Google Agentspace – Enabling AI Agents in Enterprise Workflows

What is Google Agentspace? Google Agentspace is a new platform aimed at creating an “agent-driven enterprise,” where AI agents help employees and teams with their work. Launched in late 2024, Agentspace “puts the latest Google foundation models, powerful agents, and actionable enterprise knowledge in the hands of employees” [1]. In essence, it combines enterprise search with generative AI and automation: employees (and AI agents themselves) can find information across organizational data silos, understand and synthesize it using Gemini’s intelligence, and then act on it via AI agents [1]. This addresses the key needs for scaling AI in business: an AI-ready information ecosystem that breaks down data silos, easy tools to create/adopt agents, and enterprise-grade security and compliance [2]. At Next ’25, Google announced a slate of new Agentspace features and integrations designed to make AI agents easier to discover, build, and integrate into real workflows [3].

Key Features of Agentspace

Unified Enterprise Search (with AI): At the core of Agentspace is a unified search capability that spans all of a company’s information – from documents and emails to intranet pages and databases – with the power of Google’s search and AI. Employees can “find any piece of information within the organization...with the ease and power of Google-quality search,” even if that info is in different formats (text, images, audio, video) or stored in various systems [4]. Agentspace uses multimodal, AI-powered search to understand queries and content in context. It can draw from Google Workspace data, Microsoft 365 documents, Jira tickets, Salesforce records, ServiceNow knowledge bases, or even web content – effectively breaking down silos and retrieving relevant data wherever it resides [5]. Behind the scenes, it builds an enterprise knowledge graph linking people, documents, data sources, and permissions, turning “disjointed content into actionable knowledge” for the user [6].
Chrome Integration for Search: To make this AI search seamlessly accessible, Agentspace is now integrated directly into Chrome Enterprise. Starting in preview, employees can use the familiar Chrome browser address bar (omnibox) or a dedicated search box to query Agentspace and get unified results [7]. The search happens securely within the enterprise context, respecting access controls. This means users can search their company’s knowledge base as easily as a Google web search, without leaving their normal workflow (e.g. while browsing or using web apps) [7]. By bringing AI-powered search to where employees already work, Agentspace aims to boost productivity – answers and insights are available on-the-fly, without switching context or tools.
Agent Gallery (Discovery of Agents): Agentspace introduces an Agent Gallery that serves as a one-stop catalog of available AI agents in the enterprise. This feature (generally available via allowlist) provides employees with “a single view of available agents across the enterprise, including those from Google, internal teams, and partners”, making agents easy to discover and use [8]. In one interface, a user can see all the AI assistants and automations that they have access to – whether it’s a Google-provided agent (for example, a finance-report generator), a custom agent built by their own company’s developers, or a third-party agent purchased from a partner. Administrators can even browse pre-built agents in the Google Cloud Marketplace’s new AI Agent Marketplace and enable them in the gallery [8]. (Google has over 50 partners, including Accenture, Deloitte, Salesforce, SAP, etc., contributing to this growing agent ecosystem [9] [10].) This ecosystem approach means businesses can quickly adopt proven agent solutions or share internally-developed agents, rather than starting from scratch for every use case.
Agent Designer (No-Code Agent Creation): To democratize the creation of custom agents, Google launched Agent Designer, a no-code visual interface for building AI agents within Agentspace [11]. Currently in preview (allowlist), Agent Designer lets non-technical users create their own agents by simply defining the agent’s purpose, connecting it to relevant data sources, and writing natural language prompts/instructions [11]. Under the hood, these agents can leverage the same powerful models (like Gemini) and can integrate with enterprise apps or APIs, but the employee building the agent does not need to write any code. This feature empowers business domain experts (who may not be software developers) to automate tasks or create assistants tailored to their team’s needs [11]. For example, a marketing analyst could create an agent to automatically pull campaign data from spreadsheets and generate a summary report each week, without having to ask IT for a custom tool. Importantly, Agent Designer is complementary to Google’s more advanced developer tools: it works in tandem with Vertex AI’s Agent Builder (a developer-first tool for coding agent logic). In fact, agents built in Vertex AI by developers can be published to Agentspace for broad employee use [12]. This ensures that whether an agent is crafted via code or via the no-code interface, it can live in one platform (Agentspace) for distribution and management.
Built-in Expert Agents: Google is also seeding Agentspace with ready-made expert agents that tackle common complex tasks. At launch, they announced two new Google-built agents joining the roster (alongside an existing “NotebookLM for Enterprise” agent):
