Arvind Jain is the co-founder and chief executive of Glean, a private enterprise-software company that combines workplace search, an AI assistant, and agent tooling. The durable part of his story is not an alleged rivalry with another AI executive. It is the attempt to make company knowledge usable without discarding the permissions and context attached to the underlying systems.

As of September 13, 2026, Glean remains privately held. Its latest publicly announced financing in the sources reviewed here was a June 2025 Series F at a stated $7.2 billion valuation. Both the valuation and the operating metrics around that round came from Glean, not audited public filings. They establish what the company and its investors announced, but they do not independently establish revenue quality, retention, profitability, or current market value.

The verified career sequence

Glean’s official biography of Jain says that he worked for more than a decade across Google Search, Maps, and YouTube before co-founding Rubrik and then Glean. Rubrik’s public-company records provide outside corroboration for the middle step: its 2025 annual report filed with the SEC identifies Arvind Jain among the company’s co-founders.

Those facts matter because enterprise search is not merely a language-model problem. It requires indexing information from many systems, preserving access controls, resolving identities, ranking results, and keeping the index current as documents and permissions change. Jain’s search background is relevant to that engineering problem. It does not, by itself, prove that Glean solves it better than every competitor.

In an October 2025 Goldman Sachs interview published in December, Jain discussed Glean’s founding and its expansion from search into assistants and agents. This is a public, attributable account from the founder. It is useful evidence of his strategy and chronology, not independent validation of product performance.

What Glean says the product does

Glean describes its core search product as a single interface across workplace applications. Its enterprise-search page says that connectors ingest content and that results respect the source system’s permissions. The company’s administration documentation likewise explains that users should see only content they can access in the originating application.

That permission-aware claim is central. A generic retrieval system can produce an impressive answer while leaking a document that the requester was never entitled to read. An enterprise deployment has to carry identity and authorization through ingestion, retrieval, answer generation, citations, and any action an agent takes. It also needs a deletion path when a source item is removed or a user’s role changes.

Glean markets more than search. The company positions its assistant as a synthesis layer and its agent products as a way to perform multi-step work. The categories should not be collapsed:

  1. Search retrieves and ranks source objects.
  2. An assistant composes a response from retrieved context.
  3. An agent may call tools, change records, or initiate a workflow.

Each step adds failure modes. Retrieval can miss the right document. Generation can state something unsupported. An agent can take the wrong action even when its summary is accurate. A buyer therefore needs separate evidence for retrieval quality, answer grounding, and action safety.

Financing and scale, with attribution

Glean announced a $150 million Series F in June 2025 at a $7.2 billion valuation. In the same release, the company said it had surpassed $100 million in annual recurring revenue and was processing more than 100 million agent actions annually.

Those are company-disclosed figures. Glean is not required to publish the financial statements, cohort data, or metric definitions that a listed company would normally disclose. “Annual recurring revenue” may also differ from recognized revenue, while an “agent action” can range from a low-risk lookup to a consequential write operation. The figures show the narrative behind the financing round; they should not be treated as an audit.

The round still supplies one verifiable signal: outside investors were willing to price and fund the company on those terms at that date. It does not prove that the valuation held after the transaction or that the company has a defensible margin structure. Large-model inference, indexing, connector maintenance, and customer-specific implementation can all affect unit economics.

Why enterprise context matters

Foundation-model vendors can offer general reasoning and generation. Glean’s proposed advantage is different: it tries to map an organization’s people, content, activity, and permissions across systems. If that layer is accurate and current, it can make many models more useful. If it is incomplete, an eloquent answer may still be wrong.

This makes connectors and governance more important than a chatbot demo. Glean advertises a broad connector catalog, but connector count is a vendor claim and an incomplete buying metric. Buyers should ask which APIs are used, which objects and permission types are supported, how quickly changes propagate, and what happens when a source API is unavailable. A connector that indexes only documents but not comments, row-level permissions, or deletion events may create a misleading sense of coverage.

The same caution applies to security statements. Glean’s security page is a useful inventory of the controls the company says it offers. It is not a substitute for reviewing current audit reports, contract terms, subprocessors, data residency, incident history, and the exact configuration proposed for one deployment.

A practical evaluation plan

A serious trial should begin with a defined corpus and a held-out question set, not a collection of polished demonstrations. The following measures reveal different kinds of failure:

  • Retrieval recall: does the correct source appear for representative questions?
  • Ranking precision: how often are the first results relevant and current?
  • Permission fidelity: can users retrieve anything they cannot open at the source?
  • Citation support: does each material answer claim follow from the cited passage?
  • Freshness: how long do edits, revocations, and deletions take to appear?
  • Action controls: which agent operations require confirmation, and can they be reversed?
  • Operational cost: what are the implementation, connector, model, and support costs at expected usage?

Tests should include contractors, recently transferred employees, terminated accounts, duplicate identities, restricted legal or HR material, and documents with similar titles but different dates. These cases are less attractive than a demo query, but they expose the failures that create real risk.

For agents, the evaluation must also record tool inputs, retrieved evidence, model output, authorization decisions, and downstream changes. An administrator needs to reconstruct who requested an action, what the system knew, which policy applied, and what changed. A natural-language explanation without an execution log is not an audit trail.

Limits of the public record

Public sources do not reveal Glean’s current revenue, retention, gross margin, customer concentration, or the accuracy of its search across a representative set of enterprises. They also do not show whether customers broadly allow write-capable agents into high-risk workflows. Those are diligence questions, not gaps that should be filled with estimates.

A December 2024 Axios interview with Jain is useful because the founder himself acknowledged that enterprise AI can be error-prone and hard to implement. That public caution is more informative than presenting Glean as either a certain winner or a simple wrapper around a model. The product’s value depends on the quality of its customer-specific deployment.

The defensible conclusion is narrow. Jain’s record connects search engineering, enterprise infrastructure, and the current agent push. Glean has raised substantial capital and reports meaningful commercial scale. Whether it becomes a durable enterprise layer will be decided by permission fidelity, grounded answer quality, operational adoption, and economics that public announcements do not disclose.

Source and correction note

This revision uses public company materials, product documentation, an SEC filing, and named interviews available through September 13, 2026. Company metrics and product capabilities are explicitly attributed to Glean. The earlier version relied on an unsupported account of a private investor warning, treated private-company metrics as established facts, and used competitive language that the cited record did not support. Those passages have been removed rather than softened.