Gemini Enterprise: The Coolest Things It Does, Why Companies Are Adopting It, and What the Legal Version Is

Short answer: Gemini Enterprise is Google's workplace AI — one place where employees search across the company's systems, chat with Gemini, and run AI agents that do multi-step work. It is not the Gemini app with a business login, and it's not Gemini in Gmail and Docs; it's a separate Google Cloud product that plugs into Microsoft 365, Salesforce, Jira, ServiceNow, and the rest of what a company already runs. It starts at $21 per seat per month with a 30-day free trial. There are now industry versions, led by Gemini Enterprise for Legal, which launched in preview on August 25 with four major law firms. Google's CEO says 'nearly 90% of the Fortune 100' use it. Below: what's worth trying, what the legal version does, why companies buy it, and what to check before you do.

A naming note, because Google has made this hard: the product was called Agentspace until October 2025. In April 2026 Vertex AI became the 'Gemini Enterprise Agent Platform,' so Google now calls the employee-facing product the 'Gemini Enterprise app.' This guide is about that app. Everything here comes from Google's product page, documentation, and release notes, its earnings remarks, and the press we cite; where something is Google's claim rather than an independent finding, we say so.

The coolest things you can do with it

• 1. Ask one question across every system you work in. This is the feature that sells it. Instead of searching SharePoint, then Jira, then Salesforce, then Drive, you ask once and get an answer with sources — limited to what you have permission to see. It connects to Microsoft 365, Google Workspace, HubSpot, Jira, Confluence, ServiceNow, and more.
• 2. Hand a research project to Deep Research and walk away. Google says the agent 'can work in the background for hours to deliver comprehensive, citation-backed reports' — drawing on the web and your internal documents together. Market sizing, competitor teardowns, a due-diligence first pass.
• 3. Build your own agent without code. Workflow Builder (called Agent Designer until this month) became generally available September 3. You describe a multi-step job, and the workflow can run on a schedule, on demand, or when someone @-mentions it. KPMG says its employees built nearly 700 no-code agents in the first weeks.
• 4. Put recurring work on a schedule. A Monday pipeline summary, a weekly contract-renewal check, a daily digest of support escalations — set once, delivered to an agent inbox.
• 5. Take action in Microsoft tools, not just read them. New actions for Outlook, Teams, OneDrive, and SharePoint arrived in preview on September 18, and outputs can be produced in Microsoft 365 or Google Workspace formats. For a mixed shop, that's the point.
• 6. Use it from Slack, or by voice. The Slack app handles channel mentions and multi-turn conversations as of September 10; voice input became generally available September 17.
• 7. Run science-grade agents. Co-Scientist, for generating and ranking research hypotheses, and AlphaEvolve, Google DeepMind's algorithm-discovery system, are both available — niche, and unlike anything in a rival's workplace product.
• 8. Shop an agent gallery. Pre-built agents from Google and partners including Accenture, Oracle, and ServiceNow, plus connections to outside tools through MCP and agent-to-agent standards, both generally available since August.

Two things that are no longer included, which older guides still list: the Idea Generation agent was removed in July 2026, and Gemini Code Assist was dropped from new Standard and Plus subscriptions on September 17 — Google points developers to Antigravity instead. And two things people assume are in the app but aren't documented there: computer-use agents and picking a non-Google model such as Claude. Both exist on Google's developer platform, not in the employee app.

What is Gemini Enterprise for Legal?

It's a version of Gemini Enterprise built for law firms and legal departments, announced August 25, 2026 alongside one for financial services; healthcare, life sciences, and retail are listed as coming. It's in preview, pricing isn't published, and you get it by contacting sales. The launch customers are legal teams at Cleary Gottlieb, Freshfields, Weil, and Williams & Connolly.

The pitch to lawyers is confidentiality first. Google's page: 'Client data, firm intellectual property, prompts, documents, and outputs remain strictly within your private cloud perimeter,' and 'your data is never used to train our foundation models.' On accuracy it's carefully worded: 'A hallucinated fact or an incorrect citation is a critical risk. We mitigate this by grounding AI responses in your verified internal documents and authorized legal databases.' Mitigate, not eliminate — every output still needs a lawyer's eyes.

Using Google's Gemini Enterprise, which can slot in seamlessly with other daily work tools, we can unlock greater efficiencies for our teams and help them deliver even higher quality work for our clients. — Jeff Karpf, Managing Partner, Cleary Gottlieb

What's missing from every law-firm statement: numbers. No firm has published hours saved, seats deployed, or matters handled, and no corporate legal department is named as a customer yet. It's also not an exclusive relationship — Freshfields is a named customer of Anthropic's legal product too.

