An AI agent that doesn’t know your company burns money trying to prove it. That’s the argument Glean is making as it pitches why context is critical to the success of any enterprise AI program. At its annual customer conference, Glean:GO, in San Francisco this past week, that was the only thing the company wanted to talk about, announcing a slate of new offerings designed to capture, route, and apply context so AI does what’s asked of it.
“When we talk about context, everybody has different ideas in their mind,” Arvind Jain, Glean’s co-founder and chief executive, said at a press and analyst briefing on Tuesday. “The way we talk about context is ultimately all the right information, experience, [and] the wisdom of humans that you would need to complete a task.” He said that it’s all available inside the enterprise. It’s just buried, hard to reach, and scattered across formats. Some of it is clean, structured data or official documents. The rest is messy, everyday signals people leave behind in chats, meetings, and approvals. “We collect all of it…and then we use the power of AI to organize it,” Jain said.
Context was so top of mind at Glean’s conference that it almost felt as if the company was going to have its Steve Ballmer moment. In fact, the term was uttered 61 times during the briefing, as noted by “Ground Level AI” journalist Sharon Goldman.
That repetition was the argument. Most of the industry has answered unreliable AI by building upward, adding evaluation suites, guardrails, and human review queues on top of models that keep getting things wrong. Glean sought to make the case that the fix belongs further down. An agent that gets the answer wrong doesn’t need another layer of supervision watching it fail; it needs to have been given what it required in the first place. That’s a case for retrieval-augmented generation, and it’s the ground Glean has been standing on since before agents were the point.
For years, Glean has been viewed as an enterprise search engine. “We were inspired by…the old HP quote: ‘If my company only knew what my company knows, we’d all be ten times as productive,’” Tony Gentilcore, Glean’s co-founder and head of product engineering, said during the briefing. But now, it’s repositioning itself as the gateway for AI in the workplace. Its mission has transformed from helping workers know what their company knows instantly to providing that context to AI. “We found that the same thing we built for humans to do their work is the same thing that is necessary for agents and for models to do their work,” Gentilcore added.
But Glean is in a crowded space. On one side are the agentic knowledge-work tools—Anthropic’s Claude Cowork is among the better-known, alongside Microsoft’s Copilot Cowork, OpenAI’s ChatGPT Work, and agent offerings from Atlassian, Salesforce and ServiceNow. On the other are the enterprise search vendors Glean grew up with, including Slack, Lucidworks, Elastic, and Algolia. Yet Glean argued it has an edge, releasing internal benchmarks this past week that pit Glean Assistant against Claude Cowork running Sonnet 5 across more than 180 knowledge-work tasks.
Glean said its cost per task came in at $0.58 against Claude Cowork’s $2.98, an 81 percent savings on token costs, and that graders preferred its outputs 78 percent of the time. The test used synthetic queries run against Glean’s own production data, with responses scored side by side on a five-point scale. Glean ran with automatic model routing enabled, while Cowork was held to Sonnet 5 at high reasoning throughout, the setting Anthropic recommends for balancing quality and speed.
So how exactly is Glean improving context for the enterprise?
Meet Glean Tau
Like all SaaS applications, Glean has run entirely on the cloud—users access it from their internet browsers. This week, however, the company introduced a new way of interacting with its platform. Glean Tau is a desktop-first experience that runs on a worker’s machine, bringing together local files, code, and docs with Glean’s context graph. Not only can the AI write and run code locally, but it can also analyze files and perform end-to-end workflows on the device.
Dogan said that Glean Tau has already been rolled out internally, and usage surged across different types of tasks, all without costs spiraling out of control. He pointed out that it’s because Glean no longer only taps enterprise context in the cloud. Glean Tau combines that with what’s on a worker’s device, handling local operations without the usual limits of cloud-based assistants. It fuses all relevant information about work—both locally and in the cloud—when executing a task.
