Interview, Fireside Chat
How Glean CEO Arvind Jain Solved the Enterprise Search Problem – and What It Means for AI at Work
- Glean predicts that within five years, AI assistants will execute the majority of knowledge work in domains with full data access and reasoning capabilities, transitioning from reactive tools to proactive guides that handle approximately half of a worker's tasks.
- The company aims to become the primary workplace environment where most work occurs, leveraging its existing deployed search product to facilitate the sale of AI data platforms and agentic capabilities by offering immediate value upon data access.
- Glean plans to evolve agentic capabilities by training models on individual enterprise corpora to understand specific lingo and code names, while explicitly avoiding the training of super-large foundational models in favor of partnerships with scale-focused entities.
- The organization intends to solve 90% of infrastructure burdens, including ETL, data pipelines, and governance, allowing developers to focus on business logic, and will utilize a workflow engine where users manually build automations to train the system for future automated generation.
- The AI market is projected to expand from $600 billion to a range of $12 to $15 trillion, fundamentally altering software development, though the industry is estimated to be only 2% toward a vision where AI handles any question.
- Material risks include the difficulty of building robust, stable enterprise applications that require significant time beyond simple demonstrations, compounded error rates when breaking complex tasks into steps without human input, and challenges in integrations, permissions, and parsing.
- Technical expectations involve chaining imperfect components like retrieval systems and LLMs, where failure often occurs due to stale or irrelevant information, prompting a shift toward RAG-based architectures while current implementations often lack full enterprise context.
- The outlook acknowledges that human input remains critical for building complicated workflows in the near term as error rates compound with complexity, and while every knowledge worker is expected to have access to powerful assistants, the transition to fully autonomous execution is a gradual process.