Lenny's Podcast• Jun 5, 2025• 1:06:21Interview

Anthropic's CPO on what comes next | Mike Krieger (co-founder of Instagram)

From Lenny's Podcast

Mike Krieger•CPO, Anthropic & Co-founder, Instagram

Executive Summary

Anthropic is heavily leveraging its own AI, with roughly 90% of its code now written by AI, dramatically accelerating development but shifting bottlenecks from coding to infrastructure like merge queues.
Mike Krieger, CPO of Anthropic, uses the Claude model as a primary partner for product strategy, noting that recent versions can provide novel, creative analysis that challenges his own thinking.
Anthropic's product strategy focuses on empowering developers and 'builders' by deeply integrating product features with the core model's capabilities, rather than competing directly for mass consumer mindshare.
The future of software development will involve AI agents that can autonomously handle the entire cycle from identifying user feedback to proposing, coding, and deploying a solution, a milestone Krieger believes is achievable in 2024.

12 quotes

Concerns Raised

The increasing volume of AI-generated code can break existing development infrastructure like merge queues.
Building a successful mass-market consumer app remains incredibly difficult, even with advanced AI, as shown by the shutdown of Artifact.
The deteriorating quality of the mobile web ecosystem can act as a significant headwind for content-based applications.

Opportunities Identified

Leveraging AI to automate the entire software development lifecycle, from user feedback analysis to deployment.
Building defensible AI businesses by focusing on deep user knowledge and creating novel product form factors.
Empowering non-technical roles (PMs, designers) with AI tools to build functional prototypes and accelerate ideation.
Using advanced AI models as a creative and strategic partner to generate novel insights and overcome cognitive biases.

Key Themes

AI-Driven Software Development

Anthropic has transformed its internal software development process to be AI-first, with models writing the vast majority of code. This has led to a massive increase in output, with over 70% of pull requests being AI-generated.

This represents a paradigm shift in engineering, demonstrating that the primary bottleneck is no longer coding speed but rather the infrastructure for testing, reviewing, and merging code, as well as the quality of human ideation.

The Shifting Bottleneck in Engineering

As AI dramatically increased the volume of code being written at Anthropic, the company's primary constraints shifted. The engineering bottleneck moved from individual developer velocity to systemic issues like the capacity of the merge queue and the speed of code reviews.

This is a critical lesson for any organization adopting AI-assisted coding at scale. It highlights the need to proactively re-architect CI/CD pipelines and review processes to handle a much higher throughput of code changes.

Human-AI Collaboration for Strategy and Creativity

Beyond coding, AI is becoming a legitimate partner in high-level strategic and creative work. Mike Krieger details his use of Claude as a sparring partner for product strategy, while non-technical staff use it for prototyping, demonstrating AI's utility across the entire product lifecycle.

This signals a move from AI as a simple tool for task execution to a collaborator that can augment human creativity and critical thinking, offering novel perspectives even in complex domains like business strategy.

Product Strategy for Foundational AI Companies

Anthropic is carving out its niche by focusing on the 'builder' and developer community, creating powerful, deeply integrated features that showcase the model's unique strengths. This contrasts with a direct pursuit of mass-market consumer applications, aiming to build a defensible moat through a strong developer ecosystem.

For companies in the competitive AI space, this strategy suggests that differentiation can come from focusing on a specific, high-leverage user segment and building a product that uniquely serves their advanced needs, rather than engaging in a feature-for-feature race for the broadest possible audience.

Get started free

Topics

AI-assisted coding Software Development Lifecycle (SDLC)Engineering Bottlenecks Product Strategy AI Product Management Anthropic Claude Mike Krieger Stripe Competitive Landscape AI Startups Developer Tools Human-AI Collaboration Model Capabilities SWE-bench

Processed Apr 3, 2026 yt-dlp + mlx-whisper + Gemini