In AI Agent's World - Emrecan Dogan on Building Glean's Next Gen Search and Productivity | EP77
🎯 Summary
[{“key_takeaways”=>[“Glean’s core value proposition is resolving identity and permissions across 100+ enterprise apps in real-time to ensure users only see authorized data.”, “Glean was built from the start for large enterprises where information asymmetry explodes with scale, differentiating it from solutions suited for smaller teams.”, “The platform supports an evolution from search to AI assistants (synthesis/content generation) to AI agents (end-to-end task execution).”, “AI agents can be personalized to mirror an individual employee’s unique working style or standardized across departments for consistent processes (e.g., financial reporting).”, “Glean relies on a sophisticated RAG (Retrieval Augmented Generation) architecture, arguing that fine-tuning LLMs directly sacrifices essential permission enforcement and governance.”, “Glean maintains a ‘lean, minimalist’ approach to data ingestion, continuously integrating with data sources via connectors without copying or replicating customer data onto Glean servers, preserving security boundaries.”, “The freshness and relevance of search results are enhanced by analyzing activity signals (engagement, comments, presentations) to determine the authoritative source among many potential documents.”], “overview”=>”Emrecan Dogan, Head of Product at Glean, discusses how the company is solving the complex challenge of enterprise knowledge discovery across hundreds of disparate SaaS applications by building a deep, permission-aware knowledge graph. The conversation highlights Glean’s evolution from enterprise search to AI assistants and, critically, to AI agents capable of executing tasks end-to-end, personalized to individual work styles or standardized across departments. A core tenet of Glean’s success is its RAG-based architecture, which maintains strict data governance by indexing metadata rather than copying sensitive company data.”, “themes”=>[“Enterprise Knowledge Management and Search”, “AI Agent Architecture and Personalization”, “Data Governance, Permissions, and Security in AI”, “The Role and Importance of RAG vs. Fine-Tuning”, “Scaling AI Solutions for Large Enterprises”, “Evolution of AI from Assistants to Agents”]}]
🏢 Companies Mentioned
💬 Key Insights
"As we have observed models ended up having larger and larger context windows, so that the next challenge was, hey, is RAG relevant anymore when you can supply all the documents you want through the context window interface into an LLAM? We are still landing on a very precarious problem, and it is the fact that you can supply more information doesn't mean that information is the highest quality information."
"The moment you take an LLM, let's say you have the resources to train or fine-tune a large language model with company data, and you supply in that training, let's say, some portion of company data. Now you are losing something incredibly important, which is all the permissions and governance on that data."
"But then on top of it, then we build AI assistants where you are no longer finding documents or answers, but you are actually tipping into synthesis or you are generating new content of that search. And finally, the AI agents. I think the pendulum swings from generating answers or getting answers to actually executing tasks and getting AI to do things for you end-to-end."
"What I loved about Glean at the time was the team started by solving one data, what I think as well by solving the hardest part of the enterprise AI puzzle or problem. And it is finding and understanding company knowledge across every app, across every person, every office, every project with a hybrid search engine and a very deep and fully autonomous knowledge graph."
"resolving this to understand who is Dimitri and how is Dimitri presented in different pieces of software. What is the information? What is the data Dimitri should be allowed to see and what shouldn't be provided to Dimitri? Understanding this in real time, this is really the core of what Glean brings to the table."
"I will say the big moment of truth with Glean agents, I think this holds true for the broader agents movement in the industry, is agents bring the right combination of determinism, like the traditional software paradigm is determinism. Software does the same thing again and again with the huge upside of stochasticity that LLMs bring to the table."