Glean Tau is Glean’s recognition that its context picture has a gap. Not everything employees produce lives in the cloud; plenty of material sits on laptops, in local folders, and in files that never sync anywhere an index can reach. This desktop experience is the company’s move to address that shortfall, pulling local material into the same catalog as everything else so the AI can see more of what’s actually happening inside an organization. And Glean isn’t alone in reaching for the edge: Microsoft has unveiled ways to bring model building, fine-tuning, and local deployment capabilities to Windows 11. And let’s not forget about Microsoft’s Copilot+ PCs built around on-device AI.
Officially, Glean said Glean Tau is expected to launch “soon,” but a spokesperson confirmed the timing will likely be within the next three months.
The ‘Proactive’ Push
Improved context can have ripple effects across enterprise AI. Agents will generate better responses, meaning far fewer repetitive tasks are needed, saving on token usage and also budgets. It also provides workers with peace of mind, knowing they no longer have to keep a watchful eye on agents. According to research from Glean’s Work AI Institute, human workers are spending an average of 6.4 hours each week—nearly an entire workday—on “botsitting,” checking to ensure AI is usable and has the correct context, debugging mistakes, rerunning prompts, and cleaning up “confident-but-wrong answers.”
Today’s AI is reactive, said Emrecan Dogan, Glean’s chief product officer. In other words, users have to remember to invoke it, spell out steps, build agents and skills, and remember which agent exists for what. “AI requires guidance,” he said, adding that it’s not a sustainable practice because humans have limited capacity for attention. As more models, features, tools, and capabilities are introduced, human attention will become a bottleneck in the enterprise. Workers will become overwhelmed by the situation, unable to decide which tool to use, when to use it, or what it’s good for. “We have to meet the user the moment they need it, and help them get the most value out of AI,” Dogan said.
That’s why Glean is introducing new tools built to support what it calls “proactive AI”—where human workers no longer have to supervise an agent’s work, and the agent doesn’t wait to be prompted. It may appear to be a novel concept, but other software vendors have pitched proactivity before now, including Salesforce and Asana. Despite that, Glean is positioning these updates as true digital coworkers that continuously watch how work is done and initiate tasks autonomously.
Proactive task management and coordination
Glean’s AI uses enterprise context to understand what needs attention and move work forward. It now offers three areas for improving personal productivity, starting with task management, where it takes the initiative to identify next steps and produce first drafts of deliverables. Next, it can help with email triage, surfacing what needs attention and generating a draft for a conversation, complete with context. Lastly, Glean functions as a meeting coach, helping workers prepare ahead of time, guide during the meeting, and sharpen every conversation afterward.
Independent agents
Also announced at Glean’s conference were its independent agents. Think of these as akin to Amazon Web Services’ Frontier Agents—they’re autonomous, persist between sessions, and are shared across employees. Each one also has its own identity and permissions along with a defined mandate—the scope of work it’s authorized to take on. Dogan said the independent agents also revise themselves on the go and have their own self-healing, self-improvement tools, meaning that feedback given to agents is used to improve that unit of work and improves the agent for subsequent runs.
According to Jain, there are two types of agents: those that run on behalf of a worker or ones that have their own identity. Those using the former should restrict agents to accessing or modifying only the enterprise data and information to which the human worker has rights. In other words, the agent impersonates the worker but operates within a single app, performing narrow, short-lived tasks.
As for the latter, Jain equated them to being “more like employees or colleagues of yours.” It runs continuously, has its own identity and role, and can initiate actions. This introduces a new class of problems, from role and permission design for the agents themselves, establishing their scope of responsibility, defining their behavior and collaboration norms, and setting up accountability and observability guardrails.
The proactivity and autonomy in these agents are traits they share with OpenClaw, and that’s not by accident. Gentilcore told The AI Economy that Glean started forking the OpenClaw project but stopped, deciding the approach wasn’t safe enough. “We were recreating a bunch of the stack without the guardrails and stuff like that,” he said. His team migrated everything onto Glean’s main agent harness instead. Gentilcore agreed the independent agents are essentially claw agents: “That’s what those are effectively, without the lobster analogy.”
The first three independent agents are now in beta: one for on-call management, another for voice-of-the-customer, and a sales agent.