- Deep Research Agent: An AI agent that autonomously researches complex topics on an employee’s behalf. It can pull information from both internal data and external sources (like the web or databases) and then synthesize it into a comprehensive, easy-to-read report – all in response to a single prompt [13]. This agent is meant to save employees hours of digging through documents or websites; for example, an analyst could ask the Deep Research agent to investigate a competitor’s market strategy using internal reports and news articles, and get back a concise summary with references. (Deep Research is generally available via allowlist) [13].
- Idea Generation Agent: This agent helps spur innovation by brainstorming and evaluating ideas autonomously. With a prompt from the user (e.g. “develop new product feature ideas for XYZ”), the Idea Generation agent will generate a range of novel ideas in any given domain. Uniquely, it then uses a “competitive system inspired by the scientific method” to test and refine those ideas, effectively having multiple candidate solutions compete and evaluating which ideas stand out [13]. This could involve the agent performing mini-experiments or analyses on the ideas it generated and selecting the best option(s). It’s a way to inject creative problem-solving into workflows and ensure the ideas are well-vetted. (Idea Generation is in preview via allowlist) [13].
- NotebookLM for Enterprise (previously available) is another example of an expert agent – it is designed to digest and answer questions about a user’s documents (like an AI research assistant for your notes). With these and future agents, Google provides some out-of-the-box “power tools” within Agentspace, so companies can immediately benefit from advanced AI use cases (research, brainstorming, etc.) without having to build everything themselves.
Multi-Agent Communication (Agent2Agent Protocol): Agentspace is built to handle multi-agent systems, where different AI agents can collaborate or pass tasks among themselves. To enable this, Google introduced the Agent2Agent (A2A) protocol, an open standard for communication between agents, even if they are built on different frameworks or by different vendors [14]. Google is the first major cloud provider to push this kind of interoperability, seeing it as critical for the future of AI workflows [14]. In practical terms, A2A would allow, say, a finance-report agent to invoke a forecasting agent and then an emailing agent in sequence, even if each was created by a different party or uses a different underlying model. By giving agents a common “language” to talk to each other and exchange information, Agentspace can support complex, multi-step processes that no single agent could handle alone. This also future-proofs an enterprise’s AI architecture: one can integrate best-of-breed agents from various sources and have them work together seamlessly within Agentspace.
Enterprise-Grade Security & Compliance: Recognizing that enterprises need to safeguard data when using AI, Agentspace was built with rigorous security and compliance features. It runs on the secure Google Cloud infrastructure that’s trusted by billions of users [15], and includes fine-grained controls so that AI agents only access what they’re permitted to. Key security features include:
- Data Scanning & Protection: Admins can configure Agentspace to scan enterprise systems for sensitive data (such as personally identifiable information, personal health information, or other confidential fields) [16]. If certain content is flagged as sensitive, the platform can block AI agents (and even the search function) from accessing or revealing that information [16]. This prevents accidental leakage of protected data through AI responses.
- Access Controls: Agentspace supports role-based access control (RBAC), ensuring that employees (and the agents acting on their behalf) can only retrieve data they are authorized to see [16]. An HR agent, for instance, cannot pull finance documents unless the user running it has permission. All data is also encrypted, and customers can manage their own encryption keys (CMEK) for added security [16]. Data residency requirements are respected as well, meaning companies can control where their data is stored and processed to meet regulatory compliance [16].
- Secure Deployment Environment: Agentspace essentially provides a fully managed, secure environment to deploy and run AI agents across the company. Users don’t have to worry about provisioning servers or exposing APIs insecurely – the platform handles the backend, auditing, and reliability. (This is sometimes referred to as an “Agent Engine” that runs the agents at scale for the enterprise in a governed way [17] [18].) All interactions can be monitored and logged, aiding in compliance and troubleshooting.
AI Agent Marketplace (Ecosystem): To further extend what Agentspace can do, Google has expanded its Marketplace to include third-party AI agents. The AI Agent Marketplace is a section of Google Cloud Marketplace where trusted partners offer pre-built agent solutions that tackle specific tasks or industries [10]. Enterprises can browse this marketplace and purchase/deploy partner-built agents (for example, an industry-specific customer service bot, or a Salesforce-integrated sales assistant). Once acquired, these can be enabled in the Agentspace Agent Gallery by IT admins [8] [10], making them instantly available to employees. Google indicated that partners like Accenture, Box, Deloitte, Salesforce, SAP, and others are contributing agents and integrations, reflecting a broad multi-partner support for the Agentspace ecosystem [9]. This marketplace approach accelerates adoption—companies can leverage expert-developed agents and not reinvent the wheel for common use cases.

