Summary
Harry Stebbings interviews Arvin Gin — co-founder of Glean and previously Rubric — about why frontier model companies like OpenAI and Anthropic face structural limits in capturing the enterprise app layer. Gin’s core argument: as AI models commoditize (he estimates 90% of use cases are now addressable by many models including open source), the bundling advantage model providers might expect simply doesn’t materialize. Enterprise customers are simultaneously attracted to and wary of frontier providers, fearing excessive technology dependence and loss of proprietary IP and institutional knowledge.
Gin describes Glean’s evolution from enterprise search — essentially “Google for your work life” connecting employees across hundreds of internal systems — to a full AI platform positioned as a superset of ChatGPT, Claude, and Gemini for the workplace. He shares that Glean now generates nearly 100% of its code with AI assistance, while maintaining mandatory human code review, arguing the bottleneck has shifted from writing to reviewing and that eliminating reviews (as some companies are doing) creates long-term maintainability risk.
The conversation also challenges the narrative that enterprise AI “isn’t working,” with Gin suggesting the real friction is ROI measurement, throughput bottlenecks, and organizational context gaps rather than model capability. His conclusion: for companies not training frontier models, AI providers are a massive asset — but consumption-based pricing models require disproportionate effort to generate equivalent revenue, creating a durable opening for application-layer companies like Glean.
📺 Source: 20VC with Harry Stebbings · Published July 11, 2026
🏷️ Format: Interview
