AI Agents Are Coming for the Endpoint: Why the Future Workplace Needs a New Control Point

From the PC era to the agentic endpoint

There was a time when the endpoint was easy to understand.

A PC was a personal computer.
A laptop was a mobile office.
A thin client was a window into the data center.
A browser became the new workspace.
Then VDI, DaaS, SaaS, cloud PCs and Zero Trust changed the way we looked at the edge.

But through all these waves, one principle remained almost unchanged: the user was the actor.

The user clicked.
The user opened the application.
The user copied the file.
The user approved the workflow.
The user made the mistake.

Artificial intelligence is now breaking that model.

The next endpoint landscape will not only serve humans. It will also serve autonomous agents: software entities able to understand intent, call tools, search data, trigger actions, collaborate with other agents and sometimes operate faster than the human sitting in front of the screen.

That is why the question raised in the Journal du Net article “Doit-on inventer un nouvel OS pour les agents IA ?” is so important. The article argues that AI agents do not need a new Windows, Linux or macOS, but rather a new execution, orchestration, security, memory, identity and governance layer an “Agent OS” in the operational sense.

And this idea has a direct consequence for End User Computing. If agents become actors, the endpoint can no longer be treated as a simple access device. It becomes a control point.

The AI PC is not the full answer

The market is already moving. Gartner expects AI PCs to represent 31% of the worldwide PC market by the end of 2025, with 77.8 million AI PCs shipped that year. Other analysis also points to AI PCs exceeding half of PC sales around 2026, depending on supply chain conditions and enterprise refresh cycles.

At first glance, this looks like the obvious endpoint revolution: put more AI capability into the device, add an NPU, run models locally, and call it the future workplace. But this is only part of the story. The real disruption is not that the endpoint can run AI. The disruption is that AI can run work.

An AI PC can accelerate local inference. It can improve video, summarization, translation, security detection or personal productivity. But an autonomous agent is different. It needs access to applications, data, credentials, memory, APIs, files, SaaS platforms, collaboration tools and workflow systems.

That means the enterprise challenge is not only: “Do we need more powerful endpoints?”

It is also: “How do we control what autonomous software can do from, through, or around the endpoint?”

This is where the conversation changes.

The endpoint was designed for people. Agents will expose its weaknesses.

For decades, endpoint management was built with a “human centric” way.

We secured login.
We patched operating systems.
We deployed applications.
We encrypted disks.
We controlled USB ports.
We monitored malware.
We managed browsers.
We tried to stop users from clicking on the wrong link.

Now imagine a new worker joining the digital workplace.

It never sleeps.
It can process hundreds of documents.
It can call multiple tools.
It can create tickets, draft emails, update CRM records, analyze logs, generate code, summarize meetings and request approvals.
It can also misunderstand context, leak data, call the wrong API, act with excessive privileges or be manipulated by prompt injection.

The Journal du Net article makes an essential point: in classic IT, we supervise applications; in agentic IT, we must supervise decisions. In classic IT, we assign rights to users and services; in agentic IT, we assign rights to entities that interpret intent and act across systems.

That distinction is fundamental. Because when an agent acts, the endpoint becomes part of a larger chain of responsibility.

Was the action initiated by the user?
By the agent?
By another agent?
Through which identity?
Using which data?
From which device posture?
Under which policy?
With which approval?
And can we prove it afterwards?

This is the latest EUC-related issue we’re facing.

The old endpoint model was already under pressure

Before AI agents, the endpoint landscape was already changing.

Windows 11 migration, hybrid work, ransomware, SaaS sprawl, VDI modernization, DaaS, browser isolation, contractor access, frontline workers, OT environments and regulatory pressure had already pushed organizations to rethink the classic “fat endpoint” model.

Microsoft’s Windows 365 Link is a good signal of this market direction. Microsoft describes it as a purpose-built Cloud PC device for Windows 365, with no local data, no local apps, no local admin users and centralized management through Intune. Microsoft also states that this class of Cloud PC devices reduces attack surface by requiring minimal configuration and enabling security features by default that cannot be disabled.

In other words, even Microsoft is telling the market something important:

The endpoint of the future may not be a general-purpose PC.
It may be a secure gateway to a controlled digital workspace.

AI makes that trend even more relevant.

Because the more intelligence moves into the workspace, the more the physical endpoint must become predictable, trusted, observable and replaceable.

Why autonomous agents will reshape endpoint architecture

Autonomous agents will impose four major changes on the endpoint landscape.

