Glean

What does it take to make agentic AI work in the enterprise? 🤔 Arvind Jain, our founder & CEO, recently joined leaders across the industry at Berkeley RDI's Agentic AI Summit to discuss this important question. A few standout themes from the conversation: 🔍 𝗕𝗲𝘆𝗼𝗻𝗱 𝘁𝗵𝗲 𝗯𝘂𝘇𝘇: While prototypes impress, production-ready agents are still rare, and separating substance from the noise remains a challenge. 🏗️ 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗿𝗲𝗮𝗱𝗶𝗻𝗲𝘀𝘀 𝘃𝗮𝗿𝗶𝗲𝘀 𝗴𝗿𝗲𝗮𝘁𝗹𝘆: Most companies are still experimenting. The ones moving fastest are those building from the ground up, not retrofitting AI into legacy processes. 🧠 𝗧𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗯𝗼𝘁𝘁𝗹𝗲𝗻𝗲𝗰𝗸𝘀: Hallucinations, unreliable function calling, and fragmented data hinder progress – making high-quality search and retrieval essential for context-aware agents. 🔐 𝗚𝗼𝘃𝗲𝗿𝗻𝗮𝗻𝗰𝗲 𝗶𝘀 𝘁𝗮𝗯𝗹𝗲 𝘀𝘁𝗮𝗸𝗲𝘀: Beyond access controls, agents need smart, context-aware constraints to ensure enterprise data is used safely. Thank you to Berkeley RDI and Sharon Goldman at Fortune for the great conversation and event!

OpenAI GPT-5 🤝 Glean We’re excited to support GPT-5 in Glean starting today – right from day one. 🎉 This reflects our commitment to helping enterprises try the latest AI models and put them to work with full enterprise context. With early access from OpenAI, we evaluated GPT-5 and found that it outperforms o3 across the board on correctness, completeness, and alignment with human feedback. See the performance difference for yourself, on your own enterprise agents – put GPT-5 to work today. And check out the eval blog to learn more about the performance of GPT-5: https://lnkd.in/gGZS8ttb

Introducing the Glean agent toolkit: a Python library for building context-aware AI agents that tap into your company's unique data. Define tools once and reuse them across frameworks like OpenAI, LangChain, CrewAI, and Google ADK. Steve Calvert walks us through using the toolkit to debug a CI/CD pipeline by aggregating signals from GitHub logs, Jira tickets, Confluence docs, Slack threads, and more. Thanks to MCP, we deployed the agent directly into the Cursor IDE, so engineers can troubleshoot right where they code. 💥

We’re exploring what it really takes to make AI work at work. If that’s a conversation you want to hear at SXSW 2026, vote to bring it to the stage. 🗳️ 🔷 Enterprise AI at scale with Arvind Jain, CEO at Glean, David Beitel, CTO at Zillow, and T. Alexander Puutio at Forbes: https://lnkd.in/gVTRhhmr 🔷 Getting the most out of your enterprise data with AI with Arvind Jain, CEO at Glean, and Baris Gultekin, VP of AI at Snowflake: https://lnkd.in/gAqTHEa2 🔷 AI Alignment and distributed work with Arvind Jain, CEO at Glean, and Dan Spaulding, CPO at Zillow: https://lnkd.in/gtPTUUus Your voice is important! 🗣️ Community voting makes up 30% of SXSW final selection. Voting closes August 24, so don’t miss your chance! Thanks for your help in bringing our ideas to SXSW 🙌

Hot take: Building AI agents that demo well ≠ building agents your team actually uses That's where evaluations come in – evaluations help you refine agents based on real user data. We cover how you can evaluate individual agents to make them better, and how we approach evaluations at scale to keep agents reliable across the enterprise in our latest blog 👉 https://lnkd.in/gt9uSiM6

What's it like interning at Glean? ✨ For National Intern Day, we asked our interns to share their biggest takeaways from their time at Glean – advice, lessons, and favorite moments. Swipe through to see what Christina Wu, Sriram Varun Vobilisetty, Sydney Schirrmeister, Jennifer Ying, and Abhishek Khandelwal had to say 👉

🐌 Playing around with Glean agents and the no-code workflow builder is pretty seamless. In this example, I created a simple OKR coach where I just updated the system prompt. The workflow builder feels a bit like Zapier in that I define 1) the steps in the agent (e.g. nodes in the graph) 2) system prompts for each step/node and 3) which internal data sources I want the agent to read before performing the inference. I think these simple chatbots and workflow builders are good enough for 80% of the use cases out there if your company is just looking for better answers to questions and summarizing disparate data sources on internal data. The time to value is super quick because Glean already has been approved to read a bunch of internal data sources like Gong, Coda, Slack, Confluence, and more.

Review season deja vu: “Wait, what did I even do this year?” 🤨 As Talent Development Director, Michelle shares how Glean helps teams reflect on a full year’s work with clarity and consistency. Watch why she calls it “a game changer” 👇🎥

Writing launch notes doesn’t have to be a chore 🙂↔️🙅 The Launch Documentation Agent handles it for you – pulling completed Jira tickets, summarizing features, and drafting a polished PR so you can save hours and keep shipping. It’s fully customizable: choose your LLM, set your templates and fields, and define the output format to match your team’s exact workflow. Kat Crichton from our customer experience team shows how it works 👇

I built our first AI-powered Content Design assistant at Intercom! 🤖 Creating UI content is a daily task for most of our Product Designers. But the tools that help them do it – our style guides, our messaging, our research – are scattered across docs and decks, making it hard to guarantee consistent content across teams. So I trained an AI agent to fill this gap. It’s called VERBI and it uses Glean to pull from a curated collection of internal resources. Most importantly, it can read our design system, meaning it knows how to write for every component in our product. Toggles, toasts, labels, tooltips? VERBI knows every content rule and is ready to help. This new agent can: - Write: create content from scratch in our tone of voice - Edit: instantly improve clarity, consistency and style - Review: offer content feedback and critiques - Name: suggest new names for product features - Strategise: help to rationalise content choices - Guide: deliver content guidelines on demand - Jam: find alternative words, phrases, headlines - Inform: pull in our latest research and data It’s not a replacement for good thinking or good writing. It’s how we scale both of these across a growing team, setting new benchmarks for quality and speed. It’s also not perfect (yet). I’ll probably be tweaking the system prompt for weeks to come. But it’s live, it’s working, and it’s helping us get better one iteration at a time 📈 #ContentDesign #AI #ProductDesign #UXWriting #DesignTools #Innovation