The Next Generation of Gemini: One Agent, Model Flexibility, and Less Lock-In

The next generation of business AI has arrived. On October 8, 2026, Google announced the Gemini agent, a single, universal agent for work that can plan tasks, use tools and skills, connect to enterprise systems, and execute complex, multi-step objectives.

As your trusted technology partner, Kartaca is excited to share how this new Gemini experience can streamline operations, protect enterprise data, and help organizations manage AI costs while keeping their options open as models evolve.

For CIOs and CTOs, the announcement signals a broader shift in enterprise AI. Beyond helping employees write emails or summarize documents, AI agents can connect business knowledge to operational systems, coordinate work across applications, and perform defined tasks under enterprise controls.

Making this shift requires an architecture that connects agents to reliable business data, defines their authority, and measures their outcomes.

What Is the Unified Gemini Experience?

Google is bringing its enterprise AI capabilities together around a single, universal Gemini agent.

Employees can delegate objectives through a unified interface rather than treating knowledge work, content creation, coding, and data analysis as disconnected tasks. The agent can plan the work, select the tools and skills it needs, connect to enterprise systems, and deliver completed work through the applications teams already use.

It can operate across Google Workspace, Microsoft 365, Slack, developer environments, and connected enterprise applications. It can maintain context across sessions and continue working on tasks that take hours or days, even after an employee closes their laptop.

Work can be scheduled, assigned for later execution, or triggered by events. Agents can also operate without a dedicated user interface, integrating into existing applications and workflows.

Gemini supports multi-agent orchestration, coordinating specialized agents across different parts of a larger assignment. Organizations can also create dedicated coworker agents with their own identities, email addresses, and defined roles.

Three capabilities underpin this experience: connected tools, reusable skills, and business context.

Tools connect agents to existing applications, databases, APIs, and services. Through the Model Context Protocol (MCP), organizations can connect agents to compatible systems and expose selected capabilities through standardized interfaces.

Skills describe how specific work should be completed. Teams can build reusable workflows and instructions for reviewing documents, preparing financial reports, handling customer requests, or investigating operational incidents. Shared skill and tool registries help teams reuse these capabilities across the organization.

Context and memory help agents understand organizational terminology, historical information, task requirements, and established ways of working. Gemini’s approach includes session memory for ongoing tasks, semantic memory for organizational knowledge, procedural memory for workflows, and episodic memory of previous activities.

Together, these capabilities bring AI closer to operational execution. The business value depends on how well agents connect to the systems, policies, and processes that govern the work.

Note: Availability of individual capabilities may depend on product rollout, access, and configuration.

Why It Matters for Your Business

Google’s announcement points to three important developments in enterprise AI: a persistent agent experience, flexibility in model selection, and stronger controls over deployed agents.

1. Collaborate With One Always-On Agent

Your teams can interact with a single agent through chat, web, mobile, and supported work applications. The agent can continue executing multi-step tasks and complex coding work in the cloud after an employee closes their laptop.

It can also function as a digital coworker with a persistent role, its own identity, and access to explicitly authorized information.

For example, a finance team could delegate the preparation of a monthly performance pack. The agent could gather approved data, prepare an analysis, create charts, draft an executive summary, and assemble a presentation. Employees could then review the results instead of coordinating each step manually.

The same approach applies to customer service, IT operations, sales enablement, project coordination, and software development.

Gemini can also work directly across Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, allowing employees to delegate tasks where they originate.

How Gemini Can Change the Way Teams Work in Google Workspace

Google introduces three ways to work with Gemini on Workspace: personal assistance, proactive delegation, and coworker agents that operate as team members.

Personal assistance across applications

Gemini can use context from calendars, Chat spaces, email threads, and documents to coordinate work without requiring employees to provide every detail.

For example, an employee could ask Gemini to schedule a meeting with the usual regional event leads. The agent can identify relevant people based on previous conversations and team membership, check their calendars, and coordinate a suitable time with both internal and external participants.

It can also coordinate tasks across applications. An employee could ask it to research market trends, build a financial model in Sheets, and create a presentation in Slides. Gemini can maintain the context of the overall objective rather than treating each step as a separate request.

Proactive delegation

Workspace Intelligence can identify tasks in emails and other work contexts that may be suitable for delegation to Gemini.

For example, when a manager requests a project update as a slide deck, Gemini can recognize the task and offer to delegate it to the agent. It can also prioritize messages in an inbox by surfacing important ones and explaining their relevance.

These capabilities can reduce the administrative work associated with routine requests, follow-ups, and reporting, allowing employees to focus on reviewing results and making decisions.

Coworker agents embedded in team workflows

Organizations can create coworker agents for defined roles, such as an events coordinator or project administrator. These agents can receive their own Google Workspace accounts, including email addresses, calendars, Drive access, and company directory entries.

Teams can interact with them through Google Chat spaces, mention them in conversations, or involve them in document collaboration.