Supported Integrations and AI Workflow Enablement

One of the biggest strengths of Agentspace is how it integrates with existing business tools and workflows, effectively bringing AI assistance into the daily routine of employees:

Integration with Enterprise Apps & Data: Agentspace can connect to a wide array of enterprise data sources and applications. Out-of-the-box connectors (and the search index) cover Google Workspace (Drive, Docs, Gmail, etc.), Microsoft 365 (SharePoint, Office files, emails), ticketing and CRM systems like Jira, Salesforce, ServiceNow, and more [5]. Web content and intranet websites can also be tapped. This means an agent can be configured to pull information or perform actions in these systems. For example, an agent could read an inventory database to answer a supply chain question, or create a Jira ticket as part of an automated workflow. Because Agentspace builds a unified knowledge graph, it understands the relationships between data from these sources (who owns which documents, which CRM records relate to a client, etc.) [6]. All of this happens with respect to access permissions and data governance, as described above. Essentially, Agentspace acts as the glue between AI and the enterprise’s application landscape, enabling AI-driven workflows that span multiple systems.
Developer Tools (ADK): For developers who want to create more complex or specialized agents beyond the no-code Agent Designer, Google also announced an Agent Development Kit (ADK) [19][20]. The ADK is an open-source Python framework that provides the building blocks to define an agent’s logic, tools, and memory in code [21]. It handles the heavy lifting of orchestrating reasoning steps, managing agent state, and integrating with LLMs, so developers can focus on the unique logic or toolset of their agent [21]. The ADK is model-agnostic and designed to work across LLM ecosystems [22], giving flexibility to use different models (though naturally it’s tuned for Gemini’s capabilities). This toolkit complements Agentspace: a custom agent developed with ADK could be deployed on Agentspace for enterprise use, and the A2A protocol would allow it to interoperate with other agents. ADK essentially allows businesses to build bespoke AI agents (e.g., an agent that knows how to parse proprietary data formats or interface with an internal API) while leveraging Google’s underlying agent management infrastructure.
Live Collaboration & Real-Time Agents: With the introduction of the Live API for Gemini models [23], Agentspace can support agents that operate in real-time scenarios. For instance, an agent could join a live video conference (using streaming audio input) to provide on-the-fly insights or minutes, or a factory monitoring agent could watch CCTV feed for safety issues. The ability to handle streaming input/output and long-running sessions [24] means Agentspace is not limited to static Q&A bots – it can host interactive, continuous agents that truly collaborate with humans in their workflow (listening, responding, and taking actions in an ongoing loop). This is a forward-looking integration that opens up use cases in customer support (e.g., real-time call center assistance), operations monitoring, and anywhere else “live” AI assistance is valuable.

