Microsoft Copilot Studio Drops an AI Agents That Can Use Your Apps Like You Do

Microsoft just launched a new feature in Copilot Studio called “Computer Use.”
It lets AI agents perform actions on your screen - like clicking buttons, typing into fields, scrolling through web pages, and moving between apps.

So if your process lives across a bunch of websites, popups, and half-working tools with no API - Copilot can now handle it.

📣 What Microsoft Is Saying

Microsoft describes this as an upgrade that allows agents to interact with graphical user interfaces - enabling them to complete tasks by simulating mouse clicks and keyboard inputs.

They say it gives agents the ability to “operate across systems that lack APIs or direct integrations.”
In other words: if your software doesn’t talk to anything, Copilot will still get the job done.

They also point out that the system uses deep reasoning to understand what’s on screen and decide how to proceed - even when the layout changes.

💬 In Human Words

This means your Copilot agent doesn’t need everything to be perfect.
It can:

• Recognise login pages, even if the button moved

• Fill out a form, even if the labels are weird

• Complete multi-step tasks, even when something unexpected pops up

Why?
Because it’s been trained to recognise patterns - and can figure out what to do based on what it's seen before.
It doesn’t memorize screens. It reads them.

Here’s how it compares:

This isn’t AI that waits for clean instructions.
It rolls with the mess.

🧪 What No One’s Saying Out Loud: This Is Where Deep Research Algorithms Actually Start to Matter

Until now, deep research algorithms were mostly used to create smart documents - writing reports, pulling facts, helping with research-heavy content.
We’ve seen vendors dress it up with titles like “academic-level AI” or “a chatbot with a degree.”

That sounded impressive - but it didn’t show what this kind of algorithm could really do.

This update changes that.

Because this time, it’s not being used to write about something - it’s being used to do something.

• It clicks

• It types

• It navigates systems with no instructions

• It works even when the layout is different every time

That’s possible because this algorithm works from patterns - not strict commands.
It’s not next-level chat.
It’s navigation.
It’s digital action.

Old bots needed perfect labels and pixel-perfect screens.
This one doesn’t.
That’s the real impact.

This is where we see real change - because it’s finally doing the work, not just describing it.

🔚 Bottom Line

Cost: Included in Microsoft Copilot Studio
Availability: Rolling out now
Read more.

If your daily workflow includes ugly software, weird portals, and "why is that button there?" moments - Copilot might finally be able to help.

🧊 Frozen Light Team Perspective

Everyone’s busy talking about the feature.
Yes, it’s useful.
Yes, it clicks buttons and submits invoices.
But we’re looking at the algorithm - because that’s the part no one’s talking about loud enough.

This is a deep research algorithm - and it’s working outside the lab.
It’s not summarising. It’s not writing.
It’s actually doing the things we hate doing:

• Submitting invoices

• Logging into clunky systems

• Uploading files across broken portals

• Finishing onboarding across apps that don’t connect

That’s real progress.
And we’re here for it.

But here’s why we’re really paying attention:
A few days ago, Google gave website publishers a choice -
Let us train our deep research algorithm on your content… or opt out.

And guess what?
Most of them said no.
Google’s token dropped - fast.

That’s the kind of impact these algorithms have when they don’t get access.
Now flip it.
Microsoft’s version is in the field, in motion, and getting better - because people are using it.

And here’s the uncomfortable truth:
When we use these tools, we are part of that training.
We’re not saying that’s good or bad.
We’re saying: this is the role users play in AI right now - whether we admit it or not.

Microsoft didn’t randomly drop this model into Copilot Studio.
And we don’t believe they trained it in isolation.
If we’re wrong? Sorry, Microsoft.
But we don’t think we are.

So no - you don’t have to agree.
You don’t have to participate.

But we are going to keep pointing out what no one else is saying:
Progress comes from the algorithms that are allowed to play.
And this one is playing - hard.

We’re excited.
Because this isn’t another productivity boost.
This is a new way of getting things done.
And we’re watching to see how this algorithm evolves across all major vendors.