Anthropic

Scaling enterprise AI: Fireside chat with Eli Lilly’s Diogo Rau and Dario Amodei

Dario Amodei, CEO and co-founder at Anthropic, sits down with Diogo Rau, Chief Information and Digital Officer at Eli Lilly and Company, to discuss building enterprise AI for regulated industries like life sciences. In his role at Lilly, Diogo is responsible for setting the pharmaceutical leader’s AI strategy, including how organizations use models like Claude to power clinical research and drug development. The two discuss Anthropic’s approach to building more steerable and reliable AI for enterprise deployments, our commitment to creating more skills for life sciences use cases, and the importance of building specialized models to power industry-specific solutions. Learn more about what Claude can do for life sciences: https://claude.com/solutions/life-sciences

DADario AmodeiguestDRDiogo Rauhost

Oct 20, 2025·5m·Watch on YouTube ↗

📖 CHAPTERS

1. 1

   Why enterprises need different AI incentives than consumer apps

   Dario contrasts consumer AI—optimized for engagement and growth—with enterprise AI, where the cost of errors is higher and incentives should favor reliability. He explains how these differing incentives shape model behavior and product priorities.

2. 2

   The risk of “sycophancy” and why enterprises need truth-seeking models

   They discuss model sycophancy—when an AI over-validates the user’s ideas—and why it’s especially dangerous in high-stakes domains. In life sciences, a flattering or overly agreeable model could mislead teams into expensive or harmful decisions.

3. 3

   Anthropic’s positioning: building for economically valuable, reliable work

   Dario describes design choices aimed at smarter performance on economically valuable tasks, emphasizing accuracy and dependability. The conversation frames this as a core differentiator for enterprise adoption.

4. 4

   Why domain depth matters: the biochemistry knowledge example

   Dario uses a biochemistry example to show that deeper domain competence may not matter to most consumers but is crucial for enterprises like Eli Lilly. This illustrates why specialized capability is a meaningful enterprise differentiator.

5. 5

   “Skills” and specialization as the next layer of enterprise AI

   Diogo prompts a discussion about “skills” and enterprise operating capabilities. Dario confirms this direction, describing a roadmap that includes specialized versions of Claude tailored to particular domains and workflows.

6. 6

   Specialized Claudes: fine-tuning vs. wrapping with trusted data access

   Dario outlines two approaches to specialization: improving the underlying model (e.g., fine-tuning) and “wrapping” the model with access to relevant information sources. The goal is to make the model more context-aware and operationally useful.

7. 7

   Case study: Claude for financial services and why integrations matter

   They reference Claude for financial services as an example of connecting the model to indices and ratings. Dario notes that simplifying access to domain data can dramatically improve usefulness in real workflows.

8. 8

   Toward “Claude for life sciences”: combining smarter models with scientific databases

   Dario previews work on a life-sciences-focused Claude that could blend improved intrinsic scientific capability with access to key life science resources. He highlights the breadth of relevant databases (proteins, compounds, assays) that could be put at the model’s fingertips.

9. 9

   Advice for drug discovery teams: don’t get stuck in small, incremental AI swaps

   Dario warns against focusing only on small “hill-climbing” use cases within complex processes. While starting small is natural, he argues teams should plan for ambitious transformation as model capabilities rapidly expand.

10. 10

    Plan for end-to-end workflows now to avoid multi-year adoption delays

    They close on a central message: prepare in parallel for large-scale workflow changes instead of waiting for models to be “ready,” because organizational deployment takes time. Delaying until capability arrives can add years before patient benefits are realized.