Real-World Use Cases and Impact

Google highlighted numerous real-world use cases to demonstrate how Agentspace and Gemini 2.5 are being used by enterprises to drive tangible results:

Internal Knowledge Assistants: Several companies are piloting Agentspace as an internal knowledge hub. For example, Gordon Food Service reported that Agentspace has transformed how employees access enterprise knowledge by allowing them to search across all internal data (Google Workspace, ServiceNow, etc.) in one place [25]. Queries are “grounded in [their] data” and yield results that would previously have required searching multiple systems [26]. This unified, AI-powered search is leading to better decision-making and less time wasted hunting for information [27]. Similarly, KPMG is implementing Agentspace to enhance workplace operations, enabling its staff to quickly retrieve information and insights across the firm’s vast knowledge stores [28]. By empowering every employee with an AI assistant that knows the company’s data, these organizations aim to boost productivity and make informed decisions faster.
Customer Service and Support Agents: Agentspace also facilitates the creation of customer-facing agents that can augment or automate support. A notable example is The Home Depot’s “Magic Apron” agent, which the retailer built to provide 24/7 expert guidance to customers [29]. Magic Apron can give detailed how-to instructions for home improvement projects, recommend the right products, and even summarize relevant customer reviews – functioning like a virtual store associate or handyman that’s always available [29]. This agent makes DIY easier for customers and likely reduces the load on human support lines. It demonstrates how businesses can use Agentspace (and underlying models like Gemini) to create intelligent assistants that improve customer experience and engagement.
Domain-Specific Expert Agents: In specialized fields, Agentspace-powered agents are helping professionals with complex information. For instance, at Seattle Children’s Hospital, they are launching a “Pathway Assistant” agent (powered by Gemini) to aid clinicians [30]. This agent will help doctors and medical staff by retrieving and summarizing the latest evidence-based practices and complex patient information, effectively ensuring that caregivers have quick access to up-to-date knowledge when making treatment decisions [30]. This kind of agent acts as a medical research assistant, saving time and potentially improving patient outcomes by surfacing crucial information faster. We also see financial and legal industries exploring agents: Moody’s (a financial services company) is testing Gemini 2.5 to do deeper analyses on large document sets like financial statements [31], and Freshfields (a law firm, per other Next announcements) is using multi-agent systems for due diligence. These examples show that whether it’s finance, law, healthcare, or retail, AI agents are being tailored to domain-specific challenges – from sifting through legal documents to answering customer questions – thereby streamlining workflows.
AI-Augmented Development and Operations: Agentspace, combined with Gemini’s coding abilities, also supports developer-focused agents. For example, Nokia built a coding assistant using Gemini to help developers create 5G network applications faster [32]. Such an agent can translate natural language requests into code or automate parts of the development process. Honeywell integrated Gemini into product development tasks [33], likely using agents to generate or evaluate design ideas. These use cases hint at how agents can be used in R&D and engineering to accelerate innovation (brainstorming solutions, writing code, etc.). Moreover, Google’s example with Box, Inc. showed how extractive agents use Gemini to make unstructured content actionable (e.g., reading millions of documents to pull out key points) [34]. Those insights can trigger downstream actions in a multi-agent workflow [35], essentially automating entire business processes (for instance, automatically processing procurement documents and then updating a report). This agent orchestration is exactly what Agentspace is designed for.

Overall, these examples demonstrate the versatility of Agentspace: from internal knowledge management to customer interaction and creative problem-solving, AI agents are being deployed to handle tasks that previously required significant manual effort. Businesses are seeing improvements in speed (e.g. accelerating research by auto-summarization), quality (e.g. more consistent support answers), and even new capabilities (e.g. brainstorming with an AI collaborator).

