The enterprise AI data finally exists. Glean is all over it.
a16z published hard numbers on where enterprise AI is actually landing. One of the three winning categories has a clear leader, it is in our portfolio, and the data raises a question every buyer should be asking about models and context.
The scoreboard finally exists
For two years, the enterprise AI conversation has run on anecdotes. Vendors claimed transformation. Skeptics pointed to an MIT study claiming 95 percent of generative AI pilots never reach production. Both sides argued from sentiment because nobody had hard numbers.
Andreessen Horowitz changed that this spring. Their report, "Where Enterprises are Actually Adopting AI", aggregates revenue and deployment data from the leading enterprise AI companies rather than surveying opinions. The findings cut sharply against the pilot-graveyard narrative.
More useful than the headline number is the shape of the adoption. It is not evenly distributed. Three use cases dominate: coding, support, and search. And within enterprise search, a16z names one startup as the vendor that has defined the category.
That company is Glean. We go to market with them, and this piece explains why the data validates a decision we made before the data existed.
The problem search actually solves
Every enterprise has the same pathology. Knowledge lives in hundreds of applications. The answer to a question exists somewhere in Slack, Confluence, Salesforce, Jira, email, or a shared drive, and the person who needs it spends part of every day hunting for it.
This was an annoyance in the SaaS era. In the AI era it became a structural blocker, for a specific mechanical reason: a model is only as useful as the context it can reach. An assistant that cannot see your company's actual knowledge produces generic answers. An assistant that can see everything, without respecting who is allowed to see what, is a data breach with a chat interface.
Why Glean's approach holds up
Glean's core asset is a permissions-aware knowledge graph. It connects to more than 250 enterprise applications through purpose-built connectors, indexes their content, maps how data, people, and processes relate, and enforces source-system permissions in real time on every retrieval. Everything else in the product sits on top of that graph: search, the assistant, and Glean Agents, a platform for building and governing AI agents already executing more than 100 million actions per year.
Glean's knowledge graph maps content, people, and activity to ground every answer. Image courtesy of Glean.
Two design choices matter to buyers evaluating this category. First, model neutrality: Glean supports more than 15 language models through a model hub, so when a better or cheaper model ships, you swap it in while your graph, permissions, and agents stay put. Second, governance as a first-class product: every agent action is authenticated, permissions-checked, and logged, extended by Glean Protect.
The honest question: why not just use Claude or ChatGPT directly?
This is the objection we hear most in 2026, and it deserves a straight answer. The frontier assistants are excellent, and they increasingly connect to enterprise tools on their own. If your teams already have Claude or ChatGPT Enterprise seats that can reach Drive, Slack, and email, what is Glean actually for?
The answer starts with a distinction the market keeps blurring.
A frontier assistant connects to the handful of applications each user authorizes, one connection at a time. That works for an individual. It does not work as an enterprise architecture. Glean maintains a persistent, indexed graph across hundreds of systems, enforces source permissions on every retrieval for every user, and logs every agent action centrally.
The second issue is coupling. If a single model vendor becomes your context layer, your enterprise knowledge, agents, and workflows are married to that vendor's roadmap and pricing. Glean inverts the relationship. It runs the frontier models underneath, Claude included, through a model hub spanning the major cloud AI services. When model leadership shifts, and it has shifted several times in eighteen months, you swap the engine without rebuilding the car.
Swap the model. Keep the context layer. Diagram: Ghost. Glean logo courtesy of Glean.
What Glean customers report
What the data says about how to buy AI
Step back from Glean and look at the a16z findings as a whole, because they contain a buying lesson. The AI deployments generating real revenue and real ROI share a profile: they own a specific workflow, prove value fast, and expand from there. The horizontal platforms promising to transform everything at once are conspicuously absent from the adoption leaderboard.
This is exactly the filter we apply at Ghost. We track hundreds of AI companies so that our customers do not have to.
We bring a company forward when three things are true at once: the category is real, the team is among the best executing on it, and the buyer still holds an edge. With Glean, the category is now validated by data and the leader is named. The remaining edge is architectural: making the context-layer decision deliberately, before defaulting into a bundled assistant or coupling your company's knowledge to a single model vendor.
What we are watching: how quickly the frontier labs close the connector gap, and whether graph depth, permission enforcement, and agent governance hold as the moat. Our read is that governed context across hundreds of systems is a product, not a feature.
Why we are bringing this to you now
Ghost does not exist to resell the tools you have already shortlisted. Our work sits upstream of that: identifying the companies defining a category and putting them in front of you while the timing still favors you.
Glean is past the point of being a secret, and that is precisely the point. The risk for buyers is no longer missing the company. It is making the wrong architectural decision around it: defaulting to whatever assistant came bundled with an existing license, or wiring company knowledge directly into one model vendor and discovering the coupling later. We have run these evaluations with our customers and can walk you through how Glean answers the hard questions against your specific environment and stack. When you are ready to move, we make the path from evaluation to deployment simple.