Box vs Glean (2026)

Box agents act on enterprise content, and they write as well as read. They search, analyze and create files in Box Notes, Word, PDF, Excel and PowerPoint, with availability set across AI Home, Files and Hubs, and permissions gate every action. Box serves a Box MCP server so external platforms reach Box content, publishes an AI Studio API and a developer platform, and custom agents can query images and spreadsheets. No integrations reach systems outside Box, such as a CRM or a ticketing system, so what the agents act on is content and the platform around it.

Sourcebox.com/agents, /ai/ai-studio (index copies); docs.box.com/en/box-ai/creating-and-configuring-agents; readread 2026-09-17

Agents act in outside systems as well as reading from them, through Connectors & Actions, and Glean enforces source-system permissions on every read and write. The 275+ app connectors each have their own page under glean.com/connectors, spanning documents, tickets, chat and code, and the same connectors feed Glean's index. The Indexing SDK lets teams add custom, permission-aware data alongside Glean's existing connectors, and custom actions are available through the APIs.

Sourceglean.com/ai-agents connector index and glean.com/platform/apiread 2026-09-12

Custom instructions and modes drive multi step work. Instructions can enforce step by step workflows, such as first analyze, then propose, then act, direct the agent to favor certain tools when several are available, and specify structured output. Pro Mode supports advanced agent behaviors such as recursive reasoning and enhanced document or video analysis, and Expanded Mode lets queries exceed standard limits.

The Box Agent completes multi step tasks across content, locating files, analyzing large document sets and generating new files, iterating in agentic loops. There is no workflow surface with branches and retries, no versioning is described, and whether Box Automate workflows can invoke an agent step is not stated.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents; box.com/agents (index copy); readread 2026-09-17

Glean's agentic engine lets agents reason through tasks, plan next steps and take action using enterprise context, in multi step and multi agent runs. Agent orchestration, which Glean describes as automating work across systems, routes tasks between agents and connects to external systems so automations run when and where they are needed, inside a governed runtime. An agent builder supports reasoning-based agents and complex workflows, and a deployment surface lets teams roll out, share and certify agents.

Sourceglean.com/ai-agentsread 2026-09-12

Every way into a Box agent is a person asking. Agents appear where admins publish them, in AI Home, Files and Hubs, and answer when a user asks, with suggested prompts to start an interaction. Box Automate and Relay trigger content workflows on content events, but no Box Automate workflow is described invoking an agent step, and no schedule, event or telemetry trigger starts an agent.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents; box.com/agents (index copy); readread 2026-09-17

Key events start agent runs, and agent orchestration exists to trigger agents from key events, route tasks between agents and connect to external systems so automations run when and where they are needed. People reach Glean from its own web app, from 275+ connected applications, and through Glean MCP inside Claude Code, Codex, Gemini, Cursor and Copilot. Proactive AI, which anticipates next steps like meeting prep or first drafts before anyone asks, is marked coming soon, while the event triggers carry no such marker.

Sourceglean.com/ai-agents, against glean.com/platform/enterprise-contextread 2026-09-12

Agents are grounded in specific, permission aware content. An agent's Knowledge setting binds it to specific files, or Hubs for reference. Knowledge sources are referenced with @mentions in custom instructions, with explicit priority rules when they conflict, such as a single source of truth, or ProductSpecs highest, ReleaseNotes second and FAQ lowest, which keeps agents grounded in authoritative sources.

Box AI pulls information only from the document loaded in preview and tells the user when a question falls outside it. AI Studio grounds agents in specific, admin-approved files or Hubs to reduce hallucinations, responses cite their sources, and custom agents can query images and spreadsheets. Hubs act as bindable, maintained knowledge scopes over the customer's content, and how Hub contents refresh is not described.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents; box.com/ai/ai-studio, /agents (index copies); readread 2026-09-17

A persistent index and graph sit over the customer's corpus. Glean has 275+ connectors across apps, documents, tickets, chat and code and 1000+ signals in the Enterprise Graph mapping people, content and relationships, on the principle of indexing data once and using it everywhere. The Enterprise Graph captures not just content but the relationships between people, data and activity across the organization, and it persists between runs and stays queryable. Retrieval respects permissions, enforcing source system permissions on every read, so the graph carries access state as well as content.

