The Hire Nobody Interviewed
Google held its Gemini at Work 2026 event today and introduced the Gemini agent, which it describes as a single, universal agent for work. One agent that answers questions, handles knowledge work, makes images and media, and writes and runs code, all from the same prompt box. Google’s framing is that you give it objectives instead of instructions. We’re a 100-person creative and media agency. What happens when the new hire is software?
I run technology for a 100-person creative and media agency. My reaction, somewhere around my third trip to the Bevi, was that this one is aiming at something bigger. It isn’t trying to be a better app. It’s trying to be a coworker.
One Window, One Memory
Anyone who has worked at an agency knows we have plenty of tools. What we’re short on is fewer tabs. Briefs live in Docs, decks in Slides, timelines in Asana or Sheets, and feedback hides in email threads that started three revisions ago. Every AI tool we’ve tried has been one more tab, and every one of them needed the backstory again.
The memory piece got my attention. Google describes four kinds: what it’s working on now, what it has learned about the company, how a job gets done, and everything it has done before. In agency terms, that’s the gap between a freelancer you re-brief every Monday and a coordinator who remembers the client hates orange.
Give It a Role, Then a Roster
You can give Gemini a role: personal assistant, specialist, teammate, or something closer to a chief of staff. The coworker version gets its own Workspace account, with an email address, calendar, Drive, and a spot in the company directory. Tag it in a doc comment, and it replies under its own name in version history.
I can picture it already. A traffic agent that chases status. A research agent that pulls category trends before a pitch. A production agent that builds the first rough deck. For bigger jobs it can spin up a temporary roster of sub-agents that work in parallel for hours or days. Basically a pitch team that doesn’t need pizza.
The wake up routine is what I’d use first. Workspace intelligence surfaces the email that actually matters and offers to take delegatable tasks off your plate in one click. My current morning routine involves the Bevi and staring at my inbox like it owes me money. This might be an upgrade.
The Boring Part That Matters Most
Here’s where it gets real for an agency. We handle NDAs, embargoed launches, and media budgets that are not ours to spend creatively. The governance story matters more to me than the demo.
The agent inherits the controls we already have: identity, policy, budget, and audit. Each agent gets its own identity and only sees what you share with it. Every action lands in an audit trail with the agent’s name on it. You can set a hard spend cap per project, and the agent pauses when it hits it, which means we could finally charge AI costs back to the right team. Policy is written once and applied to every agent, so a rule like “every agent-drafted email gets human eyes before it goes to a client” stops being a sticky note on someone’s monitor.
Not Married to One Model
Gemini also isn’t tied to a single model. It routes each job to whatever fits, and today that includes Claude. For an agency like ours, that’s quietly important. The best model changes every few months, and our context and client knowledge shouldn’t have to pack up and move every time it does.
So, Are We Hiring?
It’s in private preview, with wider availability promised soon for select Workspace Business and Enterprise plans, and no pricing yet. So nobody’s getting a badge photo this week.
But the question has changed. It used to be which AI tool we should buy. Now it’s which agent we trust with a client’s full context, and that’s a leadership question more than a software one. The agent will remember every client preference we’ve ever written down. It still won’t know why the client hates orange. Some institutional knowledge stays stubbornly human.
