Giving an AI agent a real place to work

Employee Experience Accessibility
Written by
Posted On
Duration of read
5  min
Share Article
Related Topics
Subscribe via email

How Advania is using Windows 365 for Agents and Copilot Studio taking repetitive project set-up out of the PMO queue. 

The short version: A PMO coordinator asks an agent in Microsoft Teams to process an opportunity. The agent checks the submission, gathers the right documents and carries out the UI-bound work in Kantata from a governed Cloud PC. People stay in control of exceptions and approvals. 

15 to 60 minutes

manual effort per set-up today 

3 to 6

set ups on a typical day

Up to 7 hours 

potential daily effort identified in early scoping 

The work was ordinary. That was the point.

When a sales opportunity is ready for delivery, the hand-off into project delivery must be clean. The documents need to be present, the data needs to be right and the project needs to be set up in the system the delivery team actually uses.

At Advania, that work previously crossed a SharePoint submission list, the document store behind Dynamics 365 and Kantata. It was repetitive, took 15 minutes to an hour per project, and came back several times a day. None of it was especially taxing , but It was work which required accuracy and attention each time, a process it felt as though there was a better way of executing, whilst freeing up PMO time to better add value elsewhere.

It also exposed a familiar gap in automation, not sitting neatly behind one API, some of it living on screens, forms and documents, with only a few exceptions still needing human intervention. Any conventional integration would either still leave gaps or require greater engineering than the problem justified.

A Cloud PC for the agent, not another shared account

Advania built the project coordinator in Copilot Studio and gave its computer use capability a Windows 365 for Agents Cloud PC pool. The agent can work through applications in the same way a person would, using the screen, keyboard and mouse, but from an environment that is joined to Microsoft Entra and managed in Microsoft Intune.

The first version is deliberately simple; a PMO user starts the process in Microsoft Teams and gives the agent an opportunity number, The agent then:

  • Checks the Deal Pack submission and confirms the opportunity is ready to progress
  • Retrieves the required files from the SharePoint document store behind Dynamics 365
  • Uses Kantata through the user interface to complete the remaining project set-up work
  • Reads the result back, reports the outcome in Teams and stops when information is missing or unclear

The human element is not designed out of the process. The agent does the repetitive navigation and data entry. The PMO keeps control of the decision points, exceptions and final approval while the service is being proven.

Why Windows 365 matters: The interesting part is not that an agent can click through an application. It is that the work happens on a managed Windows endpoint, under the same identity, access and compliance disciplines the endpoint team already understands. 

Built as an operational service, not a demo

The hard part was not getting an agent to move a mouse. It was working through the questions that decide whether an agent belongs in a live estate: whose identity it uses, what it can reach, how Conditional Access applies, what happens when a document is missing, how an administrator reviews a run and how cost is measured once the novelty wears off.

Advania brought its Modern Endpoint and Automation teams together to answer those questions. The Cloud PCs are treated as part of the endpoint estate, with Entra identity, Intune management and compliance controls. The agent is constrained to the systems it needs, its run history can be reviewed, and sensitive or ambiguous steps can pause for a person rather than guessing.

That joint approach matters. Agent projects are often treated as an AI exercise until they meet identity, networking, endpoint management and operations. By then, the design is already boxed in. Here, those controls were part of the work from the start.

Capacity

The target is to remove most of the coordinator effort from a task currently repeated three to six times a day.

Consistency 

The same readiness checks and hand-off steps can be applied on every run, with exceptions returned to a person.

Speed

The intended outcome is a project hand-off completed in minutes rather than a manual queue.

Control

The execution environment is managed as an endpoint, with identity, compliance and run evidence available to administrators.

Learning

The deployment is giving Advania practical evidence on where computer use fits, where it struggles and what customers need before production.

Important: these are early deployment findings and target outcomes. Final time saved, completion rate, intervention rate and cost per successful run should be published only after UAT and live run data have been signed off. 

Five years on, Windows 365 has a new kind of user

Windows 365 started by giving people a secure Windows desktop from the cloud. Five years later, the same platform is becoming a place for agents to work. That is a bigger shift than it may first appear. The agent is no longer an abstract bot calling a connector somewhere. It has an identity, an endpoint and a visible trail of work.

For customers, this opens a route into the applications that have resisted automation because they lack a usable API or depend on an interactive Windows session. It also raises the bar. If an agent is going to work across business systems, it needs the same serious conversation about access, policy, data and ownership as any other worker.

What this means for customers

The sensible starting point is not a list of everything an agent might do. It is one real task that people already understand, with enough repetition to matter and enough boundaries to measure.

    • Choose a process that is repetitive, screen-based and currently awkward to integrate.
    • Agree what success looks like before building: completion, duration, interventions, exceptions and cost per run.
    • Bring endpoint, identity, security and the business owner into the design at the beginning.
    • Keep a human checkpoint until the evidence supports removing it.
    • Scale only when the run data says the workload is reliable and economic.

Advania can help

Our Windows 365 for Agents Proof of Value takes one real workload, gives the agent a governed Windows desktop and measures what happens. The outcome is a practical scale, adjust or stop decision, backed by run evidence rather than a polished demo.

“We did not want an agent that looked clever in a demo and then became someone else’s security problem. We wanted to prove that it could do useful work inside the controls we already run. Windows 365 for Agents gave us the execution layer. The project is teaching us where this pattern is genuinely valuable, and where it still needs work.”

Mike Parker, Director of Digital Workplace, Advania UK

Sign up to receive insights from our experts

Get the latest news and developments from Advania delivered to your inbox

Other articles that might interest you

Sign up to receive insights from our experts

Get the latest news and developments from Advania delivered to your inbox.