Sourceglean.com/platform/enterprise-contextread 2026-09-12

Whether Box agents keep memory is not settled. The agent configuration covers details, custom instructions, knowledge, suggested prompts, availability, advanced model settings, Pro Mode and Expanded Mode, and none of these is a memory, session state or learned context surface.

The AI Studio and Box Agent material describes grounding and modes and names no memory. Agents ground in enterprise content on each run. Box's AI documentation runs to several hundred pages, so a memory feature could sit somewhere not covered here. Nothing published describes Box Agent session history or a memory store with a stated scope and lifetime.

Sourcedocs.box.com/llms.txt, /en/box-ai/creating-and-configuring-agents; tried; docs.box.com/_llms/en/ai.mdread 2026-09-17

Enterprise Memory, which learns continuously, is one of four Enterprise Context components, and what Glean publishes under that heading is the Enterprise Graph, a knowledge structure. A per user layer sits beside it. Personal Graph learns each person's goals, commitments, priorities, writing style and ways of working from their edits, replies, messages and other work signals, and Glean uses it to tailor answers and write in that person's voice.

Glean does not describe a lifetime, expiry or purge path for either, or a way for a person or an admin to read or edit what was learned. Skill reflection, launched in October 2026, updates a skill from how it is used only after a person confirms the change. Proactive AI, which builds a live view of teams and projects and anticipates next steps, is marked coming soon.

Sourceglean.com/platform/enterprise-context, glean.com/enterprise-context/personal-graph and glean.com/blog/skill-reflection-2026read 2026-10-24

Admins set guardrails on each action, and agents cannot override approval workflows. AI Studio governance applies policy-based guardrails on actions, validates inputs for prompt injection and keeps humans accountable for consequential decisions, and admins can restrict actions such as read, write, move, delete, share and email and switch capabilities such as search and content creation on or off.

Custom instructions cannot override human approval workflows, break safety policies enforced in code, access restricted data or classifications, or force tool execution when permissions block them, and instructions can tell the agent to prioritize verification before responding or to ask when unsure. Where an approval step is configured or how it is routed is not described.

Sourcebox.com/ai/ai-studio (index copy); docs.box.com/en/box-ai/creating-and-configuring-agents; readread 2026-09-17

Glean Protect checks agent actions before they run. Action Alignment checks agent actions against the user's intent and authority before execution, and write actions are checked before they execute to keep agents within their intended scope, so Glean, not the downstream system, stops a write that fails the check.

Topic Restrictions check requests against acceptable use and industry-specific policies before the model responds. Agent governance policies and different policies for select agents or users are marked coming soon, and the control on agentic browser use is in beta, while the checks above carry no such marker. Glean does not describe an approval queue that a person reviews item by item.

Sourceglean.com/platform/securityread 2026-09-12

Box holds a FedRAMP Class D certification, previously described as FedRAMP High authorized, meets FedRAMP, FIPS 140-2, ISO 27001 and PCI DSS Level 1 standards, maintains certified compliance with HIPAA/HITECH and ISO 27018, and reports SOC 1/2/3. Its information security management system is based on ISO 27001 and NIST 800-53, with SOC reports available under NDA.

Agents answer only from content the user may see and respect existing permissions, classification and retention by default. AI Studio admins restrict agent actions, including read, write, move, delete, share and email, scope knowledge to named files or Hubs, control where an agent appears, and manage which users may use it. Box states customer data is not used to train third party models.

Sourcebox.com/trust, /fedramp, /security/information-governance (index copies); docs.box.com/en/box-ai/creating-and-configuring-agents; readread 2026-09-17

Glean holds SOC 2 Type II, ISO 27001, ISO 42001, HIPAA, TX-RAMP Level 2 and GDPR, with a trust center at trust.glean.com.

