How to automate repetitive tasks on Windows with AI

Start with the job

Write the job down as steps, then ask three questions. Does it run the same way every time? Does it need reading and judgment, like deciding what an email asks for? Does it need to start on its own, at a set time or when something happens? The answers point to the right tool.

Option 1: Power Automate for desktop

Microsoft’s Power Automate for desktop builds “desktop flows”. According to Microsoft’s documentation, you make a flow from prebuilt drag-and-drop actions or by recording your own steps, then run it later, and a flow can work with apps through their on-screen controls, images or screen coordinates.

Good for: the same steps every time, in apps whose screens don’t change much, without writing code.

Harder when: the steps change from one run to the next, or a step needs judgment rather than a fixed rule.

Option 2: A script

A PowerShell script, or code in another language, can move and rename files, call programs and work with apps that offer a command line or an API. Windows Task Scheduler can start a script at a set time.

Good for: precise, repeatable work you can describe in code. For fixed, repeatable work, a script doesn’t need an AI model each time, so it often runs quickly and cheaply.

Harder when: the job lives in an app’s screens with no way in from code, or you don’t want to write and maintain code.

Option 3: An AI assistant that uses your computer

You describe the result in plain words, and the assistant reads the screen and does the steps. This is called computer use.

Good for: jobs that differ a little each time, need reading and judgment, or cross several apps, without building a flow first.

Trade-offs: it keeps asking an AI model what to do next (one answer can cover several actions). Those requests take time and may cost money, and how much depends on the job and connection. It can make mistakes, so check important results. And it has to see the job’s windows to work in them.

Where Solen for Windows fits

Solen for Windows is the third kind: an AI assistant that works through jobs with many steps on your Windows PC. You ask in plain words; it works in the job’s windows, asks before risky steps, and ends by saying what it did.

An example from our test bench: add the two action items from an email as tasks, mark one of them done, then reply to the sender and send it. That job finished in 4 of 4 runs (two AI models, two runs each, pre-release build); one model ended with Have a look rather than Done even though the work was right. Another: in a list of 60 tasks, mark one named task done, past the first page. Finished in 4 of 4 runs.

By default a request can run for up to 30 minutes, and the clock pauses while Solen waits for your answer.

Its limits today

• Windows 11 only, so far. The Beta is for 64-bit Windows 11. It has been tried on Windows 11 only: Windows 10 and ARM PCs are untested.
• It uses your mouse and keyboard. Scrolling, dragging, keys and some clicks use your real mouse and keyboard, so let it work while it works. If you start typing or clicking, its typing stops. A cursor of its own, so it can work beside you, is in development.
• It asks before risky steps. Sending to other people, paying, deleting for good, and security or access changes wait for your yes, unless your request asked for exactly that step.
• It runs on online AI models. Through Solen, your ChatGPT plan, or your own OpenAI, Anthropic or OpenRouter key; not on a model on your PC. Your request, text from the job’s windows and, when needed, pictures of those windows are sent to the model.
• You start each task. Solen does not yet watch for events such as a new file arriving. For a job that must start on its own, a script or Power Automate fits better today.
• Very long jobs aren’t there yet. The lab notes show where it stands.

A simple way to choose

They mix well: a script for the fixed part, and a plain-words request for the part that needs judgment.