Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
Anthropic, Glean & OpenRouter: How AI Moats Are Built with Deedy Das of Menlo Ventures
In this episode, Deedy Das of Menlo Ventures returns to discuss the transformative power of AI in enterprise software and venture capital. From his firsthand experience with Glean’s evolution to early bets on Anthropic, Das sheds light on how foundational AI infrastructure is reshaping industries, investment strategies, and the future of engineering. The conversation dives into the nuances of market dynamics, model performance, and the quiet revolution underway in AI research and deployment.
Deedy Das explores how Glean evolved from a niche search tool into a $7B AI-native platform by solving deep enterprise integration challenges. He highlights Anthropic's unprecedented growth, now capturing 32% of the enterprise LLM market, driven by superior model quality and strategic product design like Claude Code. The discussion covers shifting market shares, the importance of defensible moats in AI investing, and the rise of the $100M Ontology Fund, which backs early-stage startups such as OpenRouter and Goodfire. Emphasis is placed on mechanistic interpretability for model transparency, the economic forces shaping AI adoption, and the reflexivity that amplifies dominance in AI infrastructure. Meanwhile, AI coding tools are transforming development but introduce security risks and potential erosion of engineering skills. The fund’s focus on research-driven innovation signals a long-term vision for responsible, high-impact AI advancements beyond immediate monetization.
05:28
05:28
SaaS tools limiting API access doesn't make business sense as Glean doesn't cut into their revenue.
10:42
10:42
Search tools lack virality and rely on strategic integration for adoption.
18:16
18:16
Claude Code is a critical innovation in AI, demonstrating early value of agents to end-consumers.
25:45
25:45
Claude can charge a premium due to its better model, especially attracting coding users
28:12
28:12
Improving model intelligence may not significantly increase user base for ChatGPT.
33:33
33:33
The model layer is harder to build and may capture more value than the app layer.
41:25
41:25
The Ontology Fund has achieved a high graduation rate to the next funding round for its portfolio companies.
47:30
47:30
Mechanistic interpretability is necessary as AI makes critical decisions
1:02:18
1:02:18
Voice dictation product achieves over 80% zero edit rate
1:07:15
1:07:15
The best technology doesn't always win in markets due to timing, demand, and investment trends.
1:13:55
1:13:55
OpenAI plans to scale from 2 to 30 gigawatts of compute this decade.
1:18:49
1:18:49
AI-assisted coding can turn engineers' brains off like a drug
