make the AI native to the creative suite before someone else makes the creative suite native to the AI.

OpenAI made the same move from the developer tools side. Codex now has background computer use on Mac, persistent memory, scheduled automations, and over 90 plugins spanning CircleCI, GitLab, and Microsoft Suite. Codex can see, click, and type across any Mac application with its own cursor. If that sounds like it competes with every desktop automation tool and workflow product on the market, that's because it does.

The vertical integration play

In economics, this has a name: vertical integration. A company expands into adjacent layers of its value chain rather than relying on partners. Netflix did it when it stopped licensing shows and started making them. Apple did it when it stopped buying Intel chips and designed its own. The logic is always the same: once you control the core asset, the margin in the next layer up looks too good to leave to someone else.

For the AI labs, the core asset is the model. The next layer up is the application. The applications closest to the model are the most natural targets: design tools and coding environments. They're the ones where the model's capabilities translate most directly into product features, and where the existing incumbents' moats depend on workflows the model can automate.

I think the most uncomfortable implication is for companies that built their businesses as model-powered tools. If you're a startup that raised capital to put an AI layer on top of design, coding, or content creation, your model provider just became your competitor. The same API you're paying for is now powering a first-party product with better integration, lower latency, and zero margin to protect.

Adobe's counter is the shrewdest move available: make itself indispensable infrastructure the labs build on top of. But that requires something most tool companies don't have — a suite so deep and so widely adopted that even the model providers need it. For everyone else, the question is blunt: if the lab can ship 80% of your product as a feature, what's your remaining 20% worth?

---