Rethinking The Infrastructure Behind Intelligent Workspaces

Elena Koryakina is Chief Product & Technology Officer at Parallels, leading product strategy, engineering, DevOps, AI, and infrastructure.

getty​Enterprises are adding copilots, chatbots and AI agents at a remarkable pace. Yet much of the discussion about AI still focuses on what users see: the interface, the model and the latest AI feature.

The more important changes are happening beneath the surface.

As AI moves from answering questions to taking action, the infrastructure supporting the digital workspace must evolve with it. An AI agent that retrieves information, interacts with applications or executes workflows becomes another actor in the enterprise, requiring its own identity, permissions, policy controls and secure access to applications and data.

Technology leaders should look beyond how many AI capabilities they can deploy and ask whether the environment supporting them can securely, consistently and at scale support those capabilities.

The first wave of generative AI was largely about experimentation. Employees interacted with copilots and chatbots, asking questions and evaluating responses. Agentic AI fundamentally changes that relationship.

Instead of answering questions, AI agents can perform work on a user’s behalf, coordinate across multiple systems and complete business processes with minimal human intervention.

Imagine an agent onboarding a new employee. It retrieves HR information, creates accounts, provisions applications and notifies managers.

The conversation soon shifts away from model quality.

Who authorized the agent? What systems can it access? What actions is it permitted to perform? How are those actions monitored and audited?

These are infrastructure questions.

The industry is beginning to respond. Microsoft, for example, has introduced purpose-built identities for AI agents. Enterprise identity will increasingly need to encompass both people and the AI agents acting on their behalf.

Many organizations are not yet prepared for that shift. Deloitte’s 2026 “State of AI in the Enterprise“ research found that only 21% have a mature governance model for autonomous AI agents. The conversation is shifting from AI adoption to the infrastructure required to operate it securely and at scale.

Treating AI like any other enterprise platform is a good place to start. Organizations should establish governance, security and architecture standards before individual teams begin deploying agents at scale.

Traditional identity systems were designed around human users. Employees authenticate, receive permissions based on their role and operate within defined access policies.

AI agents don’t fit neatly into that model.

Allowing an agent to simply inherit a user’s credentials creates unnecessary risk. An agent may require access to a small subset of the resources available to the individual it represents.

A better approach is delegated identity, where agents receive only the permissions needed for a specific task. Organizations can then determine who authorized the agent and whether its actions remained within policy.

As a practical first step, technology leaders should assign AI agents their own identities with task-specific permissions that can be monitored and audited.

Trust must also extend beyond identity.

Governance must extend beyond individual agents to the services, APIs, tools and data they use. As AI ecosystems become more connected, organizations need confidence that agents are using trusted tools, working with approved data sources and operating within consistent policy.

Model Context Protocol (MCP) is emerging as a standard for connecting AI systems with enterprise applications and data. As adoption grows, organizations will need trusted mechanisms for determining which tools agents can access.

Approved MCP registries may become an important governance layer as the ecosystem matures. Organizations should also establish a centralized approval process for AI tools and integrations, much like many already do for SaaS applications. A trusted inventory of approved models, tools and data connections becomes increasingly valuable as AI adoption accelerates.

As organizations move from a handful of assistants to hundreds of specialized agents, they’ll need orchestration platforms that coordinate identities, policies, applications and governance.

The execution environment must evolve as well.

Employees increasingly interact with AI through browsers, SaaS applications, virtual desktops and cloud services. At the same time, AI agents may communicate directly with enterprise applications and data without any human interaction.

Traditional endpoint-centric security becomes increasingly difficult to maintain in that environment.

Instead, the digital workspace itself can become the policy enforcement layer.

Protected AI execution environments can enforce consistent policies across models, applications and data while giving organizations visibility into agent activity. Technologies such as browser isolation add another layer by separating AI interactions from the endpoint and providing a consistent place to monitor behavior and enforce policy.

Data protection must evolve alongside AI. Every prompt or workflow creates another path for sensitive information to leave the organization. DLP strategies should extend to AI interactions, allowing organizations to identify and control sensitive data before it reaches external models.

Trying to secure every AI service individually becomes increasingly impractical as the number of models and services grows. Instead, organizations should define consistent controls across the workspace, including which AI services employees can use, what data can be shared and where AI interactions should occur.

The digital workspace is evolving from a way to deliver applications to an orchestration layer that connects people, AI agents, applications, enterprise data and security policies.

Within that environment, identity establishes who or what can act. Policy defines the boundaries, trusted connections determine which services agents can use and data controls govern what information can move between systems.

Enterprise AI maturity isn’t measured by how many agents an organization deploys. It’s measured by whether those agents can be trusted to operate as part of the business.

1. Do AI agents have identities that are separate from human users?

2. Can we control exactly what systems and data they can access?

3. Can we explain and audit every significant AI action?

4. Are AI interactions governed consistently across our digital workspace?

5. Can we introduce new AI services without redesigning our security model?

The organizations that lead in AI will be those that rethink the infrastructure of intelligent workspaces and build the governance needed to support AI at scale.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?