The most transformational use of AI in the enterprise won’t come from copilots or new AI-powered tools. It will come from AI agents doing actual work—logging in, reading the screen, sending keystrokes and mouse movements—using the same desktops, apps, and browsers as your knowledge workers.

This won’t be a party trick or a hack. It’s the most practical, secure, and scalable way for AI to enter your enterprise. And it’s just around the corner.

“But that’s ridiculous!”

“Inefficient!”, “Lame!”, “Too much overhead!”, “AIs aren’t humans, they’ll just go direct with APIs”, “Temporary at best!”

I get it. An AI using a GUI sounds like a cartoonish Rube Goldberg machine. But if the goal is for AI to integrate into the work done by human knowledge workers, then the fastest, safest, and most realistic way to get there is to let AI use the same interfaces and tools as the humans.

This is the same reason humanoid robots are shaped like humans. While these generalist robots are not optimal for any one task, they need to operate in the human-shaped world—moving through doors, climbing stairs, and using tools made for people. Their shape isn’t optimized for efficiency, it’s optimized for compatibility with the built environment.

The same is true for enterprise workspaces. If AI needs to interact with apps, processes, email, files, and SaaS, (while being secure, compliant, and observable), then using the existing enterprise workspace via the existing access methods is the only viable option.

The knowledge workspace is human-shaped

Companies are structured and organized around humans and our contraints. We’re slow. We forget things. We need sleep, context, approvals, escalations, and guardrails.

Every enterprise process exists because humans are inconsistent. That’s why we have forms, systems, trackers, reviewers, escalations, and sign-offs. The entire architecture of knowledge work evolved around humans doing work in imperfect human ways.

So if AI is to assist humans in that system, it will naturally need to live within that system. While that may not be pretty, it works and it’s already in place. These are the things enterprises care about.

This is what transformational AI really looks like

Most of today’s AI in enterprise software falls into two categories:

  • Copilots and AI features built into commercial products (Office, Saleforce, etc.)
  • Custom apps and bots built by IT (dashboards, chatbots, wrappers)

These are both excellent uses of AI. But the AI in these cases is limited to single products or workflows and doesn’t fundamentally change how businesses operate.

There’s a third category which I’m focused on today:

  • Workplace AI agents that act like human workers

When does AI become transformational? When it works with your existing stack, running on your existing infrastructure, following your existing governance models, just like your existing knowledge workers.

Letting AI agents operate GUIs like humans brings real, immediate benefits:

  • No re-platforming: Use the tools and infrastructure you already have.
  • No app rewrites: AI can navigate any app with a UI.
  • Full auditability: Session recordings show exactly what happened.
  • Familiar policies: DLP, ZTNA, monitoring, & everything else you already do today.
  • No learning curve for IT: Support, governance, and incident response all stay the same.
  • Easy to start: Begin with one agent and one task. Scale when you’re ready.
  • Fallbacks are simple: If an agent fails, a human picks up where it left off.
  • Workers can drive it: Let people experiment and co-pilot their own AI agents.
  • Ongoing maintenance is identical: Update, retrain, monitor, and adjust things just like you would a human or script.

This is how transformation happens in the real world

Transformation is never clean or fast. You don’t swap out enterprise systems overnight, you adapt step by step over time. The intermediate steps might look strange by themselves, but when taken in context with the overall arc (whose story isn’t finished), each step makes sense.

This is why I wrote about how electric motors didn’t transform factory layouts overnight. Especially in today’s enterprise environment where everything depends on traceability, compliance, and trust, taking small, safe steps is a must.

AI agents operating your computer will start small, automating a few repetitive tasks, then augmenting a few teams, and helping a few workers. But as the changes compound, you’ll want an execution environment that’s already secure, governable, observable, and extensible.

That’s Citrix.

We’ve already built that workspace: a dynamic, secure, policy-driven environment where work happens, regardless of who (or what) is doing the work. Human or AI, nothing changes: same identity, same guardrails, same observability, same control.

You don’t need to invent a new execution layer for AI. You already have one.

While we’re not yet in a world where computer-using agents are practical at scale, we’re close. And when they arrive, they won’t need special treatment. They’ll log in like anyone else and get to work.

So the question isn’t whether your workspace is ready for AI. The question is whether your AI is ready for the workspace you already have.

Read more & connect

Join the conversation and discuss this post on LinkedIn. You can find all my posts on my author page on the Citrix blog (or via RSS).

Video of my most recent talk

In May I gave the closing keynote at the EUCtech Denmark 2025 conference, called The Future of Work in an AI-Native World. I talked about a lot of what I covered today and walked through how AI will evolve and impact the workplace in the coming years. You can watch it on YouTube.

My upcoming talks

  • AppManagEvent: Closing Keynote: AI & the Future of Enterprise Apps — Utrecht, Netherlands, Oct 10
  • MAICON 2025: AI at Work: The Employees’ Revolution! — Cleveland, Ohio, Oct 14-16