1. Identity will move beyond the human user

In the past, the endpoint answered one main question: who is logged in? In the agentic workplace, that is not enough.

We will need to know:

Who is the user?
Which agent is acting?
Who owns that agent?
What is the agent allowed to do?
Is it acting independently or on behalf of a human?
Can it access local resources?
Can it trigger remote actions?
Can it read sensitive data?
Can it write, delete, send, approve or execute?

This requires a new identity model where humans, devices, applications and agents are all governed entities.

The endpoint must participate in that identity chain. It cannot remain a passive device.

2. Local data will become even more dangerous

For years, EUC architects have tried to reduce the amount of data stored on endpoints. The reason was simple: lost laptop, stolen data, ransomware encryption, unmanaged copies, shadow IT.

“With agents, this risk increases”

An agent with access to local files, cached credentials, browser sessions or temporary downloads can become a powerful exfiltration path. Even if the user does not intentionally leak data, the agent might process or expose information that should never have been in scope.

That is why “no local data” becomes more than a thin client feature. It becomes an AI governance principle.

IGEL’s Preventative Security Model is relevant here because IGEL OS is designed around a read-only, locked-down endpoint where unauthorized changes cannot execute or persist, and where no data is stored locally on the endpoint.

In an agentic world, reducing endpoint persistence is not only a security choice. It is a way to reduce the blast radius of autonomous action.

3. The endpoint must become easier to reset than to repair

The classic enterprise endpoint often carries years of complexity: local applications, drivers, agents, policies, VPN clients, security tools, user profiles, cached data and exceptions.

“This creates operational friction“

When something goes wrong, IT must investigate, remediate, reimage, patch, recover or replace. In the age of AI agents, this model becomes too slow.

If an endpoint is suspected of being compromised, polluted, misconfigured or exposed to risky agent activity, the desired answer should be simple:

Reboot.
Return to known-good state.
Reconnect to the controlled workspace.
Continue working.

“This is where immutable endpoint design becomes strategic“

The Journal du Net article compares the future Agent OS challenge with Kubernetes: containers existed before Kubernetes, but Kubernetes made large-scale operation possible. We can apply a similar logic to endpoints. Thin clients, Linux endpoints and cloud workspaces existed before AI. But AI agents will make controlled, industrialized endpoint operations far more important.

4. User experience will become a policy surface

In the past, user experience was mostly about performance and ergonomics. In the agentic workplace, user experience also becomes a governance layer.

When should the user validate an agent action?
When should the endpoint block a workflow?
When should the session require stronger authentication?
When should screen capture, clipboard, local printing or file transfer be disabled?
When should an agent be visible to the user?
When should the user know AI-generated content is being used?

The EU AI Act is moving in this direction. The European Commission published draft guidelines in May 2026 for transparency obligations under Article 50 of the AI Act, aiming to help providers and deployers comply consistently.

This means endpoint experience, compliance and AI transparency will increasingly overlap. The endpoint will not just display the workspace. It will enforce part of the trust model.

Where IGEL fits the gap

IGEL does not need to become the “OS for AI agents“. That is not the point.

The point is that AI agents will need a safe, controlled and resilient execution environment around them. Some of that control will live in the AI platform. Some will live in identity. Some will live in SaaS governance. Some will live in the cloud workspace.

But the endpoint remains the first and last meter of the digital workplace.

This is where IGEL fits the gap

IGEL as the trusted access layer

IGEL OS can act as a secure access layer between the user, the physical device and the digital workspace. Instead of turning every endpoint into a complex, general-purpose attack surface, IGEL reduces the local footprint and focuses the endpoint on secure access.

For architects, this matters because the future AI workspace will likely be distributed across:

Cloud PCs
VDI and DaaS
SaaS applications
Enterprise browsers
AI copilots
Agentic workflows
Data platforms
Identity systems
Security control planes

The endpoint should not become the place where all this complexity accumulates. It should become the place where access is controlled.

IGEL as a preventative security control

IGEL’s Preventative Security Model positions the endpoint as a proactive participant in Zero Trust, with IGEL OS embedding security into the endpoint and using real-time context and dynamic configurations to inform access decisions.

This is important for AI because autonomous agents increase the need for preventative controls.

Detection will still matter.
EDR will still matter.
SIEM will still matter.
SOC operations will still matter.

But when agents act at machine speed, prevention becomes more valuable. You do not want to discover after the fact that an agent copied sensitive files, triggered the wrong workflow or used a compromised endpoint session.

You want the endpoint architecture to make certain classes of risk impossible, or at least significantly harder.

