Danai Data Book a discovery call

Blog In the news

Google just gave Gemini its own work email. ChatGPT, a workplace AI agent or a governed agent team: 5 questions that tell you which your business needs

Herman Wa · · 3 min read

Illustration

The news. On 8 October 2026 Google began turning Gemini into an agent for businesses. It takes objectives rather than single instructions, plans the work, hands parts of it to sub-agents and connects to systems such as Google Workspace, Microsoft 365, Slack and Jira. It also gets its own Workspace account and email address, so colleagues can tag it, email it and add it to group chats, and its actions are logged against the agent rather than a person (TechCrunch).

Agents that act, not just answer, are arriving inside the tools many UK firms already pay for. That makes an old question urgent: when is a single assistant enough, and when do you need agents working under rules, with a record a person can check?

Not every problem needs a team of AI agents. Plenty of work is done well by a single assistant, and it would be wasteful to build more than you need. The trick is knowing where the line sits.

Here are five questions we ask at the start of every conversation.

1. Will anyone ask why the decision was made?

If the answer will be used once and forgotten, a chatbot is fine. If a board, an auditor, a client or a regulator might ask in six months why you chose this option, you need a record of what was considered, who disagreed and what evidence was used.

A chatbot leaves no record. A governed team ends every session with minutes, sources and a tamper-evident log.

2. Does the question have more than one legitimate angle?

"Summarise this contract" has one right answer. "Should we enter the German market this year?" has several defensible ones depending on whether you weight cash, risk, timing or customer demand most heavily.

When a question has competing angles, one voice will quietly pick one and present it as the answer. A team gives each angle its own specialist and makes them argue it out under rules.

3. Is a confident wrong answer expensive?

Language models sound equally sure whether they are right or wrong. For a draft email that does not matter. For a pricing change, an acquisition or a compliance decision it matters a great deal.

A governed team states confidence, requires sources and lets any agent object, as long as the objection comes with a fix. Weak points get found before you act on them, not after.

4. Does the work need different kinds of expertise?

If the task needs finance, regulation, operations and market knowledge at once, a single prompt tends to blur them together. Specialist agents, each with its own memory and its own sources, keep them separate and accountable.

5. Will you ask the same kind of question again?

A team becomes more valuable the more often it is used. Its agents remember previous sessions, the standing orders improve as you learn what you want, and on a retainer we track how often the team's forecasts actually came true.

If it is a one-off, the set-up is probably not worth it.

Scoring yourself

  • Mostly no: use a good assistant, with sensible rules about what data goes into it. You do not need us.
  • Two or three yes: a small team on one workflow is likely to pay off. That is what our Discovery Sprint is for.
  • Four or five yes: you have a decision process that is probably already costing you time in meetings, rework and second-guessing. A governed team will give you speed and a record.

A note on honesty

We would rather tell you a chatbot is enough than sell you a team you do not need. The first call is thirty minutes, and you will get a plain answer on which side of the line you are on.

Sources

Read next