Identity & Access verifies who is making a request through the customer's SSO provider before AI runs, covering both user and agent identity, and Data Permissions enforces source-system permissions on every read and write, so an agent does not automatically access everything the user can. An always on audit trail records what happened.

Four items are marked coming soon or beta, agent governance policies, per agent and per user policies, Enterprise Graph enriched threat detection, and security scans of memory.

Sourceglean.com/platform/securityread 2026-09-12

Admins see AI usage by agent and by user, but not a step by step record of each agent run. Admin AI Insights adds an AI usage dashboard in the Admin Console with AI queries over time, AI usage by product, and top agents by usage, both chargeable and non chargeable.

Admins can download detailed user activity reports that break down unit spend by individual user or specific agent, and the Admin Console carries extensive audit trails for more than 300 actions. AI Studio governance includes maintaining audit trails and logs, and the AI Unit insight card runs about two hours behind. There is no per run trace of an agent's steps, tool calls or retrieved content. Citations show the sources behind an answer.

Sourceblog.box.com/introducing-admin-ai-insights, box.com/security/it-admin-controls (index copies); docs.box.com/en/box-ai/understanding-ai-units-in-box; readread 2026-09-17

Every control point is logged to an end-to-end audit trail as it happens, with events available to stream into a SIEM.

Six control points are logged, identity and access, data permissions, sensitive content, topic restrictions, runtime threat protection and action alignment, so the trail shows who asked, under what identity, what was permitted, what was screened and what the agent was allowed to do.

It is Glean's own record of what its agents did, separate from the logs the customer's Slack, Jira or Salesforce already keep. Agent observability, a separate surface, reports adoption, error rates, upvotes, downvotes and ROI as operational metrics.

Sourceglean.com/platform/security and glean.com/ai-agentsread 2026-09-12

Customers choose the region where their content is stored. Box Zones expanded to 10 regions, adding Switzerland, Singapore and Israel. Box is managed cloud with that regional choice, there is no on premises or customer cloud deployment, and the other seven regions are not named.

Sourcebox.com/security/it-admin-controls and Box Zones announcement (index copies); readread 2026-09-17

Customers can host Glean in their own cloud tenant, and Glean says SafetyCulture chose it in part for its security-first approach to enterprise data and the ability to host Glean within its own cloud tenant. TX-RAMP Level 2 points to a deployment posture US state governments can procure, and a trust center sits at trust.glean.com. Glean publishes no deployment guide for the cloud tenant option, and no single tenant, customer managed, on premises or air gapped deployment.

Sourceglean.com/platform/securityread 2026-09-12

Admins get a builder for their own agents. AI Studio starts from New Agent, where the admin supplies a name, custom instructions, knowledge bindings, up to four suggested prompts, availability and model settings. Box lists example use cases a customer could build, such as extracting NDAs and liabilities from data room documents or pulling KYC details from onboarding forms. There is no prebuilt agent set, template gallery or catalog.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents; box.com/ai/ai-studio (index copy); readread 2026-09-17

The Agent Library lets teams discover trusted, reusable agents, so every team can build, deploy and discover new agents on one platform, and a deployment surface lets them roll out, share and certify agents. The library holds independent agents built for different jobs that customers can adopt from the catalog. Glean does not publish a count of prebuilt agents.

Sourceglean.com/ai-agents and its Agent Library sectionread 2026-09-12

Admins choose the model for each agent. In an agent's Advanced settings, they select Auto or a model from the list of other providers available in the organization. Pro Mode is on by default, so the agent uses higher thinking levels and may access more capable and newer models, and Box's AI units guidance recommends Standard models or Auto for routine work and Premium models for complex reasoning. Models come from leading providers such as Anthropic, Google and OpenAI. Bringing your own model or customer hosted inference is not offered, and choice is from Box's provider list.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents, /en/box-ai/understanding-ai-units-in-box; box.com/ai/ai-studio (index copy); readread 2026-09-17

Glean Intelligence optimizes AI for price-performance, matching each task to the right model across 40+ open and frontier models, with Auto routing. Customers and admins get a Model Hub for access to the latest models, flexible model choice as part of the Open Platform, an AI Gateway that governs every AI interaction and which models may be used, and Usage Controls for AI usage and spend. Multi cloud support and the Bedrock, Vertex and Azure OpenAI hosting relationships are infrastructure, not separate model providers. Glean does not show how a model is selected for each step.

