The Twenty Minute VC (20VC)
Episode: Anthropic CPO Mike Krieger: Where Will Value Be Created in a World of AI | Have Foundation Models Commoditized | When Do Model Providers Become Application Providers | What Anthropic Learned from Deepseek
Release Date: March 3, 2025
Host: Harry Stebbings
Guest: Mike Krieger, Co-founder of Instagram and Chief Product Officer at Anthropic
Introduction
In this compelling episode of The Twenty Minute VC (20VC), host Harry Stebbings welcomes Mike Krieger, the renowned co-founder of Instagram and current Chief Product Officer at Anthropic, a leading AI company. The discussion delves deep into the evolving landscape of artificial intelligence, exploring where future value will be generated in an AI-driven decade, the commoditization of foundation models, the transition from model providers to application developers, and the insights Anthropic has gleaned from competitors like Deepseek.
Where Will Value Be Generated in an AI-Driven Decade?
Mike Krieger opens the conversation by addressing the critical question of where value will be created as AI continues to advance. He asserts that the most durable value resides in areas where companies possess differentiated go-to-market strategies, specialized industry knowledge, or unique data access—ideally a combination of these factors. Mike emphasizes sectors like finance, legal, and healthcare as prime candidates due to their complexity and the substantial groundwork required to excel within them.
“The thing that's going to give you legs and be durable over the long run is being able to sell into those places, have something that you understand about those places uniquely, and then get better at being deployed there over time.”
— Mike Krieger [05:22]
Foundation Models Commoditization and Differentiation
Harry probes whether the next wave of AI will benefit existing vertical SaaS companies or foster entirely new startups. Mike responds by highlighting the dual potential:
-
Existing Vertical SaaS Companies: These companies can integrate AI to enhance their offerings without alienating their current customer base. However, they must be cautious not to overpromise capabilities that current models cannot deliver reliably.
-
New Startups: Agile startups can experiment with AI innovations more freely, capitalizing on advancements as models improve. Mike underscores the importance of startups continuously iterating and leveraging AI advancements to unlock new value propositions.
“Don't wait around for the models to be perfect, be exploring in this space, be frustrated by the current generation of models, and then be very aggressively trying the next one so that you can feel like you can now finally deliver on the thing that you saw in your head.”
— Mike Krieger [07:56]
Model Quality vs. Product in User Experience (UX)
The discussion shifts to the intricate relationship between model quality and product design. Mike argues that AI models and product UX are now inextricably linked. Designers and product managers must account for the non-deterministic nature of AI, ensuring that the integration of models enhances rather than disrupts user experience.
“You can't separate the two anymore. Designing a scaffold and a product around a fundamentally non-deterministic system means that model quality, prompting, and backend optimizations are direct components of product design.”
— Mike Krieger [20:29]
AI in Software Development and Coding
When exploring AI's impact on software development, Mike shares his experiences with Anthropic’s Claude Code, a tool designed to assist in coding by finding and editing code files efficiently. He envisions a future where developers transition from writing code to delegating tasks and overseeing AI-generated solutions, emphasizing the evolving role of software engineers as managers and collaborators with AI.
“The role ends up looking like being more of a manager and delegator to these things rather than just a partner in the loop.”
— Mike Krieger [44:54]
Challenges and Blockers in AI Development
Mike identifies the primary technical challenge as creating training environments that accurately reflect real-world, multi-step processes. Current models excel in narrow tasks but struggle with broader, more complex interactions that require deeper understanding and adaptability.
“Figuring out how to better break down complex environments and think about them holistically is the biggest blocker.”
— Mike Krieger [13:31]
Synthetic Data vs. Human Data
The conversation touches on the future of data in model training. Mike advocates for a hybrid approach, combining original human data with synthetic environments that allow models to explore diverse scenarios. He cites examples from gaming, such as Pokémon, to illustrate how synthetic data can enhance a model's ability to handle uncertainty and varied approaches.
“It absolutely has to be a mix. The best models will come from a combination of great human data and the ability to generate diverse synthetic environments.”
— Mike Krieger [16:02]
Model Selection and Abstraction Leaks
Harry raises concerns about the increasing complexity of model selection for end-users, likening it to choosing between different versions of Google. Mike agrees, highlighting the concept of "leaky abstractions" where the underlying complexities of AI models expose themselves to users. He emphasizes the need to simplify user interactions with AI, avoiding the necessity for users to understand model differences.
