Episode Summary
Executive Summary: The episode centers on Glean’s enterprise AI search product and a broader debate about the future of AI in business. Arvind Jain argues that most AI applications will be built on open-source/open-domain models, while Glean remains model-agnostic and uses strong permissions controls to safely answer questions across company systems. The conversation also explores search vs. chat, AI’s impact on knowledge work, compliance, and how enterprises will adopt AI at scale.
Main Topics: Glean’s enterprise AI search assistant (Priority: 5/5): Arvind explains Glean as an internal Google/ChatGPT for companies, letting employees ask questions across corporate knowledge sources and get answers grounded in their organization’s data. Permissions, privacy, and enterprise security (Priority: 5/5): A major focus is how Glean preserves access controls across Slack, Google Drive, Jira, GitHub, and other systems, ensuring users only see information they are authorized to access. Model strategy: GPT-4, Gemini, and open source (Priority: 5/5): Glean is model-agnostic, uses multiple LLMs, and can support customer-hosted/open-domain models. Arvind argues open source will dominate most AI inference and application building long term. Search vs. chat as the future interface (Priority: 4/5): Arvind argues the future will not be two separate interfaces; instead, search and chat will converge into one adaptive interface that returns links, summaries, or direct answers based on intent. Enterprise knowledge graphs and productivity analytics (Priority: 4/5): The discussion explores how Glean can connect fragmented work artifacts across tools and potentially surface productivity insights for individuals and managers, though that is not its primary mission. Compliance, legal, and governance use cases (Priority: 4/5): Beyond day-to-day search, Glean can support restricted compliance workflows such as e-discovery, privacy requests, and broader governance by indexing enterprise content while honoring retention/deletion rules. OpenAI, Google, and the competitive AI landscape (Priority: 4/5): Arvind says OpenAI is still ahead in text models, but the gap is narrowing quarterly. He is optimistic about Google’s AI position and believes open models will ultimately win market share.
Key Arguments: Glean solves a distinct enterprise problem: knowledge is spread across many systems, so employees need a safe way to search and synthesize across them. The product is permissions-aware by design, indexing content and access rules so users only see what they are allowed to see. Glean does not train or fine-tune customer data into GPT-4; it uses LLMs only for summarization/synthesis after retrieving authorized snippets. Enterprise trust in major cloud providers (Microsoft, Google, AWS) is high enough that many customers accept hosted models if VPC/security controls are in place. Open source/open-domain models will likely power the majority of AI inferencing and app-building long term because open source is hard to beat on momentum and adoption. The future of search is an adaptive interface, not a strict chat-vs-search split; the system should answer with the right format for the user’s intent. AI will make existing governance gaps more visible because it can surface information that was already poorly permissioned but previously hard to find. Google remains strong in AI due to talent, infrastructure, and data centers, even if product rollout feels slower because of caution and organizational complexity.
Data Points: Predicted share of AI work based on open models: 80% - Arvind predicts most AI inferencing and app-building will be based on open-domain/open-source models. Google tenure: 2003 to 2014 - Arvind says he worked at Google for about 11 years. LLM usage at Glean: GPT-4, Gemini, and others - He says Glean is model-agnostic and uses multiple large language models. Primary current deployment model: GPT-4 - He says most deployments currently use GPT-4, mainly via Azure. Customer deployment time: Within a day - He says Glean can be up and running in a 2,000-person enterprise very quickly. Enterprise focus size: A few hundred to the largest enterprises - Glean targets mid-market through very large organizations. Developer vetting rate on ad read: 1% - Lemon.io ad mentions only 1% of applicants get in. Developer experience threshold: 3+ years - Lemon.io ad says developers are hand-picked with at least three years of experience. Slack admin activity window: Last 30 days - Jason references Slack admin reporting that shows messages sent and login days across the last 30 days. Twist listener discount on OpenPhone: 20% off for 6 months - Ad read for OpenPhone. Twist listener discount on Lemon.io: 15% off for 4 weeks - Ad read for Lemon.io. Twist listener discount on Gelt: 15% off for first year - Ad read for Gelt.
Pivotal Quotes: "I think in the future, the majority of AI work is going to be based on open source models." — Arvind Jain: On the long-term winner between closed and open models. "Glean is an AI-powered search engine. We are an AI-powered assistant that helps people get more work done." — Arvind Jain: Defining the core product. "We are not actually training or fine-tuning models like GPT-4. We're actually using them only as summarization and synthesis engines." — Arvind Jain: Explaining Glean’s data/privacy approach.
Implications: Enterprise AI will hinge on safe retrieval, permissions, and cross-system context more than raw model capability. Open-source models are likely to gain share, while enterprises will demand tight governance and flexible deployment options.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.