Claude Science: Anthropic Just Gave Researchers a Claude Code Moment

Anthropic shipped something at the end of June that I think people are underrating: Claude Science, an AI workbench for researchers. And the most important sentence in the whole launch is the one that sounds least exciting: it is not a new model.

Claude Science runs on the same Claude models everyone already has — including Opus 4.8 — but wraps them in a purpose-built harness: connections to 60+ scientific databases, prebuilt toolkits for genomics, proteomics, structural biology, and cheminformatics, and the ability to autonomously carry out real computational work from a high-level instruction. Sound familiar? It's the Claude Code playbook, pointed at science instead of software.

The pattern worth internalizing

This is now the third or fourth proof that the frontier isn't just bigger models — it's better harnesses. The same intelligence, given the right tools, context, and permission structure, becomes 10x more useful in a domain. Software got Claude Code. Science just got its version. Every serious vertical — law, medicine, finance, chip design — is going to get one.

If you're building products (like I do at nights), that's the lesson to steal: your moat isn't access to the model, because everyone has that. It's the scaffolding you build around it — the tools you connect, the workflows you encode, the domain data you wire in.

The detail I love

Anthropic is funding up to 50 research projects with grants up to $30,000 each — applications open until July 15, projects running September through December. Handing scientists both the instrument and the funding to prove it out is exactly how you bootstrap an ecosystem.

Agent harnesses are the new APIs. Learn to build them.