How it compares with the other legal AI launches

What it costs

For comparison, Anthropic lists Claude Enterprise at $20 a seat per month plus usage at API rates, and Microsoft 365 Copilot — the product most companies are actually weighing this against — is $30 a user in our Copilot review, on top of a Microsoft 365 license. OpenAI doesn't publish a ChatGPT Enterprise price.

Why companies are adopting it

• It works with what they already have — including Microsoft. The buyer's fear with any Google product is lock-in to Workspace. Gemini Enterprise reads and now acts inside Microsoft 365, which is why firms that live in Outlook and SharePoint can still buy it.
• The data promise is in writing. For Business, Standard, and Plus: 'You own your data, not Google,' and 'your data — including prompts, outputs, and training — isn't used to train Google models or models for any other customer.' Google also says it 'never sells customer data' and 'doesn't use customer data for advertising.'
• Google covers the copyright risk. Gemini Enterprise is on Google's list of indemnified generative AI services, meaning Google takes on certain intellectual-property claims arising from its use. Read the service terms for the exclusions; we haven't.
• Compliance is built in rather than bolted on — data residency, customer-managed encryption keys, access transparency, and a prompt-injection screen Google calls Model Armor, in the Standard and Plus editions.
• Employees actually use it. KPMG reported in 2025 that 'nearly 90% of KPMG employees accessed the tool within two weeks,' and its executive Stephen Chase called it 'the fastest adopted technology our firm has had.' (KPMG is also a Google partner that sells these deployments, so weigh accordingly.)
• Momentum, by Google's account. On Alphabet's July earnings call, Sundar Pichai said nearly 90% of the Fortune 100 use Gemini Enterprise and named PepsiCo, Intel, HSBC, Bell Canada, and Macy's. Google Cloud revenue grew 82%. Google also says Gartner named it a Leader in its first Magic Quadrant for enterprise AI assistants.

Read that last figure carefully. Google has never published a seat count or revenue number for Gemini Enterprise, and 'Gemini Enterprise platform' in its earnings language covers the developer platform as well as the employee app — so '90% of the Fortune 100' means those companies use something under the name, not that their employees all have it. No independent survey yet compares its seats with Copilot's, ChatGPT Enterprise's, or Claude's. The spending data that exists points elsewhere: Ramp's September index has Anthropic and OpenAI at 44% and 40% of businesses and doesn't break out Google.

What to check before you buy

• A real vulnerability, since fixed. In December 2025, researchers at Noma Security disclosed 'GeminiJack,' a flaw that let a booby-trapped shared document, calendar invite, or email quietly pull corporate data out through Gemini Enterprise — no click required. Google confirmed the finding and said it was mitigated. Any tool that reads everything an employee can read inherits this class of risk; ask your vendor how it's tested.
• A model-safety incident, not a product breach. Google confirmed on September 18 that during a security evaluation in May, a Gemini model reached the systems of three real companies because the test environment had internet access it shouldn't have. That involved a model under test, not Gemini Enterprise or any customer's data — but it's part of the same month of disclosures we covered at OpenAI, and a fair question for any vendor.
• Preview means preview. Gemini Enterprise for Legal, and many of the connectors that make it useful, aren't generally available. Don't plan a rollout around a preview feature.
• The product keeps changing names and contents. Four renames and two removed features in under a year. Get the feature list you're paying for in the contract.
• Permissions are your job. Cross-system search shows people whatever they already have access to — including files that were shared too widely years ago. Clean up permissions before you turn on search, not after.
• Search quality varies by connector. Google's own documentation notes that when it searches a system live rather than indexing it, 'search quality might be lower.'

How to try it

• Small team: start the Business edition's 30-day free trial from the product page — Google says no IT setup is needed.
• Larger company or regulated industry: Standard and Plus also have a 30-day trial, but purchase goes through Google Cloud sales.
• Law firm or legal department: Gemini Enterprise for Legal is preview-only, by request, at cloud.google.com/ai/legal.
• First week: connect two systems, not ten; try the cross-system question, one Deep Research report, and one scheduled workflow. That's enough to know whether it earns a seat.

Our read

Gemini Enterprise's best argument isn't the model — it's the plumbing. The thing employees notice is that one question reaches every system they use, including Microsoft's, and that agents can then act there. That's a harder thing to build than a chatbot and Google has built it, with the data and indemnity promises buyers need in writing. The legal version is the same bet pointed at the most document-heavy, confidentiality-obsessed profession there is, and four elite firms signing on at launch is a real signal — though not yet a result. What's missing is proof: no seat counts, no law-firm outcomes, no independent head-to-head. If your company runs on Google Cloud or wants one assistant across a mixed Microsoft-and-Google estate, the trial is worth a month. If you're choosing on model quality alone, that race is being run elsewhere this month. Our consumer-side take is in the Gemini review.