Claude Science: the moat is the workflow, not the model
Anthropic's Claude Science is 60+ skills and connectors on top of existing Claude models — proof that the value in AI is the workflow layer, not the model.
Anthropic just shipped a flagship product that is, pointedly, not a new model. Claude Science, launched June 30, is an AI workbench for researchers — 60+ curated skills and connectors wired into the tools scientists actually use. It runs on the same Claude Opus and Sonnet models you can already call. The interesting part for anyone running a business isn't the biology. It's the admission underneath: the value here is the workflow layer, not the raw model.
What actually happened
Claude Science is a workbench where a coordinating agent has access to over 60 preconfigured skills and connectors spanning genomics, proteomics, structural biology, and cheminformatics. It plugs directly into the databases and models researchers live in — UniProt, PDB, Ensembl, ChEMBL, and NVIDIA's BioNeMo toolkit (Evo 2, Boltz-2, OpenFold3) — and produces auditable artifacts a scientist can actually cite.
TechCrunch put the strategy plainly: Claude Science "bets on workflow, not a new model," to win over scientists. Anthropic itself is explicit that this isn't a more capable model for biology — it's the same Claude, wrapped in the connectors and skills that turn a general chatbot into something a domain expert will keep open all day. It's in beta for Pro, Max, Team, and Enterprise, with discounted academic seats and a grant program offering up to $30,000 in credits to 50 projects.
Why it matters for your business
Here's the operator translation. Everyone is fixated on which model is smartest this month. Anthropic — the company selling the models — just demonstrated that the smartest model, undifferentiated, isn't the product. The product is the 60 connectors, the skills that encode how the work is actually done, and the auditable output. That layer is where the defensibility lives, and it's the layer you can own.
You are not going to out-train Anthropic. You don't need to. Your edge is the same as Claude Science's: the connectors into your specific stack, the skills that capture how your business actually operates, and outputs your team trusts. Swap the model underneath next year and the workflow keeps working. That's the whole game — and it's a build, not a subscription.
Key takeaways
- Anthropic launched Claude Science on June 30 — a workbench of 60+ skills and connectors, not a new model
- It runs on existing Claude models; the differentiation is the workflow layer, not raw capability
- TechCrunch summarized the strategy as betting on "workflow, not a new model"
- Your defensibility works the same way: own the connectors, skills, and trusted outputs — keep the model swappable
Want the workflow layer that's actually yours? We build the connectors and skills around your business — the part that keeps its value when the model underneath changes. See how we architect it or tell us what you'd wrap an agent around.
Sources: Anthropic, TechCrunch.
- #claude-science
- #ai-agents
- #workflow
- #automation
- #anthropic
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Probook's $40M and the AI OS for home services
a16z and Sequoia put $40M into Probook, an AI operating system for HVAC, plumbing, and electrical shops. What vertical AI software means for operators.
Read itMeta's AI agents are behind schedule. Yours will be too.
Zuckerberg told staff Meta's AI agent development hasn't accelerated in four months — on a $145B budget. The operator lesson for small businesses betting on agents.
Read it