Google has just taken a step worth looking at calmly: it has unveiled a Gemini capable of executing tasks, not just answering questions, and it has decided to launch it first in the corporate world before bringing it to the general public. For anyone running a private company, the news isn’t about Google: it’s about the fact that AI agents in the enterprise have stopped being a laboratory experiment and have become a management decision. And management decisions are made with judgement, data and a clear use case, not with headlines.
What Google announced (and why it starts with businesses)
As reported by TechCrunch (Sarah Perez, 8 October 2026), at a Google Cloud event the company presented a unified agent within Gemini that can receive objectives, not just instructions: it plans the work, uses tools and connects to the organisation’s internal systems to complete it.
Some elements of the announcement that deserve attention from a business standpoint:
- Connection with real work systems: Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git and analytical databases, among others, as well as servers that follow the MCP protocol (Model Context Protocol).
- Traceability: a "task tray" where you can see what the agent is doing, how it distributes the work across subagents and what steps it has taken, with an audit log attributed to the agent and not to a person.
- Model choice: by default the system chooses the model, but the user can pick another one, including third-party models such as those from Anthropic.
- Spending control: Google announced flexible spending options, intelligent routing and real-time consumption limits.
- Release order: businesses first, consumers later, in order to solve the "hard problems" of security, scale and performance beforehand, in Sundar Pichai’s words.
That last point is the most revealing one. When a vendor decides to launch its most ambitious technology in corporate environments, it is saying that the value lies in real work: case files, orders, incidents, reconciliations, reports. Exactly where an SME loses hours every week.
From assistant to agent: the difference that really matters
A conversational assistant drafts an email if you ask it to. An agent receives an objective ("prepare the proposal for this client using their historical data and leave it pending approval") and chains together several steps using tools and systems until it is completed.
The practical difference is who does the intermediate work. In the assistant model, the person remains the "glue" between systems: copying, pasting, checking the ERP, searching for the document, updating the spreadsheet. In the agent model, that glue is automated and the person moves on to supervising and deciding. It is a change of role, not just of tool.
What this demands of your own house
Here comes the part that no product announcement tells you: an agent is only as good as the systems and data it connects to. If customer data lives in three different places, if the item master has duplicates or if permissions were never reviewed, the agent doesn’t fix the mess: it amplifies it and, what’s more, it does so faster. That is why at Tisa we have insisted on the same order for years: first tidy processes and data, then automation, and only then AI.
How to ground AI agents in the enterprise without losing your bearings
You don’t need to reinvent the company to get started. You need to choose the first use case well. These are the criteria we apply in our projects:
- It should be repetitive and high-volume. If it happens twice a year, it isn’t a good candidate.
- It should have a measurable outcome. Hours saved, errors avoided, response times. If it can’t be measured, it can’t be defended before management.
- The data should exist and be reliable. An agent that consults incomplete information produces answers that look like the truth: that is what we call hallucination, and in an administrative process it comes at a high price.
- There should be a human control point. Approvals, amount limits, reversible actions. Traceability is not an extra: it is the condition for being able to use it.
- The cost should be predictable. AI consumption is billed by usage. Without caps and without monitoring, the cheap pilot turns into an uncomfortable invoice.
Realistic examples by sector
With the experience we have in retail, distribution, construction, the food industry, hospitality, sports centres and fleet management, the natural candidates tend to be the same ones:
- Distribution: preparing responses to requests for quotation by cross-referencing order history, stock and customer terms, leaving the proposal ready for commercial review.
- Hospitality: classifying and routing incoming requests from agencies and direct customers, with the booking recorded by a person or by a controlled workflow.
- Construction: gathering scattered site documentation and putting together the draft progress report.
- Fleets: cross-referencing maintenance reports, workshop invoices and cost allocation per vehicle to detect deviations.
- Services and sports centres: following up on unpaid invoices and renewal notices with personalised communication.
None of these examples requires replacing the ERP. It requires connecting properly what you already have.
Where your ERP fits into all this
If your company works with Microsoft Dynamics 365 Business Central, the foundation is already there: finance, purchasing, inventory, sales or project data is in a single, structured system. From there, Power Platform (Power Automate, Power Apps, Power BI) allows you to build automations and dashboards, and custom development covers what the standard doesn’t reach.
The AI market moves fast and no vendor has the definitive answer. Our recommendation is pragmatic: keep your management core solid, choose the AI layer according to business criteria —fit with your processes, total cost, integration and support— and avoid getting locked into a decision you can’t reverse in two years.
Let’s talk about your first use case
At Tisa we have been helping private companies turn technology into concrete results since 1987: fewer errors, more visibility and processes that scale without driving up IT costs. If you are considering where to start with AI applied to your business, we can help you identify the data you already have, establish reasonable governance and define a first measurable use case.
Write to us at info@grupotisa.com or call us on (+34) 971 305 885 and we’ll assess it together, with no obligation. More information at grupotisa.com.
Source of the announcement discussed: TechCrunch, "Google brings agentic AI to Gemini, starting with businesses", 8 October 2026.