For example, a marketing team could ask an events coordinator agent to prepare a launch-readiness document and post it to the team’s Chat space. A team member could also mention the agent in a Google Docs comment to request an edit. The agent can suggest changes and respond in the comment thread under its own identity, with its activity reflected in the document’s version history. These agents operate under their own identities and can access only the context made available to them. Access follows the sharing and membership controls already used by the team.

For IT leaders, each coworker agent needs a clear owner, a defined role, appropriate access, and a process for reviewing its work.

Google Workspace is evolving into an environment where AI can coordinate work across documents, communications, calendars, and team processes. Agents can participate in existing workflows rather than becoming another interface employees must manage.

2. Choose the Best Model and Keep Your Options Open

Google is separating the Gemini agent experience from the underlying model choice. The agent can select different models for different tasks, using Google’s Gemini family and Anthropic’s Claude models, with additional private and open models planned for the future.

This allows organizations to match models to workloads instead of routing every request through the most powerful and expensive option.

A frontier model may suit complex planning, nuanced reasoning, or difficult technical problems. Faster, lower-cost models may handle routine classification, information extraction, or repetitive sub-tasks.

For a multi-step workflow, the agent can coordinate these capabilities according to the task’s requirements. Smart Routing and multi-model orchestration are intended to balance performance, quality, and cost.

For example, an agent preparing a management report might use a more capable model to interpret a complex business question, a faster model to classify and extract information from supporting documents, and another suitable model to draft the final narrative.

Model flexibility also helps organizations adapt as the AI landscape changes. Google states that business context, skills, and data can remain in place as the underlying model evolves.

This can reduce dependence on a single model family, provided organizations also consider integration design, data portability, access controls, model-specific capabilities, and provider-specific dependencies.

For CIOs, the goal is to avoid unnecessary coupling between business processes and a particular model, making it easier to evaluate alternatives as accuracy, performance, cost, and business requirements change.

3. Control Your Agent Fleet With Clear Security and Spending Limits

As agents gain the ability to execute tasks, IT teams need to manage them as operational identities with specific permissions and measurable responsibilities.

Google’s announcement describes several capabilities intended to support enterprise governance:

• Agent identity and authorization: Give each agent its own identity and fine-grained, role-based permissions. Attribute actions to the agent that performed them.
• Auditing and observability: Maintain audit trails of agent actions and monitor activity to identify anomalies.
• Agent Sandbox: Isolate agent execution within a dedicated environment with its own network boundary.
• Agent Gateway: Enforce organizational policies on network traffic to, from, and between agents.
• Real-time spending limits: Set project-level AI spending caps. When a cap is reached, the project’s agent work pauses until an administrator chooses to resume it.

These controls help organizations define what agents can access, which actions they’re authorized to take, how their activity is monitored, and where execution must stop.

An agent that analyzes financial transactions shouldn’t automatically have permission to execute payments. An agent that diagnoses a production issue shouldn’t necessarily be allowed to change system settings. Preparing an action and executing it require different levels of authority.

Agent autonomy should be designed around business risk, not granted simply because a technical integration makes an action possible.

Data and Analytics: Turn Business Questions Into Operational Insights

Google is introducing data and analytics capabilities designed to help technical teams and business users turn natural-language questions into actionable insights.

For data engineers and ML teams, Gemini can help generate PySpark code, provide editable notebooks for testing, train models, and troubleshoot or fix pipeline issues. These capabilities can accelerate repetitive development tasks, giving teams more time to focus on data architecture, model quality, and business problems.

For business users, Google describes capabilities that use BigQuery and Knowledge Catalog to generate and save operational reporting queries. Once a query has been saved, teams can run reports on demand without incurring token costs, with consistent, verified answers through established reporting logic. This can make routine reporting more accessible, allowing employees to ask questions in natural language.

Knowledge Catalog connects agents to organizational schemas, business definitions, and established rules. This matters because business questions often depend on definitions that aren’t self-evident. For example, departments may interpret net margin, active customer, or qualified opportunity differently.

Enterprise AI requires trusted business context. Organizations should establish authoritative sources, define metrics consistently, and validate generated answers. This is particularly important when agents work across multiple data platforms or retrieve information from documents, PDFs, and internal knowledge repositories.

Making Unstructured and Distributed Data More Accessible

Google also introduces capabilities intended to help organizations use data that may be difficult for AI systems to interpret or access consistently.

Smart Storage is designed to enrich unstructured objects, including PDFs, images, scans, and audio recordings, with contextual information while keeping that information alongside the original objects. This can make previously difficult-to-search content more useful to AI workflows while retaining the existing security posture of the stored data.

For example, an organization could make scanned documents and archived files more accessible to workflows that retrieve relevant evidence, interpret content, or answer questions about organizational records.

Borderless Lakehouse addresses the challenge of data distributed across cloud platforms, data systems, and business applications. Google’s approach allows Gemini to query Amazon S3 and Azure Data Lake without variable egress fees, access Salesforce Data 360, SAP, and Workday without copying their data, and federate Apache Iceberg tables across Databricks Unity, Snowflake Horizon, and AWS Glue.