Sourceglean.com/platformread 2026-09-12

Outside agents can reach Box content, and Box publishes developer APIs. Box-native agents run in Box Agent and Box AI Studio, and third-party platforms, including Claude, Microsoft Copilot, GitHub Copilot, Glean and custom LLMs, access Box content through the Box MCP server. Box also publishes a developer platform (developer.box.com) with APIs and an AI Studio API, and agents generate files in Box Notes, Word, PDF, Excel and PowerPoint for use elsewhere. Box agents consuming external MCP tools is not described.

Sourcebox.com/ai/ai-studio, /agents (index copies); readread 2026-09-17

Outside callers reach Glean through its Search API, Chat API and Agents API, an Indexing SDK that lets teams add custom, permission-aware data alongside Glean's existing connectors, and a Web SDK to embed Glean UI and context-aware chat or agent experiences.

The Agents API follows the LangChain Agent Protocol, with client libraries in TypeScript, Go, Java and Python, and an ungated developer portal sits at developers.glean.com.

Glean MCP feeds governed enterprise context into Claude Code, Codex, Gemini, Cursor and Copilot, letting those tools drive Glean, while the 275+ connectors run the other way, letting Glean drive other systems.

Sourceglean.com/platform/api and glean.com/blog/go-agents-launchread 2026-09-12

Agents are tested before release, and evaluation tests gate where they can be published. AI Studio includes a Playground where admins validate behavior, refine instructions and confirm grounding before broader deployment, using custom suggested prompts against real Box content and swapping models in the same session.

Availability expands as an agent passes evaluation tests for increasingly complex knowledge configurations, from single files to multiple files and Hubs, so passing those tests is the release path. The evaluation tests are not described in detail, and no scored regression suite is published.

Sourcedocs.box.com/en/box-ai/creating-and-configuring-agents; box.com/ai/ai-studio (index copy); readread 2026-09-17

Builders can test agents before they publish them, and Glean does not describe how a test is scored. The agent builder versions every change automatically, keeps drafts and offers rollback, and lets builders test outputs and adjust instructions.

Headless agent building, generally available since October 2026, lets developers define an agent in files from Claude Code, Codex or Cursor, run it with representative inputs, "run evaluations" and refine it before publishing, with changes reviewed and rolled back through Git. Debug and trace views became generally available in May 2026. Glean does not describe what an evaluation measures, a pass criterion, scored test cases, a quality gate or a controlled experiment. Agent observability reports adoption, error rates, upvotes, downvotes and ROI as live operational metrics.

Sourceglean.com/ai-agents/agent-builder, glean.com/blog/glean-headless-agent-builder and glean.com/press/glean-introduces-the-enterprise-agent-development-lifecycle-codifying-how-enterprises-build-govern-and-measure-ai-agentsread 2026-10-24

Box agents work on content, not screens. They search permitted files, analyze document sets and generate files inside Box, and external systems reach Box through its MCP server. There is no browser, desktop or screen control, since the agent's tools are content operations, so breakage when an interface changes is not a concern.

Sourcebox.com/agents, /ai/ai-studio (index copies); docs.box.com/en/box-ai/creating-and-configuring-agents; readread 2026-09-17

Agents reach other systems through 275+ API connectors and custom actions, and Glean offers no hosted browser, desktop session or remote or local computer control. Glean Protect lists keeping agentic browser use in check as a guardrail in beta. Nothing shows Glean's own agent driving a browser.

Sourceglean.com/platform/security and glean.com/platformread 2026-09-12