Where Value Will Be Created in the AI Era: Insights from Instagram Co-Founder and Anthropic CPO Mike Krieger

1. AI Today and Tomorrow: "We're Still on Day One"

Mike Krieger emphasizes that AI technology is still in its early stages and that models will become increasingly differentiated over time. He notes that AI has not yet become an indispensable tool in most people's work, and looks optimistically at the potential ahead.

"We're still on day one — a day where AI has not yet become an essential tool in most people's work."

• Key points:
  • AI models will grow more differentiated over time.
  • AI is not yet an essential tool for most people.
  • The real value of AI is likely to emerge from differentiated data and knowledge in specific industries (e.g., finance, law, healthcare).

2. How AI Creates Value: Differentiated Data and Market Access

Mike explains that AI can generate long-term value in specific industries through differentiated data and strategic go-to-market approaches. He highlights that while complex industries like healthcare are difficult to enter initially, they offer significant long-term value.

"The sustainable value of AI comes from deep understanding of specific industries and differentiated data."

• Key terms:
  • Differentiated data: Proprietary data accessible only within a specific industry.
  • Go-to-market (GTM): A strategic approach tailored to a specific industry.
  • Healthcare: A complex but long-term high-value opportunity for AI.

3. Incumbents vs. New Startups: Who Benefits from AI?

AI technology presents opportunities for both existing vertical SaaS companies and new startups — but each group faces different challenges.

• Incumbents:
  • Must integrate AI while retaining existing customers.
  • Risk losing trust through overpromising.
• New startups:
  • Face the challenge of limited early data.
  • Must design products around the future potential of AI models.

"Startups must feel frustrated by the limits of current models, yet actively experiment with next-generation models."

4. AI Model Differentiation: "Claude Feels Like Claude, GPT Feels Like GPT"

Mike says that AI models will develop increasingly distinct characteristics over time. Anthropic's Claude model shows particular strength in areas like coding — a result not of chance, but of strategic focus.

"Claude's strength in coding is not a coincidence — it's the result of deliberate strategic focus."

• Key points:
  • AI models are differentiated based on their strengths in specific domains.
  • A model's character and tone have a significant impact on user experience.
  • Claude: Strong in coding and agent-based planning.

5. The Convergence of AI and UX

AI product design is increasingly merging with UX. Mike emphasizes that handling the nondeterministic nature of AI models is a core challenge in product design.

"Designers and product managers need to get comfortable designing for nondeterministic systems."

• Key terms:
  • Nondeterministic systems: Design that accounts for the unpredictability of AI models.
  • Prompt optimization: The process of tuning user inputs for the model.
  • Integration of UX and model quality: Product design and AI model quality cannot be separated.

6. The Pace of AI Progress and Its Limits

AI is advancing rapidly, but Mike points out that significant challenges remain. He emphasizes the importance of building environments that better reflect real-world complexity so that AI models can become more capable collaborators.

"For AI models to become more useful collaborators, they need to better reflect the complexity of the real world."

• Key challenges:
  • Improving training environments: Building training setups that reflect real-world complexity.
  • Evaluation systems (Evals): Better tools are needed to measure model performance.

7. The Future of Data: Human Data vs. Synthetic Data

The discussion also touched on the roles of human and synthetic data in advancing AI models. Mike believes the combination of both will produce the most powerful models.

"The best models will come from a combination of human data and synthetic data."

• Key points:
  • Human data is essential for initial training.
  • Synthetic data is useful for exploring diverse scenarios and improving models.

8. AI and Human Interaction: "Model Personality and Vibes Matter"

Mike says that the personality and "vibe" of an AI model have a significant impact on user loyalty — an important factor that goes beyond raw technical performance.

"People love Claude's personality and vibe. It's something beyond technical performance alone."

• Key terms:
  • Model personality: The characteristics of a model as felt through interaction.
  • Vibes: The sensory quality of the experience of using a model.

9. AI and Coding: The Changing Role of Developers

AI is fundamentally changing how coding is done. Mike predicts that future developers will primarily focus on generating ideas and reviewing AI-generated code.

"The developer of the future will generate ideas and review AI-generated code."

• Key points:
  • AI will automate much of the coding work.
  • Developers will increasingly focus on code review and idea generation.

10. AI's Long-Term Impact: Extending Human Lifespan

Mike believes AI has significant potential to extend human lifespan by accelerating drug development and clinical trials.

"AI will extend human lifespans by accelerating drug development and clinical trials."

• Key terms:
  • Drug development: Accelerating the discovery of new medicines through AI.
  • Clinical trials: Reducing the time required to write clinical trial reports with AI.

Conclusion: The True Potential of AI

Mike Krieger emphasizes that while AI is still in its early stages, it can create long-term value through differentiated data and domain-specific knowledge. He believes AI has the potential to fundamentally transform human life — and that realizing that potential will require continuous innovation and collaboration.

"AI is not just a tool. It has the potential to fundamentally transform human life."