This can help organizations use AI without first consolidating all datasets into a single location. Keeping data in its existing environment may reduce unnecessary duplication, although integration, permissions, performance, and governance still need to be evaluated for each use case.

The platform can also use registered datasets to determine which data is required for an analysis, generate SQL, Spark, or Python code, execute the analysis on Managed Spark Service with Lightning Engine or in BigQuery, and produce charts or dashboards in the tools analysts already use.

Together, these capabilities connect natural-language requests to structured data, unstructured content, and analytical workflows.

Before scaling AI-driven analytics, technology teams should evaluate data quality, access permissions, metadata, lineage, and underlying business logic. Making more data accessible is valuable only when agents can interpret it correctly and use it within appropriate boundaries.

From AI Pilots to Production: A Practical Roadmap

Moving from experimentation to production requires a deliberate approach to workflow selection, integration, security, and measurement. We recommend evaluating agent opportunities across five dimensions.

1. Identify High-Value Workflows

Start with a clearly defined business problem. Look for repetitive, information-intensive processes involving data gathering, established rules, or coordination between teams.

Potential starting points include preparing management reports, processing invoice exceptions, assembling customer-service cases, reviewing documents, coordinating meetings, preparing project updates, and investigating routine IT incidents.

Define a baseline before implementation. Cycle time, manual effort, error rates, exception volumes, and operating costs help establish whether the agent delivers measurable value.

2. Connect the Agent to Enterprise Systems

Map the applications, databases, APIs, and information sources the workflow depends on.

Identify authoritative sources and determine whether the required data is accessible, up to date, and appropriately governed.

Where existing applications lack suitable integration interfaces, custom connectors and services built around standards such as MCP can expose selected capabilities to agents.

The goal is to connect agents to existing systems of record while preserving the business rules and controls those systems enforce.

3. Establish Reliable Grounding

Agents need to retrieve and interpret the right information in the correct business context.

Depending on the use case, this may involve indexed grounding over approved knowledge sources, federated access to information that remains in its original systems, or a combination of approaches.

Define authoritative sources, establish metadata and permissions, and test how the agent handles incomplete, conflicting, or outdated information.

For analytics workflows, agree on business definitions and validation rules before allowing agents to generate reports that influence decisions.

4. Design Security and Autonomy Together

Assign each agent a defined identity, an accountable owner, and the minimum permissions necessary for its role.

Determine which actions it can perform independently, which require approval, and which are prohibited. Configure audit trails, policy enforcement, monitoring, and escalation paths accordingly.

Testing should verify that agents respect access restrictions, avoid unauthorized actions, handle integration failures safely, and escalate ambiguous cases, in addition to evaluating response quality.

5. Measure Performance and Manage Costs

Evaluate the complete operating model, including model usage, connected services, integration development, monitoring, maintenance, and human review.

Use the appropriate model for each task, establish spending limits, and monitor the cost of completed business outcomes rather than focusing exclusively on individual model calls.

Track adoption and reliability alongside financial metrics. A workflow that performs well in a demonstration may behave differently when it encounters real data, exceptions, and changing business conditions.

Use these results to refine the workflow, improve grounding, adjust permissions, and determine where additional autonomy is justified.

How Kartaca Can Help You Capitalize on Gemini

Transforming business workflows into efficient, agent-first processes requires the right strategy, grounding, and security guardrails. Our specialized team can help you:

• Identify high-value workflows. Assess business processes, prioritize use cases, define success metrics, and identify opportunities where agents can deliver measurable operational improvements.
• Build custom MCP connectors. Connect legacy databases, applications, and systems of record through secure integrations and open standards, including the Model Context Protocol.
• Establish enterprise grounding. Connect agents to approved organizational knowledge using appropriate indexing, retrieval, and federated-access patterns. Define trusted sources, business terminology, and validation rules to improve consistency.
• Design secure agent architectures. Establish identities, permissions, auditability, execution boundaries, and approval mechanisms that align with your business requirements.
• Optimize the operating model. Evaluate model selection, integration costs, monitoring requirements, and spending controls to support sustainable deployment.
• Enable AI-powered collaboration. Identify opportunities for Gemini in Google Workspace, from cross-application task delegation to team-based coworker agents, while maintaining appropriate access controls and clear responsibilities.
• Lead change management. Help employees delegate work to agents, review outputs, manage exceptions, and collaborate effectively with digital coworkers.

As a Google Cloud Premier Partner, Kartaca combines cloud engineering, data expertise, and practical business understanding to help organizations move from isolated AI experiments to production-ready solutions.

Ready to identify your next AI opportunity? Contact us today to discuss how Gemini’s unified agent experience can help your organization improve productivity, connect business systems, and build secure, measurable AI workflows.

Let’s build an AI foundation that can evolve as models, tools, and business needs change, while keeping data, governance, and business outcomes at the center.