Glean Transform
Glean also introduced Glean Transform this week. It’s a “living map” that draws on the platform’s enterprise graph to watch how work moves through an organization, across email, support tickets, documents, and meetings. Glean Transform continuously scans the organization for areas where AI could improve workflows. A report is generated, along with suggestions for agents or skills that can be deployed quickly with a single click. The tool also measures business impact.
The tool appears to be aimed at companies that treat AI adoption as a single project with a finish line—pick the workflows, deploy the agents, declare victory. But then, executives wonder why it isn’t generating value. Glean’s argument is that the map keeps changing. “Transformation is not a one-time initiative,” Dogan said. “It’s a constant evolution, opening up new possibilities to reimagine how work gets done in your environment, in your enterprise.”
Glean Transform has not been given a launch date, only that it’s listed as “coming soon.”
Team Chat
“We are moving AI from single player to multiplayer,” Dogan said on stage at Glean:GO, drawing one of the loudest rounds of applause of the keynote. The line came as the company introduced Team Chat, which lets employees work alongside each other, Glean’s Assistant, and agents in the same thread or shared artifact.
Slack and Microsoft have made similar moves, pushing workers to collaborate with their digital “peers” to build up the agentic enterprise, the Frontier Firm, or whatever marketing label is preferred. Glean’s Team Chat becomes an interactive canvas for work, enabling one-on-one and group conversations with humans and also including digital colleagues, such as the previously mentioned independent agents.
While there is utility in Team Chat, Glean could also use it to bolster its context graph, thereby keeping this flywheel going.
Team Chat is currently available in beta.
Other Announcements
Some of the other new tools in Glean’s announcement slate include interactive dashboards and an interface that combines structured data with unstructured business context, such as tickets, conversations, and documents. The dashboards refresh in real time as information changes, providing organizations with the context needed to make decisions. This is available today in beta.
Glean has also expanded its AI Gateway, which launched in July. Now, it supports more AI entry points, enforces restricted topics policies through Glean Protect, and has expanded its governed MCP access to include organizational skills and personal memory. This is also now in beta.
Lastly, Glean has announced context-aware AI threat detection, which will warn of any risky agentic behavior in context. Using its Enterprise Graph and context layer, the company claimed it can detect whether an agent is acting legitimately. It is listed as coming soon.
Pursuing a SaaS Revenue Model Over Token Consumption
This week’s news comes three months after Glean announced that its annual recurring revenue (ARR) surpassed $300 million. It’s an amount the company said has tripled since early 2025. Gentilcore said adoption is scaling inside those accounts. “Eighty-five percent of our customers deploy this wall-to-wall,” he said, adding that Glean sees 45 percent daily active users over monthly active users—meaning that of the employees using Glean in a given month, nearly half are on it on any given day.
“What’s our vision at this point? We started thinking that we’re going to empower every person, that we’re going to bring the context to every person,” Gentilcore said. “I think this has grown into something much bigger, which is the definitive context and intelligence layer for enterprise AI. This is about connecting, governing, protecting, building the right context one time for your entire enterprise, and then connecting that to everywhere that you use AI.”
Could Glean become the first of the AI infrastructure companies to top $1 billion? Jain dodged the question, saying only that there were internal plans he wasn’t at liberty to share and that “we keep getting surprised in what the market looks like.” Instead, he said Glean is pursuing a SaaS-like model rather than usage-heavy token revenue.
“We feel that today, AI is very expensive for customers, and as a company, we are trying to be very cost-conscious for our customers,” Jain said. “We think of ourselves as the company that’s going to help every enterprise manage their AI costs. From that perspective…think of us as having SaaS type of revenue model, not a revenue model where if a user uses the product a lot, we make a lot of revenue from that user.” The framing matters: Glean sees itself as a cost manager for enterprise AI rather than a beneficiary of runaway consumption.
Disclosure: I attended Glean:GO as a guest of the company, with travel and accommodation expenses covered. However, what I write reflects my own reporting and analysis. No one reviewed or approved this piece before publication.