IGEL as a bridge to the cloud workspace

The AI endpoint discussion is often framed as “AI PC versus cloud.”

That is too simplistic.

Some AI workloads will run locally.
Some will run in the cloud.
Some will run in SaaS.
Some will run in private infrastructure.
Some will run inside controlled agent platforms.

The real architecture question is workload placement.

Where should the data live?
Where should the model run?
Where should the session execute?
Where should the policy be enforced?
Where should the audit trail be kept?

IGEL fits well in this hybrid reality because it allows organizations to extend the life of existing endpoint hardware while moving the Windows, application and data execution layers into more controlled environments. This is especially relevant at a time when PC refresh economics are under pressure. IDC has recently revised its PC outlook downward amid memory and supply chain disruptions, forecasting an 11.3% decline in global PC shipments in 2026.

AI is not only changing software architecture. It is also putting pressure on hardware economics. That makes endpoint reuse, lightweight operating models and cloud workspace strategies more attractive.

The future endpoint will be judged by control, not by power

The endpoint debate has often been reduced to specifications. CPU / RAM / Disk / GPU / NPU / Battery / Screen / Weight.

Those metrics still matter. But the agentic enterprise will add new criteria:

Can the endpoint prove its integrity?
Can it prevent persistence?
Can it avoid local data exposure?
Can it enforce workspace policies?
Can it recover quickly?
Can it support Zero Trust access?
Can it reduce operational complexity?
Can it integrate with cloud-delivered workspaces?
Can it support AI without becoming an uncontrolled AI playground?

The AI PC will be useful for some users and some workloads.

But for many enterprise scenarios (healthcare, finance, manufacturing, retail, call centers, contractors, shared workspaces, regulated environments, frontline workers, IT/OT) the bigger question will be control.

Not every endpoint needs to become an AI workstation. Many endpoints need to become trusted gateways into an AI-enabled workspace.

A new architecture pattern: Agentic EUC

We may need a new term for this shift.

Call it Agentic EUC

Agentic EUC is not about putting a chatbot on every desktop. It is about redesigning the digital workplace for a world where humans and agents collaborate inside governed workspaces.

An Agentic EUC architecture should include:

Secure endpoint layer: Immutable, minimal, centrally managed, with no unnecessary local data.
A cloud workspace layer: Windows 365, Azure Virtual Desktop, Citrix, VMware, Omnissa, Amazon WorkSpaces or other controlled workspace models.
An identity and policy layer: User identity, device posture, agent identity, conditional access, privilege boundaries and approval workflows.
An AI agent layer: Copilots, specialized agents, orchestration frameworks, model access, tool connectors and workflow automation.
An observability layer: Tracing of actions, prompts, tools, data sources, costs, validations and policy decisions.
A governance layer: Lifecycle, ownership, risk classification, compliance, auditability and business value measurement.

This is where the endpoint becomes strategic again. Not because it runs everything. But because it controls the entrance to everything.

The lesson from the past

We have seen this movie before.

When PCs arrived, IT lost control of computing power that had previously lived in the data center.
When the web arrived, applications escaped the local machine.
When mobile arrived, the perimeter dissolved.
When cloud arrived, infrastructure became programmable.
When SaaS arrived, business units became technology buyers.
When VDI and DaaS matured, the desktop became a service.
When ransomware exploded, the endpoint became the battlefield.

Now AI agents are arriving. And once again, the endpoint has to evolve. The mistake would be to think that the future endpoint is simply a more expensive PC with an NPU. The real future endpoint is a trusted, policy-driven access point into an intelligent enterprise.

Conclusion: IGEL and the endpoint after the AI tipping point

AI agents will not wait for endpoint teams to be ready. They will enter the enterprise through browsers, SaaS platforms, copilots, automation tools, developer environments, collaboration suites and business applications.

Some will be approved. Some will be invisible. Some will be useful. Some will be risky. All of them will force IT leaders to rethink control. The Journal du Net article is right to ask whether agents need a new OS. But in End User Computing, we should ask a parallel question: Do endpoints need a new role in the age of autonomous agents?

“The answer is yes“

The endpoint must become less persistent, less exposed, less complex and more governed. It must become a clean, resilient and trusted access layer. It must support human productivity while limiting agentic risk.

That is precisely where IGEL has a strong role to play.

Not as the brain of AI.
Not as the agent platform.
Not as another layer of complexity.

But as the secure edge of the agentic workspace. Because in the AI era, the endpoint that matters most may not be the one with the most local intelligence. It may be the one that knows exactly what should never happen locally.

Share this content: