Google's Gemini agent takes goals, not steps, across Workspace, Microsoft 365 and Slack. Here's the PM test before you let it near a project
Google announced a single 'universal' work agent on October 8 that runs in Workspace, Microsoft 365 and Slack. It is in private preview, but for a GC's project team the question to settle now is whose name an agent acts under.
Google's new Gemini agent is built to take a goal instead of a list of steps, and it works across Google Workspace, Microsoft 365 and Slack. It is in private preview, so no GC can buy it yet. The decision a project team can make now is whose authority an agent like this acts under.
What did Google announce?
On October 8, 2026, at its Gemini at Work event, Google Cloud announced what it calls a universal agent for work. Per 9to5Google and Quartz, the agent:
- takes a described outcome and decides how to get there, rather than following step-by-step prompts;
- runs in the Gemini Enterprise app, Workspace, Microsoft 365, Slack, the command line and mobile and desktop apps;
- can be called through an API, so other software can embed it without its own interface.
VentureBeat's coverage adds that these agents can run long tasks and get their own Gmail, Calendar and Drive storage. We could not confirm that detail beyond that one report, so treat it as unverified until Google documents it.
Availability is limited. The agent is in private preview, Google says wider availability is coming for select Workspace Business and Enterprise plans, and no pricing has been published.
Why should a GC's project team care?
Most project teams already live in a mix of these tools: Outlook and Teams at the GC, Gmail and Slack at a sub or design partner, Procore and Drive in between. An agent that works across all of them is aimed at exactly that mess.
The work it points at is the follow-up grind: chasing open RFIs, assembling a weekly look-ahead from scattered emails, compiling a closeout checklist from the spec and the submittal log. A PM who says "get me every open RFI older than 10 days and draft the nudge to each recipient" is giving it a goal, not steps.
That is also where the risk sits. A goal-based agent makes choices you did not spell out. On a project, those choices land in a contract record.
What should a PM decide before turning one on?
Three questions, in order:
| Question | Why it matters on a project |
|---|---|
| Whose name does the agent act under? | An RFI or notice sent from a PM's account reads as the PM's statement. Notice provisions in your contracts assume a person sent it. |
| What can it read? | Google's help page says Gemini in Workspace sees what the user can see, and admins can switch off its access to Drive and Gmail. A PM with access to bid files, pay apps and HR threads gives the agent the same reach. |
| Who approves outbound messages? | Anything to an owner, architect or sub should pass a human first, at least until you have a track record. |
An agent that can reach a platform you do not administer, such as a sub's Slack workspace, adds a fourth issue: you cannot set that company's policies.
Should a mid-size GC act now?
Do not rush a pilot you cannot yet join. Do use the preview window to write the rules, because those rules apply to every agent product, not just Google's.
- Pick one internal task. Compiling an open-items list from a closed project is a safe first test.
- Use a closed job. Compare the agent's output with what your team actually did.
- Check the log. Can you see what it opened and what it sent? If not, it is not ready for live work.
- Keep outbound human-reviewed. Drafts only, sent by a person.
Also note what is still unproven. A goal-based agent demo is not a project record, and nothing yet shows how it behaves on a messy RFI log with conflicting dates and half-named subs.
The takeaway
The tool will matter less than the policy. Before any agent gets near a live job, decide whose identity it uses, what it can open, and who approves what leaves the building. Write that down while the product is still in preview, and you will be ready to test whichever vendor's agent reaches you first.
- What is Google's Gemini agent?
- It is a work agent Google announced on October 8, 2026 that takes a goal, works out the steps, and returns a result. Google describes it as one agent available in the Gemini Enterprise app, Google Workspace and third-party services, including Microsoft 365 and Slack.
- Can a construction company use the Gemini agent today?
- Not broadly. Reports say it is in private preview, with wider availability planned for select Workspace Business and Enterprise plans. No pricing has been published.
- Can the Gemini agent see everything in my company's Drive and Gmail?
- Google's Workspace help page says Gemini has the same access to Workspace data as the user does, and that admins can turn off its access to Drive files and Gmail messages. That guidance covers Gemini in Workspace generally; confirm the specifics for the new agent once it reaches your plan.
- Should a project manager let an AI agent send RFIs or emails to subs?
- Not until the firm has decided whose name the message goes out under and who approves it. Start with internal drafting tasks on a closed project, and keep a human on anything that reaches an owner, architect or subcontractor.