Skip to content
Sentynel

#opinion5 min readBy Sentynel

An agent that clicks through your screens inherits the process behind them

Microsoft now gives agents their own Cloud PCs to work in apps without an API. Useful as a bridge, but an agent on old screens repeats the old process faster.

Share

On 17 November 2026 Microsoft said that agents can now get "a computer of their own": Windows 365 for Agents gives them Cloud PCs where they "see the screen, click, type and move data between apps," including legacy and on-premises apps "without an API." It is a real answer to a real problem, because most companies run on systems nobody will rewrite this year. But an agent that works through the screens a person used also inherits every step that person had to take, including the ones that only exist because the systems were never connected. Treat computer use as a bridge to a redesigned operation, not as the operation.

The easy take: automate the clicks, keep the systems

The appeal is obvious. No integration project, no vendor to call, no API to negotiate. The agent opens the same order screen your team opens, copies the same fields into the same ERP and moves on. Microsoft reports that its own business operations team uses Windows 365 for Agents to process orders and has cut processing time by 60%.

Microsoft order processing with Windows 365 for Agents
cut in order processing time, as reported by Microsoft for its own business operations team

Source: Microsoft Windows Experience Blog, 17 November 2026.

That is a meaningful number, and nothing here argues against it. The question is what the number measures. It measures how fast the existing path is walked. It says nothing about whether the path should still have those steps.

A screen is a decision someone made for a person

Every screen in a legacy system encodes a decision about how a person should work: which fields to type, in what order, what to check before saving, which second application to open to look something up. Many of those steps exist because two systems never talked to each other, so a person became the connection between them.

When an agent takes over that screen, the person stops being the connection. The agent becomes it. The gap between the systems is still there; it is now filled by something faster and harder to see. Nobody has to ask why the order number is typed twice, because nobody is typing it anymore.

That is the trap. Copying data between screens used to be visible pain, and pain was what eventually got a process redesigned. A computer-use agent removes the pain without removing the cause.

What the agent inherits

An agent working on the old screens takes on more than the clicks:

  • The order of the steps. If the old flow waits for a manager to open a screen before an order moves, the agent waits too.
  • The workarounds. The spreadsheet someone keeps on the side, the field used for something it was never meant for, the note typed into "comments" because there was no other place for it.
  • The fragility. A screen that changes after an update, a pop-up nobody expected, a session that times out. A person adapts without thinking. An agent needs every one of those cases written down.
  • The missing record. If the old process never logged why an order was edited, the agent will not log it either, unless someone decides it should.

None of these are reasons to avoid computer use. They are the list of things to look at before deciding which steps should survive.

Use the screen as a bridge, with an end date

Computer use earns its place in three situations: a system with no API that will be retired anyway, a step that happens rarely enough that integration is not worth it, and the first weeks of a redesign while the real connections are being built. In all three, the screen is temporary.

So treat each screen-based agent like a bridge in a construction plan:

  1. Write down which systems it connects and why a person was doing it. That sentence is usually the redesign brief.
  2. Mark which steps exist only because the systems are disconnected. Those are the steps to remove, not to speed up.
  3. Give the bridge an owner and an end date. If nobody owns retiring it, the bridge becomes the permanent operation.
  4. Measure the work, not the clicks. Time per order is a start. Cases resolved, errors corrected later and time waiting between steps tell you whether the operation improved.

Connecting the sources is what changed the cycle

In Sentynel's OCESA case, ticket sales, sponsorship results, ratings and media evidence sat in separate places. An agent could have been taught to open each source and copy the numbers into a report. The work that changed the result was different: one reporting operation was built to connect the five evidence sources to the question being asked. Sentynel's project case record reports the reporting cycle going from weeks to minutes and a −70% reported operating cost, with the cost comparison normalized.

OCESA · reporting cycle
WeeksEvidence spread across separate sources
MinutesFive evidence sources connected into one reporting operation

Source: Source: Sentynel project case record. Results are reported project outcomes; the cost comparison is normalized.

The agents in a redesigned operation still act, still read and still write. They do it inside a flow where the systems are connected on purpose, so there are fewer screens to click through in the first place.

A first test

Pick one process where a person spends most of the day moving data between two applications. Before giving that work to a computer-use agent, list every step and mark the ones that would disappear if the two systems shared a record. If more than half would disappear, the agent is not the project. The connection is. Use the agent to keep the work moving while you build it.

Sources

  1. https://blogs.windows.com/windowsexperience/2026/11/17/microsoft-ignite-2026-building-windows-for-the-next-era-of-work/
Share
← All sygnals