Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this returning episode of the Latent Space podcast, Deedy Das, now Partner at Menlo Ventures, shares his journey from Glean to venture capital and provides insider insights on the explosive growth of Anthropic. (11:00) The conversation covers Glean's evolution from "boring" enterprise search to a $7B AI-native company, Anthropic's meteoric rise to become potentially the fastest-growing software company in history, and the shifting dynamics in enterprise AI market share. (25:00) Das discusses managing the $100M Anthology Fund, investing in next-generation AI infrastructure and research companies like Goodfire and OpenRouter, and explores critical questions about the future of coding, AI safety, and venture capital in the age of artificial intelligence.
- Main themes: Enterprise AI transformation, venture capital evolution, the intersection of AI research and commercialization, and the changing landscape of software engineering in the AI era
Speakers
Deedy Das
Partner at Menlo Ventures and manager of the $100M Anthology Fund. Previously worked at Glean during its transformation from enterprise search to a $7B AI-native company. Das has been instrumental in Menlo's investments in Anthropic across multiple rounds and has invested in over 40 AI companies through the Anthology Fund, including notable successes like OpenRouter and Goodfire.
Alessio Fanelli
Founder of Kernel Labs and co-host of the Latent Space podcast. Expert in AI and enterprise technology.
Swyx
Editor of Latent Space and co-host of the podcast. Prominent AI researcher and commentator focused on the intersection of AI and developer tools.
Key Takeaways
Enterprise Search Requires Deep Technical Investment, Not Just AI Layering
Das emphasizes that Glean's success came from solving fundamental enterprise search problems before AI acceleration, not from simply adding LLMs to existing systems. (04:00) The company spent years addressing critical challenges like permission-aware search, handling fresh data with limited feedback loops, and creating viral adoption mechanisms in non-social products. This foundational work became invaluable when AI transformed the landscape, demonstrating that sustainable competitive advantages come from doing hard technical work that competitors won't replicate.
Model Layer Captures More Value Than Application Layer in AI Stack
Das argues that the hardest technical challenges typically capture the most value in any technology stack. (33:30) While app-layer companies build valuable products, it's significantly easier for model providers like Anthropic to enter application spaces than for applications to build frontier models. This dynamic suggests that despite short-term success of AI applications, the fundamental model layer will likely capture the majority of long-term value, similar to how infrastructure companies historically outperformed applications built on top of them.
Research Investing Requires Balancing Deep Technical Bets with Product Commercialization
Investment in AI research companies like Goodfire (mechanistic interpretability) and Prime Intellect (distributed training) represents a high-risk, high-reward strategy. (45:00) Das emphasizes following talented teams with strong technical competence toward problems that will likely exist in ten years, even if current commercial applications aren't obvious. The key is identifying when breakthrough research can translate into massive business opportunities, requiring patience with the research timeline while maintaining focus on eventual commercialization paths.
AI Coding Tools Risk Creating Dependency That Weakens Engineering Skills
Das expresses concern about "vibe coding" where engineers rely heavily on AI assistance without deeply understanding the code being generated. (77:00) This creates a dangerous cycle where the joy and skill-building aspect of solving hard problems is replaced by constant AI prompting. The risk is particularly acute for junior engineers who may never develop the fundamental problem-solving skills that senior engineers use to evaluate AI-generated solutions. This trend could fundamentally change the craft of engineering, potentially weakening the overall technical capability of development teams.
Corporate Venture Funds Succeed When Structured as Independent Investment Vehicles
The Anthology Fund's success stems from being managed externally by Menlo Ventures rather than as an internal Anthropic corporate venture fund. (40:00) This structure avoids the common corporate VC pitfall of prioritizing strategic usage over investment returns. Das explains that traditional corporate VCs often fail because they invest based on "who uses my stuff the most" rather than fundamental business quality. The external structure allows for better investment decisions while still providing strategic value to Anthropic through ecosystem development.
Statistics & Facts
- Anthropic has achieved what Das calls "the fastest growing software company of all time" trajectory: zero to $100M ARR in one year, $100M to $1B in one year, with projections of $1B to $9B this year. (16:00)
- OpenAI's enterprise API market share has declined from 50% in 2023 to 25% in 2025, while Anthropic's share has grown from 12% to 32% in the same period, based on Menlo Ventures' enterprise LLM API spend survey. (23:00)
- In 2024, OpenAI spent $7B total on compute, with only $2B going to inference (serving all ChatGPT users and API customers) while $5B went to R&D and training new models. (76:00)
