The next workspace Isn’t just for humans

Picture a global manufacturer that finds a critical defect in one of its products. Within minutes, a hundred things have to happen at once: orders paused, suppliers and customers notified, inventory reallocated, regulatory paperwork prepared, executives briefed. Even a year ago, that meant pulling hundreds of people into virtual war rooms, each logging into a dozen applications to work through one task after another.

Increasingly, the first responders won’t all be people. Some of the work will be handled by AI agents that flag affected customers, update ERP systems, and assemble the first draft of a compliance package, while the people stay focused on the calls that need judgment. By the time the first executive briefing starts, several workstreams are already moving, with humans and agents working the problem together.

The question is no longer whether AI can do this work. It’s where the work happens. That sounds like an implementation detail. I’ve come to believe it’s one of the defining architectural questions of enterprise AI.

Over the past six months, AI has moved from conversation to execution. Enterprises aren’t experimenting with chat anymore; they’re putting agents to work writing software, securing infrastructure, processing claims, onboarding vendors, or reconciling invoices. The companies finding real momentum have stopped asking whether agents belong in the enterprise. They’re asking where those agents actually run, and whether that environment is one they can trust.

Every major shift in enterprise computing has redefined what a workspace is for. The personal computer gave each employee a digital desk. The internet connected those desks, and the cloud made them accessible from anywhere. Through all of it, one assumption held: the only workers using workspaces were people. That assumption is starting to break.

Gartner® recently predicted that “by 2028, Fortune 500 enterprises will scale to over 150,000 agents, creating massive agent sprawl and critical IT management challenges, up from 15 in 2025.”[i]

These aren’t chatbots waiting for a prompt. They’re digital workers that navigate enterprise applications, reconcile transactions, prepare reports, and coordinate workflows alongside people. Like any worker, they need an identity, permissions, policies, and accountability. They need somewhere to do the job. That somewhere is the enterprise agentic workspace, no longer only the place where people are productive, but the operating environment where humans and agents work side by side.

There’s a popular belief that agents will reach enterprise software entirely through APIs. The reality looks different. Talk to any enterprise that runs the software behind banks, hospitals, and government agencies, and ask how many of their mission-critical applications expose modern APIs. According to McKinsey, 70% of enterprise software used by Fortune 500 companies was developed over 20 years ago. These systems hold decades of institutional knowledge, custom workflows, and compliance logic that aren’t going anywhere just because AI arrived. They are too important to abandon and complex to modernize, so agents have to meet them where they already are, through the same screens people use every day.

Which is why the harder problem was never teaching an agent to click. It’s making sure those clicks happen inside an environment the organization can actually trust. Most companies don’t stall because their agents aren’t capable enough. They stall because they can’t safely scale autonomous work. Ungoverned agents create more than security risk; they create drag. Security teams can’t audit what they can’t see, compliance teams can’t certify what they can’t trace, and agent owners can’t steer what they can’t observe. So, every meaningful action waits on a human approval; agents stay stuck in pilots, and the productivity gains never quite arrive.

Here’s the pattern I keep coming back to, after deploying agents across our own teams: an agent earns trust the same way a new hire does. You don’t give someone unrestricted access to every critical system on their second day. They start with guidance, they learn about the business, and they take on more as they gain experience. Agents are no different. Trust isn’t granted; it’s earned, whether the new worker is a person or a piece of software.

An agent running straight off a laptop inherits everything the laptop has: your operating system, your credentials, your network, your files. There’s almost no line between the person and the autonomous worker. A governed cloud workspace redraws that line. Each agent gets its own identity, permissions, an isolated session, and clear network boundaries. Every action is observable, every decision is accountable, and a human can watch a session in real time, the same over-the-shoulder check-in you’d give someone in their first week. Governance stops being something you bolt on afterward and becomes part of the architecture.

At the AWS Summit in New York City, we introduced a set of services built on a shared idea: agents create the most value when they build knowledge over time inside trusted boundaries. AWS Continuum, our new agentic security service, addresses code vulnerabilities at machine speed. AWS Context gives agents real business understanding by connecting enterprise data, relationships, and institutional knowledge. Amazon Bedrock AgentCore handles orchestration. And Amazon WorkSpaces for Agents provides the governed environment where agents work with applications exactly the way employees do today, even when no API exists.

With WorkSpaces for Agents now generally available, organizations can give their agents a secure cloud desktop to operate enterprise applications and run autonomous work, without rebuilding decades of existing software. It’s framework-agnostic, it connects to the identity systems you already run, and the economics are simple: you pay only while an agent is actually working. Most of that trust isn’t new. Since 2014, Amazon WorkSpaces has powered millions of enterprise sessions across financial services, healthcare, manufacturing, retail, and government, under demanding regulatory requirements around the world. We didn’t add trust for the agentic era. It was already there.

Over the next five years, every enterprise will run three workforces at once: people, agents, and the traditional software systems that have always done the quiet work in the background. What separates the leaders won’t be who has the smartest model; models are converging, and they’re getting more capable for everyone. The companies that will lead won’t just have the smartest AI, they will build the most trustworthy place for intelligence to work, creating workspaces where people and agents accomplish more together than either could alone.

For more information, read more about how to get started with WorkSpaces for Agents.

Ruslana Zbagerska is Vice President of Amazon WorkSpaces at AWS.

[i] Gartner, Emerging Tech: Tech Innovators in AI Agent Management Platforms, Ethan Cai, Alfredo Ramirez IV, Kiumarse Zamanian, Christine Tenneson, Aakanksha Bansal, 27 May 2026