20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

In this wide-ranging discussion with Harry Stebbings, Arvind Jain explores the evolving AI landscape and challenges several prevailing assumptions about AI's impact on enterprise software and workforce dynamics. Jain emphasizes that 90% or greater of enterprise use cases can now be fully handled by many different models, including open source options, representing significant commoditization of the model layer. This commoditization undercuts the bundling strategy that has traditionally benefited companies like Microsoft, as consumption-based pricing models eliminate inherent bundling advantages—forcing companies to deliver 10 times the work for the same revenue.

On the competitive threat from frontier model providers like Anthropic and OpenAI, Jain argues that most AI companies should view these providers as strategic assets rather than competitors. While enterprises are concerned about dependency on frontier model companies for their operational workflows and institutional learning, Jain maintains that the real value creation happens at the application layer where context, workflow integration, and domain expertise matter most. He notes that Anthropic's vertical offerings (legal, health, design) have remained shallow and haven't cannibalized existing software, instead expanding the market.

Regarding ROI and AI adoption, Jain identifies a critical throughput problem: most enterprises are deploying AI poorly by connecting it to systems in rudimentary ways and letting models brute-force their way to finding necessary context. This approach burns tokens inefficiently and requires significant investment to structure context properly. Real value exists in specific domains like customer support where productivity gains are measurable, but broader engineering productivity remains difficult to quantify despite developers writing nearly 100% of code through AI tools.

On workforce implications, Jain takes a contrarian stance against the prevailing narrative of team shrinkage. He argues that companies maintaining or growing headcount while competitors shrink will ultimately outcompete them because they can build 10x better products or produce 10x more output with additional human capacity. This directly contradicts claims from many Fortune 500 CEOs who are cutting staff. Jain believes that while per-person productivity will increase, so will demands and product complexity, requiring larger teams. He plans to grow Glean from 1,000 to 5,000-10,000 employees, expecting composite roles blending engineer, product manager, and designer responsibilities.

On open source adoption, Jain reports that enterprises are increasingly interested in open source models, with cost being the primary driver rather than data privacy concerns. The inflection point came when models like GLM 5.2 came within three months of frontier model capabilities. However, adoption of Chinese open source models faces geopolitical hesitation—not from technical concerns but from paranoia about backdoors and competitive disadvantage. He predicts majority of enterprise workloads will run on open source within three years.

Regarding sovereign models and geopolitical AI competition, Jain notes that the desire for sovereign models was stronger a year ago than today, with most nations accepting that they won't build competitive models themselves. The exception is China, which has produced the only substantial non-US open source models. He discusses the tension between regulatory barriers on Chinese models and the need for US open source development, expressing belief in the US system's ability to innovate without capturing open source through regulation.

On token economics and model pricing, Jain notes that contrary to historical expectations, per-token prices increased 15% in recent months despite assumptions they would fall. He questions the fundamental business model sustainability if models become much cheaper, noting that the already loss-making frontier model businesses would face existential threats. Internally at Glean, they implemented a triage agent that costs $1 million monthly—expensive even relative to the on-call team it replaced—raising questions about true efficiency gains.

Jain also critiques current startup ecosystem dynamics, arguing that excess capital is creating unsustainable compensation structures where startups pay engineers $500K-$1 million annually while investors and founders accept these economics. He believes this path isn't sustainable against competitors like Google who don't need to overpay for talent. He advocates for more discipline in capital deployment and proper unit economics rather than relying on continuous fundraising.