31%. That's the share of US small businesses using AI who told the Federal Reserve it had brought in more sales. 71% said it made them more productive.
Three stories this week, one message. AI is joining businesses as a member of staff, with its own login. The regulator says responsibility for what it does stays with you. And the payoff goes to businesses that point it at a specific problem, not the ones that simply switch it on.
1. Google's Gemini agent gets its own Workspace account
Google CloudOn 8 October Google Cloud announced the Gemini agent, which it calls "a universal agent for work". It works inside Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar. It can also be reached from Microsoft 365 and Slack.
The part that matters is the "coworker agent". You describe a role and Gemini creates it. In Google's words, the agent "receives its own Workspace account, including an email address, calendar, Drive, and presence in your company directory." Staff add it to a Chat space or @mention it like anyone else. Google says it "sees only what you share with it", access follows the sharing your team already uses and every action goes into an audit trail under the agent's name.
Google's announcement doesn't give prices or say which Workspace plans get it. TechCrunch reports it's coming to businesses before consumers.
Why it matters: a coworker agent is a new starter that can act on whatever is shared with it. Most businesses set their sharing up years ago with people in mind, and plenty of files are open to the whole company. The audit trail helps, but only if someone reads it.
Practical takeaway:
- Treat an agent like a hire. Give it one named owner, a defined job and access to only what that job needs.
- Tidy company-wide sharing first. If Drive is full of files open to everyone, fix that before an agent joins "everyone".
- Put the audit trail in someone's diary. Monthly is enough to start. IT providers, your clients will want help with this, and our AI for MSPs page shows where we fit.
2. The ICO says AI agents are no excuse for poor compliance
On 8 October the UK's data protection regulator published a report saying ten of the biggest foundation model developers, the companies whose AI sits underneath most chatbots and assistants, have made or committed to data protection changes after its scrutiny, including Amazon, Apple, Google, Microsoft and OpenAI.
The same day it opened a six-week call for evidence on agentic AI, meaning AI that completes tasks and uses tools "often with limited human oversight". It wants to hear from the organisations that deploy agents as well as developers. It closes on 20 November and will feed a statutory code of practice on AI and automated decision-making. The ICO's Richard Nevinson put it plainly: "the fact AI agents act with autonomy is not an excuse for poor compliance."
Why it matters: read that line as aimed at you as much as at the AI companies. If you switch on an agent that handles customer or staff data, the regulator will look at what you did to control it. It applies directly if you have UK customers or staff. If you don't, expect your own regulator to ask the same questions.
Practical takeaway:
- List every agent that can reach personal data. Include the ones built into software you already pay for.
- Write down what each one may do without a person checking. That's the accountability question the ICO is asking.
- Respond if you're already using agents. It shapes the guidance you'll be measured against.
3. Fed data: 46% of US small businesses use AI, 31% of them see more sales
On 8 October economists at the Federal Reserve Bank of New York published an analysis of the Fed's Small Business Credit Survey, carried out between September and November 2025. 46% of US firms with at least one employee were using AI tools and another 15% planned to within a year. Of the firms using it, 71% said it made them more productive but only 31% reported higher sales. Writing and marketing was the most common use, at 83%. Just 7% had fully built AI into how the business runs.
AI users were much more optimistic about the year ahead. The share expecting revenue to rise beat the share expecting a fall by 48 points, against 21 for non-users. The link was strongest among firms in weaker financial shape and those struggling with technology such as their website or cybersecurity. Whether they actually get those results, the authors say, "is a subject for future research".
Why it matters: 71% more productive, 31% more sales. That's the gap between AI that saves someone time on emails and AI pointed at something that makes money. The Fed's own reading is that AI "may be helping some small businesses overcome specific operational or technical constraints". It isn't replacing staff either. 77% of AI users saw no change in labour costs.
Practical takeaway:
- Pick a constraint, not a tool. The slow quote, the unanswered enquiry, the website that doesn't convert.
- Measure before you start. If you can't say what it looked like before, you won't know whether AI helped. Our piece on why AI pilots fail covers this.
- Plan for changed jobs, not fewer people. Decide what the time AI saves will be spent on.
What this week means for your business
AI is getting a login and a regulator watching how you supervise it. The value doesn't come with the login. It comes from one clear job, the right access and someone checking the work.
That's the work we do. We build production AI around a specific problem with permissions and checks designed in, and we run it so they don't drift. The AI Readiness Diagnostic is free and no-obligation. You keep the statement of work, the wireframe and the ROI case whether or not we build anything.
Questions business owners are asking
Will Google's Gemini agent be able to see all my company's files?
Not automatically. Google says a coworker agent sees only what you share with it and access follows the sharing your team already uses. So files shared with the whole company, or with a Chat space the agent joins, are within reach. Review broad sharing before you create one.
Do data protection rules still apply when an AI agent does the work?
Yes. On 8 October the ICO said the fact AI agents act with autonomy is not an excuse for poor compliance. If an agent handles customer or staff personal data for your business, you're still answerable for how that data is protected and for what the agent is allowed to do without a person checking.
How many small businesses are using AI?
In the Federal Reserve's Small Business Credit Survey, carried out in late 2025, 46% of US firms with at least one employee were using AI tools and another 15% planned to within a year. Of the firms using AI, 71% reported higher productivity, 31% higher sales and 77% no change in labour costs.