How Agentspace Enables AI-Driven Workflows

Google Agentspace is more than just a collection of bots; it represents a new way of working. By integrating AI deeply into enterprise workflows, it allows organizations to augment their workforce with AI “co-workers”. Here’s how Agentspace fundamentally enables AI-driven workflows in businesses:

Empowering Every Employee with AI: Because of its user-friendly interface (like searching via Chrome or using a gallery of agents) and no-code creation tools, Agentspace makes AI accessible to non-technical staff. Every employee can have AI at their fingertips to assist with daily tasks – whether it’s drafting a document, answering a complex query, or generating ideas. This democratization means AI is no longer limited to data science teams; it becomes a ubiquitous assistant across the company. Google’s vision, as the tagline suggests, is “AI for every employee.” With Agentspace, an employee might use a Deep Research agent to prepare for a meeting, then use an Idea Generation agent during a brainstorming session, and consult a custom sales data agent to plan a client pitch – all in one day, seamlessly.
Seamless Integration into Existing Workflows: Agentspace’s integrations ensure that adopting AI doesn’t mean disrupting current workflows or switching between dozens of new apps. Instead, AI infuses the tools people already use. Searching in Chrome brings back internal data powered by AI; an agent integrated with Gmail or Chat can offer real-time suggestions or automate responses; a plug-in in Google Sheets might fetch data via an agent query. By fitting into the flow of work, Agentspace allows AI-driven processes to feel natural and to actually save time (as opposed to being a separate, cumbersome process). For example, an employee drafting a budget proposal in Google Docs could invoke an agent (via a sidebar) to pull in the latest sales figures and even write a first draft of an analysis, all without leaving the document.
Cross-Silo Coordination and Multi-Step Automation: Agentspace, especially with its multi-agent support, enables complex automations that involve multiple steps and systems. Consider a workflow like onboarding a new employee: normally, this spans HR systems, IT ticketing, training content, etc. With Agentspace, one could create an Onboarding Agent that automatically pulls the new hire’s info, generates accounts in various systems (through integrations/APIs), compiles a personalized training document (using Gemini to summarize relevant manuals), and answers the new hire’s questions. This agent might internally utilize a chain of sub-agents (one for account setup, one for document creation, one Q&A bot), coordinating via the Agent2Agent protocol. The end result is a much faster, smoother onboarding process, with minimal manual intervention. This is just one example of AI-driven workflows – others might include automated financial reporting cycles, incident response processes (multiple agents handling detection, analysis, and resolution), or intelligent customer service escalation (where an AI triages and only passes to humans what it couldn’t solve). Agentspace provides the platform to design and run these orchestrations reliably.
Continuous Learning and Improvement: By leveraging Gemini 2.5 and future models, Agentspace benefits from ongoing improvements in AI capabilities (reasoning, knowledge, etc.). Agents can be updated or retrained with new data, and thanks to enterprise controls, they can incorporate organization-specific knowledge. This means the AI agents get smarter and more useful over time. For instance, the more employees use the Deep Research agent on company wikis and data, the better it may get at retrieving the most relevant insights (especially if fine-tuned or given feedback). In effect, an Agentspace deployment can learn from an enterprise’s collective interactions, creating a virtuous cycle where AI-driven workflows become increasingly efficient and accurate. Moreover, enterprises maintain control: they can update the agents’ prompts or logic using Agent Designer or ADK as business needs evolve, ensuring the AI continues to align with their goals and policies.

In conclusion, Google Agentspace represents a holistic approach to embedding AI into the fabric of enterprise work. With its robust features (unified search, agent creation tools, galleries, security) and the powerful Gemini 2.5 model behind many of its capabilities, Agentspace enables organizations to create an “AI agent ecosystem” inside their business. Supported by official integrations and a growing marketplace, companies can mix-and-match solutions and have AI agents collaborate to handle everything from routine queries to complex projects. The announcements at Cloud Next 2025 emphasize that this is not a distant vision but an emerging reality: businesses are already deploying these tools and seeing transformative improvements. As one Google VP put it, Agentspace “delivers AI for every employee,” heralding a future where human creativity and decision-making are amplified by readily accessible AI assistance in every workflow [2]

References