We put Gemini in front of thousands of people at Virgin Media O2 before chasing productivity theatre, with predictable spend, EU data trust, and agents in the tools they already open.

Most people at a big company don't live in an IDE.

They live in email, calendars, Teams channels, half-finished briefs, and "can you find that conversation from Tuesday?" That was the world we were actually trying to improve.

So at Virgin Media O2 the workforce story wasn't "pick a clever model and watch the token meter." It was Gemini Enterprise: rolling out AI at scale to 4,000+ people, with a known, predictable cost instead of per-token pricing.

That's the centre of gravity. Get the tool in front of thousands of people. Spend you can explain, without needing to publish the invoice.

(I can't publish the actual costs here. The point is the shape of the commercial model, not the number on the page.)

Putting capable AI in front of thousands of people means meeting them where they already work, not asking them to become engineers first.

The Gemini Enterprise footprint covered things people actually use:

• Image generation

• Chatbots

• Agents

• Including agents in Outlook and Teams

• And Gemini Notebook for working with their own material

4,000+ is my cleared claim for that footprint. I'm still not dressing it up as a productivity scoreboard. It's how widely we rolled it out.

One of the quieter wins in the workforce rollout was Gemini Notebook, the public name for what many people still know as NotebookLM.

What it does, in practice: you bring curated sources (docs, notes, the pile of PDFs on someone's desktop) and you get grounded research against that material, instead of a generic chat that doesn't know the brief. Alongside Gemini Enterprise, it also sits in the enterprise compliance story we'd already bought into, not a side-door consumer tool with a shrug about data.

For a company rolling AI out to thousands of non-engineers, that matters. Not everyone wants an IDE. Plenty of people want "here's the pack. Help me make sense of it." Gemini Notebook fitted that gap: practical, source-grounded work for the people who run the company day to day.

Per-token pricing is fine when you're experimenting alone at a laptop.

It's a different conversation when you're putting AI in front of 4,000+ people across a company. Suddenly finance wants a number they can budget. Leaders want a cost they can defend. Nobody wants a surprise bill because a busy week of agents went loud.

Gemini Enterprise gave us predictable spend for that scale, the kind of commercial shape you can take to finance without turning every prompt into a line item. I won't quote what that cost was. I will say the alternative (open-ended token burn at workforce scale) is how pilots die in procurement.

That sounds boring. It's also how you actually ship AI past the pilot without the pricing model becoming the blocker.

Cost isn't the only thing that kills a workforce AI rollout. Trust does.

People, rightly, worry about what happens to the data they paste into a prompt. Customer details. Internal threads. Calendar context. The quiet fear under every "just try ChatGPT for this" moment at work.

Gemini Enterprise ensured an EU data ringfence and ZDR (zero data retention). That combination removed a lot of the worry about what data went into AI, subject to GDPR.

I'm not going to pretend compliance language is exciting. I am going to say this is the human side of enterprise AI: if people don't trust where the data goes, they won't use the tool you just spent months rolling out. Predictable cost gets you past finance. Ringfence and ZDR get you past the knot in someone's stomach before they hit send.

The agents people feel are the ones in the tools they already open every morning.

In Outlook and Teams, we had agents that could:

• Find conversations across channels

• Search calendar availability

Not vague "manage your email" marketing. Find the thread. Find when people are free. The boring coordination tax that eats a day in fragments.

If you've ever hunted a decision across three Teams channels and a calendar nobody updated, you know why that matters more than another demo reel.

4,000+ is my cleared claim for how widely we rolled Gemini Enterprise out. I'm still not claiming every seat used every capability equally. Putting it in front of thousands isn't the same as everyone using it every day.

And I'm not publishing commercial figures. Only that predictable cost beat per-token surprise for a rollout this size, and that EU ringfence plus ZDR mattered for GDPR-era trust.

Rolling out AI at this scale with a known cost model, and data boundaries people could live with, is the start of the story, not the ending.

We didn't put Gemini in front of 4,000 people to turn them into engineers, or to discover our bill at the end of the month, or to shrug at where the data went.

We did it so the company could adopt AI at a size that mattered, with spend we could plan for, and with data boundaries people could trust, in the tools they already lived in, including a notebook where curated sources actually ground the answer.

Footnote: the Gemini Enterprise licence also included Antigravity (Google's VS Code-based AI IDE, primarily on Gemini models) as part of the deal, not a separate purchase. Useful context; not the workforce story.