How to Use Microsoft Copilot Studio to Automate Manual Browser Tasks With No API

If you spend time every week clicking through a web portal, copying data by hand, and pasting it somewhere else, this article is for you. That task feels minor until you add up how often it breaks your focus and how many hours it quietly consumes each month. If you are a manager, ops professional, or team lead who runs on Microsoft 365, you can set up an agent in Copilot Studio that does that clicking for you while you work on something else. By the end of this article, you will know exactly how to build it.

Why This Matters

The specific pain here is not the volume of data. It is the portal with no API and no export button. Whoever built it did not anticipate that you would need the information somewhere else. So every week, someone on your team logs in, clicks through several screens, reads the fields that matter, and copies them manually. It is not difficult. It is just slow, repetitive, and it fractures concentration every time it happens. Tasks like this are invisible to most automation tools because they require a human to operate the browser the way a human would. That is exactly the gap that computer-using agents are designed to close.

The System

Microsoft made computer-using agents in Copilot Studio generally available recently. The core capability is that the agent reads what is on screen and interacts with it, without needing an API connection or a custom integration. Here is how to build a working version.

Step one: identify the task. You need a browser-based workflow that follows the same sequence of clicks and reads every time. Manual data collection from a project tracker, a supplier portal, or an internal reporting tool are all good candidates.

Step two: write out every step as a numbered sentence, as if you are explaining the process to someone who has never seen the screen. Be specific about field names, section labels, and where to navigate. This document becomes your agent instructions.

Step three: open Copilot Studio, create a new agent, and paste your instructions into the task configuration. Keep the scope tight on the first run. Narrow boundaries are how you catch errors before they compound.

Here is the prompt structure that worked in testing:

PROMPT: Open [portal URL] and log in using the saved credentials. Navigate to the projects section. For each project listed on the first page, read the status field and the last updated date. Enter each project name, status, and date into the open spreadsheet, one row per project. When the first page is complete, stop and wait.

Step four: run the agent and review the output manually. On the first run, expect one or two errors, particularly where the portal uses inconsistent field labelling. Update your instructions to account for those variations, then run again. The second run should be clean.

A realistic setup time is 60 to 90 minutes across two sessions. That is worth it if the task repeats weekly. For something you do once a month, the calculation is less clear.

Note on access: Copilot Studio is part of Microsoft Power Platform and requires a licence. If your organisation uses Microsoft 365, Power Platform access may already be available to you. IT may need to enable it.

This is the kind of practical AI workflow covered every week in the AI at Work newsletter on LinkedIn. Subscribe to get one tested system per week, directly in your feed.

What to Expect

A 25-minute manual task runs without you while you focus on something else. After the initial setup and one round of instruction refinement, the agent handles the routine collection reliably. Over a month, that is a meaningful block of time returned to deeper work. The quality is also more consistent than copy-paste by hand, because the agent follows the same instructions every time and does not get distracted halfway through a page.

Start Here

Pick one task you do manually in a browser that follows the same clicks in the same order every time. Write each step as a numbered sentence, as if you are explaining it to someone who has never seen the screen. That document is your agent instruction set. Everything else in Copilot Studio builds from it. You can find more workflows like this on AI at Work.