The labs are eating the app layer

Figma is trading at $18. In August 2025 it was $142. The latest 6% drop came after The Information reported that Anthropic is building a natural-language design tool for websites and presentations, paired with its upcoming Opus 4.7 model. No public demo. No beta signup. No pricing page. Figma's market cap fell by hundreds of millions on the rumour alone.

That reaction is the story. Not what Anthropic announced, but how instantly the market repriced every company sitting between a foundation model and the end user.

The conventional reading of the past 48 hours is that AI labs are "expanding into new verticals." I think that framing misses what's actually happening. Anthropic, OpenAI, and Microsoft aren't expanding. They're reclaiming territory the app layer was keeping warm.

TechCrunch reported that OpenAI acqui-hired Hiro Finance, a Ribbit Capital-backed personal finance startup whose product shuts down on 20 April. The team joins OpenAI to build financial reasoning directly into ChatGPT. The logic is blunt: if your model can already answer questions about money, budgets, and investments, why let a startup extract value from a capability you already have? Fold the expertise in-house, turn off the standalone product, move on.

The AI Insider reported that Microsoft is testing always-on autonomous agents in Copilot. These aren't the usual "ask me a question" assistants. They monitor your email, calendar, and workflows continuously, executing multi-step tasks scoped to your organisational role without being prompted. Every standalone productivity tool that handles scheduling, email triage, or task management just got a countdown timer.

The Ford moment

There's a useful parallel from a century ago. Ford Motor Company didn't stop at building cars. It bought rubber plantations in Brazil, iron mines in Michigan, and a railroad to connect them. When you control the core asset, owning the supply chain around it is cheaper than paying intermediaries. The AI labs have reached the same inflection. The model is the engine. Design tools, financial advisors, and autonomous workflow agents are the rubber plantations.

The uncomfortable part for SaaS founders: what many of them thought was a moat was always a timing advantage. While Anthropic, OpenAI, and Google were racing on benchmarks and parameter counts, the application layer had room to grow. Now that frontier models are converging and capability gaps are flattening, the labs are looking up from the engine room. They see revenue sitting in the app layer. They have the models, the distribution, and the capital to go get it.

The way I see it, the companies that survive this aren't the ones with the best UI or the cleverest prompt engineering. They're the ones with proprietary data, earned trust in regulated industries, or workflows so deeply embedded that ripping them out costs more than the AI alternative saves. If your product is fundamentally a better interface for something the model can already do, you're one product announcement away from Figma's chart.

The question every product builder should be asking today isn't "how do I use AI?" It's "what do I have that the labs can't replicate by Tuesday?"

---