Anthropic buys into biotech AI

Anthropic is getting serious about biology

Anthropic just spent $400 million on an eight-month-old startup.

The company is acquiring Coefficient Bio, which builds AI models for medical research tasks like small molecule discovery and biomarker analysis.

This looks like a speed move, not a safe one. Buying a young team this early suggests Anthropic thinks the real value is in talent and direction, not proven product, especially in areas where AI could cut research timelines by up to 100x.

It also shows where competition is going. Labs are no longer just chasing better chat models; they are moving into fields where AI can produce economic value faster and more defensibly than consumer tools.

Somewhere, a lab bench is still covered in notebooks and pipettes while this gets wired into software.

If this works, do AI labs start looking more like pharma companies?

Coding tools are becoming operators

Cursor just changed what its product does.

Cursor 3 lets developers spin up AI agents that complete coding tasks on their own, going head-to-head with tools like Codex and Claude Code that already have millions of users.

This is a shift from assistive coding to delegated work. The product is no longer helping you write code line by line; it is taking a task and running with it, which changes how developers measure trust and control.

It also tightens the competition. Cursor used to sit on top of models, but now it is competing directly with the labs that make them, which is a harder position to hold over time.

You can imagine a developer watching logs scroll by while the agent keeps working.

If the agent does most of the work, what does “coding” mean day to day?

A quiet vendor just became a big risk

A little-known data vendor just forced Meta to stop work.

Meta paused its projects with Mercor after a breach tied to a compromised AI tool called LiteLLM, now linked to a wider supply chain attack affecting thousands of organizations.

Mercor sits in a sensitive spot, hiring large contractor networks to build proprietary datasets for labs like OpenAI and Anthropic, and those datasets are treated like core IP. The fact that one vendor can interrupt multiple top labs at once tells you how concentrated and fragile this layer has become.

This will push labs to audit vendors harder and possibly pull more data work in-house, even if it slows them down in the short term. It also raises the chance that competitors learn how models are trained if any of that 200 GB to 3 TB of claimed stolen data is real.

You can picture someone refreshing Slack at midnight, waiting to see if their project still exists.

If this layer breaks, who actually controls AI progress?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.