Google has announced new agentic AI capabilities for Gemini, starting with business use cases. According to the company, Gemini is being developed beyond a conversational assistant into an AI agent that can plan and execute tasks, work across business applications and systems, and coordinate work through subagents.

Google also said the agent can use multiple AI models and can be assigned its own workplace identity, including an email address. While the company presentation points to broader automation and delegation inside office environments, the practical significance for businesses will depend on how these features are deployed, governed and integrated with existing systems.

What happened

Google says Gemini is gaining agentic capabilities designed for enterprise workflows rather than simple prompt-and-response interactions. Based on the announcement, the new approach includes several elements:

• Task planning and execution: Gemini is intended to carry out multi-step work rather than only answer questions.
• Work across business apps and systems: the agent is positioned as a tool that can interact with software used inside organisations.
• Delegation to subagents: Google says a main agent can assign parts of a task to specialised subagents.
• Use of multiple models: the system can reportedly combine different AI models for different parts of a workflow.
• Workplace identity: Google says the agent can have its own organisational identity, including an email address.

At this stage, the announcement is primarily about Google’s product direction and enterprise positioning. It does not by itself mean that all organisations can immediately or safely hand operational processes to autonomous AI systems without further controls.

Why it matters for European businesses

The announcement matters because it moves AI further into operational territory. Many businesses already use generative AI for drafting content, summarising documents or assisting employees. Agentic AI goes further by attempting to do work inside systems, not just generate output for a human to review.

For European businesses, that shift raises both productivity opportunities and governance questions. If an AI agent can access email, internal tools, documents, CRM platforms, support systems or e-commerce operations, it could reduce manual work across sales, marketing, administration and customer service. At the same time, it increases the importance of permissions, auditability, security controls and human oversight.

There are also regulatory and compliance implications. Businesses operating in Europe may need to assess how agentic AI fits into existing obligations around data protection, information security, record keeping and internal accountability. Where AI systems process personal data or interact with regulated workflows, legal and compliance teams are likely to require clear rules on access, approval steps and monitoring.

The idea of giving an AI agent its own workplace identity is particularly significant. A dedicated identity may help organisations manage access rights more cleanly than sharing human accounts, but it also creates new identity and access management questions. Companies will need to understand what the agent can access, what it can send, who is responsible for its actions, and how misuse or errors can be traced.

Who may be affected

• SMEs adopting AI tools: smaller businesses may see agentic AI as a way to automate repetitive office tasks without building custom software.
• Marketing and e-commerce teams: teams managing campaigns, product content, customer communications and reporting could be early users if the tools connect to common business platforms.
• IT and security teams: they will likely need to review identity, permissions, integration architecture and monitoring before rollout.
• Founders and operations leaders: decision-makers may face pressure to adopt AI automation while still managing risk and accountability.
• Regulated businesses: organisations in sectors with stricter data, audit or approval requirements may need more cautious deployment models.

What companies should consider

Businesses evaluating agentic AI should treat it differently from a standard chatbot deployment. The key question is not only whether the model produces useful answers, but whether it can be trusted to act appropriately inside real workflows.

• Map use cases carefully: start with low-risk, repetitive processes where the impact of mistakes is limited.
• Review access permissions: define exactly which systems, data and actions an AI agent can use, and apply least-privilege principles.
• Keep human checkpoints: approvals may still be necessary for customer communications, financial actions, publishing, or sensitive data handling.
• Assess identity controls: if agents receive their own workplace identities, ensure they are included in identity governance, logging and offboarding processes.
• Check data protection implications: if personal or confidential business data is involved, review data flows, retention and processor arrangements.
• Test auditability: companies should be able to see what the agent did, why it acted, and which systems it touched.
• Avoid broad rollout too early: agentic AI may be useful, but organisations should validate reliability and control before connecting it to critical operations.

For European businesses, Google’s announcement is less about a single new feature and more about the direction of workplace AI. As major vendors push from assistant tools toward autonomous task execution, the competitive question will increasingly be whether companies can adopt these systems productively and govern them responsibly.