Anthropic Launches Claude Science, AI Market Splits Into Specialist Tiers

Why this matters: Claude Science targets pharma, biotech, materials science, and academic research—domains where hallucination is catastrophic and domain knowledge is currency. These are high-margin, high-stakes users willing to pay premium prices for reliability. By specializing, Anthropic is essentially saying that general-purpose LLMs are becoming table stakes, not differentiation. The real money is in verticalized products where you've baked in enough domain validation and accuracy guarantees that enterprises feel comfortable making research decisions based on your outputs.

This is a significant competitive signal. OpenAI has been pushing toward GPT-as-platform (letting others build the specialized layers). Google is investing in vertical-specific models like TabFM for tabular data. Meta is betting on neurotechnology applications. The AI infrastructure market isn't consolidating—it's fragmenting into increasingly specialized tiers, and founders need to decide: are you building general-purpose tools, or are you going deep into a domain?

The strategy tracks with what we're seeing elsewhere in today's news. Google's TabFM demonstrates that foundation models for specific data types (structured tables) unlock new possibilities for enterprises sitting on years of operational data. Meta's brain-to-text work opens entirely new interfaces and accessibility use cases. These aren't minor tweaks—they're different markets with different buying processes, different compliance requirements, and different willingness to pay.

The employment data cited in the Ramp analysis is also relevant here: firms adopting generative AI are shifting hiring patterns, but not uniformly. Domain expertise becomes more valuable when your AI system needs someone who understands both the domain *and* the model's limitations. This creates a second-order effect: as AI companies verticalize, they need deeper partnerships with domain experts, which means opportunity for specialized consulting and integration layers.

One note of caution: Godot's decision to ban AI-authored code contributions is a small signal with bigger implications. Open-source maintainers are explicitly losing trust in AI-generated code quality and verifiability. This isn't about ideology—it's pragmatism. If you're building products that depend on open-source dependencies, you should be thinking about code provenance and validation. The assumption that AI-generated code is equivalent to human-written code is increasingly unsafe.

The throughline: AI is moving from "build one general model, let a thousand applications bloom" to "build deep expertise in specific domains, charge accordingly." If you're founding around AI right now, the question isn't whether to use Claude or GPT. It's whether you have a specific domain where you can build defensible advantages through accuracy, integration depth, and trust. Generic wrappers around LLMs are becoming commoditized. Specialists win.