Episode Summary
Executive Summary: Arvind Jain explains how LLMs transformed enterprise search from keyword matching into semantic understanding, enabling Glean to evolve from a search product into an AI assistant and workflow platform. He emphasizes that SaaS APIs, cloud scale, and permission-aware governance made enterprise AI viable, but that relevance, freshness, and security still require classic search infrastructure. Glean now aims to create AI-first employees and personalized digital teams.
Main Topics: LLMs changed the core search paradigm (Priority: 5/5): Jain argues search has shifted from brittle keyword lookup to deep understanding of both questions and documents, allowing systems to match intent to concept and answer directly rather than merely return links. Glean’s technical evolution from semantic search to AI assistant (Priority: 5/5): The company began with embedding-based semantic matching before RAG and generative AI were mainstream, then expanded into a ChatGPT-like assistant that uses both public knowledge and internal company data securely. Why enterprise search became feasible now (Priority: 5/5): Jain says earlier enterprise search failed because pre-SaaS systems were hard to access and not interoperable. SaaS APIs, connectors, and cloud infrastructure solved data access and scale, making turnkey enterprise search possible. Governance, permissions, and security as a product pillar (Priority: 5/5): A major theme is that enterprise AI must enforce permissions, authority, and sensitivity at the document and query level; Glean effectively became a security/governance layer to make AI safe in companies. From search product to apps/agents for business processes (Priority: 4/5): Beyond answering questions, Glean now supports curated apps/agents that use approved content and perform tasks inside workflows, driven by customer demand for functional AI that can replace parts of business processes. Go-to-market: enterprise sales over pure PLG (Priority: 4/5): Jain explains why Glean could not rely on self-serve PLG alone: the product is company-wide by nature, requires broad indexing, and needs top-down rollout, though he still sees PLG as an important lead channel. AI adoption requires education and behavior change (Priority: 4/5): He notes users often still behave like old search users, typing short queries instead of long instructions. Companies must teach employees how to use AI, and leaders should value AI fluency as a workforce capability.
Key Arguments: LLMs are foundational because they let systems deeply understand both user intent and document meaning, which makes search far less brittle than keyword-based methods. Enterprise search only became practical after SaaS replaced on-prem, versioned software with API-accessible, interoperable systems that can be indexed continuously. Classic IR still matters: models alone are not enough; search systems must prioritize freshness, authority, correctness, and structured presentation of information. Enterprise AI cannot be built safely without permissions-aware infrastructure, because most company knowledge is private and exposure of sensitive data is a serious risk. Glean’s product had to evolve from a search box into an assistant and then into curated apps/agents that can both answer questions and perform work. The hard part of AI adoption is not only ROI, but also training employees to use AI effectively and become AI-first workers. For a product like Glean, enterprise sales is structurally necessary because value requires indexing the whole company’s corpus; PLG can complement, not replace, that motion. The biggest remaining challenge is not just hallucination, but retrieving the right information at all, since enterprise knowledge is often incomplete, stale, or poorly organized.
Data Points: Founding timeline: Late 2018 / early 2019 - Jain says Glean was conceived in late 2018 and started in early 2019. Search experience duration: Almost 30 years - He says he has been working on search for almost 30 years. Customer document scale: More than 1 billion documents - He cites one large customer with over 1 billion documents in its company corpus. Internet scale reference: 1 billion documents - He notes that in 2004 the entire internet was around 1 billion documents, comparing public-web and enterprise content scale. Enterprise data systems at Rubrik: 300 different SaaS systems - He says Rubrik had information spread across roughly 300 SaaS systems, motivating the original internal need for Glean. Enterprise knowledge privacy: 90% - He estimates about 90% of company knowledge is private in some form and subject to permissions. AI productivity target: 90% of your work - He describes the future assistant vision as doing 90% of work for the user. Workforce transformation goal: 10X - He says the goal is to build a personal team that makes each individual a 10X contributor.
Pivotal Quotes: "LLMs have completely changed it... it has allowed us to really deeply understand a question that a user is asking." — Arvind Jain: On how search has changed from keyword-based retrieval to semantic understanding. "Any AI experiences that you build inside the company has to think about security and governance and permissions like at a fundamental level." — Arvind Jain: On why enterprise AI must be permission-aware and secure by design. "We had to create the market for this." — Arvind Jain: On the challenge of selling a category that did not previously exist in enterprise software.
Implications: Enterprise AI winners will combine model intelligence with search discipline, security, and enterprise integration. The market is moving from simple retrieval to permissioned assistants and agents that can execute workflows, but adoption will depend on education, governance, and workflow-level ROI.