“The overall experience is one of why would I choose one over the other? We suffer from this problem as well. So model selection needs to be collapsed further.”
— Mike Krieger [18:22]
Release Speed and Product Marketing
The rapid pace of AI model releases creates a "product marketing nightmare," with constant updates making it challenging to maintain stable product messaging. Mike discusses Anthropic’s strategies to balance rapid iteration with the need for stability, such as opting features into beta and carefully managing enterprise customer expectations.
“Every time there's a model launch, I assume every one of those labs is either watching the launch stream or evaluating their next steps. It's about getting used to the bumpy ride.”
— Mike Krieger [26:35]
AI and Human Relationships
A thought-provoking segment explores the potential for AI to become integral to human relationships, particularly among younger generations. Mike expresses both optimism and caution, acknowledging AI’s role in augmenting social interactions while recognizing the irreplaceable value of genuine human connections.
“There is a lot you can learn from what it feels like, but it is absolutely insufficient as the whole. AI can be a helpful piece of future human interaction but not the entirety.”
— Mike Krieger [57:46]
Global Perspectives: Europe and China in AI
Mike addresses the underestimated capabilities of China in AI development, stressing the importance of recognizing China's advanced research teams and innovative startups. He also highlights Europe's unique strengths, such as stringent data privacy laws and societal values, which can influence global AI practices.
“It's a mistake to underestimate China's ability to train at the frontier, especially with access to massive compute resources and innovative startup ecosystems.”
— Mike Krieger [33:33]
AI and Longevity
In response to a question about AI's impact on human longevity, Mike expresses optimism about AI accelerating medical research, particularly in drug discovery and clinical trials. He cites examples where AI has significantly reduced the time needed for research processes, underscoring AI's potential to contribute to medical breakthroughs.
“AI is helping close the loop on drug discovery and clinical trials. For instance, Novo Nordisk uses Claude to complete clinical trial reports in 20 minutes instead of weeks.”
— Mike Krieger [61:22]
Quickfire Round
In a rapid-fire segment, Harry poses several concise questions to Mike, eliciting candid responses:
-
What has OpenAI done better than you?
“They've moved faster at shipping V1s, even ahead of where the model sometimes is.” -
What have they done worse than you?
“Probably personality and having the features they built be cohesive.” -
Which alternate model provider do you respect most?
“They’ve balanced first-party product development and an API that people use at scale as well.” -
If you could rebuild the Anthropic product stack from scratch, what would you do differently?
“Simplify the information architecture so users don’t have to differentiate between projects, artifacts, and chats.” -
What have you changed your mind on in the last 12 months?
“The importance of investing more in first-party products alongside our API offerings.” -
What is a major technical or product challenge on the horizon in AI that no one’s talking about?
“Ensuring discernment and privacy in increasingly knowledgeable models, preventing them from revealing sensitive information.”
Conclusion
The episode with Mike Krieger provides invaluable insights into the future trajectory of AI, emphasizing the importance of specialized knowledge, thoughtful product design, and the delicate balance between rapid innovation and user stability. Mike's reflections on Anthropic's strategies, challenges, and the broader AI landscape offer a nuanced perspective for investors, entrepreneurs, and technologists navigating the rapidly evolving world of artificial intelligence.
Notable Quotes:
-
Mike Krieger [05:22]:
“The thing that's going to give you legs and be durable over the long run is being able to sell into those places, have something that you understand about those places uniquely, and then get better at being deployed there over time.” -
Mike Krieger [07:56]:
“Don't wait around for the models to be perfect, be exploring in this space, be frustrated by the current generation of models, and then be very aggressively trying the next one so that you can feel like you can now finally deliver on the thing that you saw in your head.” -
Mike Krieger [20:29]:
“You can't separate the two anymore. Designing a scaffold and a product around a fundamentally non-deterministic system means that model quality, prompting, and backend optimizations are direct components of product design.” -
Mike Krieger [44:54]:
“The role ends up looking like being more of a manager and delegator to these things rather than just a partner in the loop.” -
Mike Krieger [26:35]:
“Every time there's a model launch, I assume every one of those labs is either watching the launch stream or evaluating their next steps. It's about getting used to the bumpy ride.” -
Mike Krieger [61:22]:
“AI is helping close the loop on drug discovery and clinical trials. For instance, Novo Nordisk uses Claude to complete clinical trial reports in 20 minutes instead of weeks.”
Note: Advertisements and non-content sections have been omitted to focus solely on the substantive discussion between Harry Stebbings and Mike Krieger.