Power Automate Desktop - Simply Explained

0:00 Welcome to the M365 FM podcast, where we explore the people, ideas and technologies shaping the future of work. Power Automate Desktop is a free robot already sitting on your windows PC, and most people never switch it on. It's Microsoft's answer to work. You do the same way over and over in apps that were never built to talk to anything else. You don't write code to use it because you show it the task once and it copies you. The same tool can run while you watch it or while you're fast asleep, and those two versions don't cost the same. Welcome to the M365 FM podcast, where we explore the people, ideas, and technologies shaping the future of work. By the end, you'll know exactly what Power Automate Desktop is, how it works, and what it costs you. What Power Automate Desktop actually is.

0:41 Let's start with the plain version, because the name does you no favors. It's a free windows app that automates the boring, repetitive parts of working on a computer. Microsoft built it to handle all repetitive desktop processes, and that phrase covers more ground than it sounds like it does. The category it belongs to has a scary name, and the name is the only scary part. It's called robotic process automation or RPA, and it simply means software that pretends to be a person at the keyboard. Instead of asking an app for permission, the robot moves, the mouse, clicks the buttons, and types the text. It works exactly the way you would. IBM describes RPA as using software robots to do tasks a human would normally do, and that's the whole idea that matters for a simple reason. A lot of the software you use every day was never designed to connect to anything else. It's an old finance system, or a terminal screen, or a supplier's website with no download button.

1:26 There's no clean way for a modern tool to reach inside software like that. So the work gets done by a human slowly, forever. RPA is the workaround. If a person can see the screen and use the mouse, the robot can too. You might expect a tool like this to demand programming, and it doesn't. Power automate desktop is low code, which means you build by recording your own clicks or by dragging ready made actions into place. You're assembling a task, not writing software. Microsoft designed it for exactly the people who feel shut out by developer tools. Think of someone at home who checks a weather site every morning, or a small business owner copying figures off supplier invoices. There's also the employee typing data into a company system that nobody has improved in years. That last part is the part worth sitting with. It's built for anyone who does rule based work on a machine. And if you're watching this, that probably includes you.

2:08 The question is where the automation actually runs, because that's the difference between this and every other Microsoft automation tool you've met. Desktop flows versus cloud flows. If you've used Power Automate before, you've probably used cloud flows and the name tells you most of it. A cloud flow runs on Microsoft's servers, not on your machine. It watches for something to happen, like an email arriving or a file landing in SharePoint. Then it runs a set of steps across the apps you've connected. Those connections happen through connectors, which are just pre-built plugs that let a flow talk to an app. A standard connector covers the Microsoft family, things like outlook teams, and SharePoint. A premium connector reaches further into systems like Salesforce, SAP, or Oracle, and it costs more to use. The flow reaches those apps through an API, which is the official doorway a program uses to talk to another program.

2:53 When an app has a good API and a good connector, everything is clean and fast. The real tension here is what happens when there's no door. Plenty of software has no API and no connector, so a cloud flow simply can't reach it. It can't log into your ancient accounting package, and it can't click through a supplier's website that was built in another era. That's the gap. Desktop flows fill a desktop flow runs on a windows machine, not on a server, and it drives the mouse and keyboard through the actual screen. It doesn't ask the app for permission. It acts like a user because as far as the app can tell, it's one. Put the two side by side and the picture gets clear. A cloud flow is the polite guest who walks in through the front door and uses the intercom. A desktop flow is the worker who climbs in the way a person does through the screen. When there's no front door at all, one talks to systems, the other uses them by hand, just faster and without getting bored.

3:39 You might think one is the better choice, but they're not rivals. They're two halves of the same product, and Microsoft built Power Automate to cover both the clever part and the part will come back to is that you can wire them together. The cloud half watches for a trigger, and the desktop half does the clicking. How the recorder turns your clicks into actions. So you know what a desktop flow is. Building one comes down to two paths. The first is the actions library, a long list of ready made steps you drag into place. The second is the recorder, and the recorder is where most beginners start because it feels like cheating. The recorder works the way you'd hope you tell it to start recording. You do the task once and it watches. It tracks your mouse and keyboard activity in relation to the on screen elements you touch, and it captures each action separately.

4:21 When you stop recording, it turns everything you did into a list of steps you can edit. You did the job once, and now the flow knows how to do it forever. Each of those steps is called an action, and it's worth knowing what one looks like. An action is a single move, like open this file or click this button or type this text into this field. A desktop flow is just a stack of actions run in order top to bottom. Once you see it that way, the whole tool stops feeling mysterious. This is also the point where you realize you've been doing a robot's job by hand for years. The recorder has a smart trick underneath it. To click a button reliably, it needs to identify that button. It does that through the Accessibility Framework's windows apps used to describe themselves. The modern one is called UI automation, and it's the recommended choice for most apps because it gives the robot detailed information about each element.

5:05 The older one is called a Microsoft Active Accessibility, and you reach for it with legacy apps that never exposed the modern details. In plain terms, the robot learns what a button is, not just where it sits, so it can still find it if the window shifts. Sometimes an app won't cooperate at all, and it refuses to describe its own screen. For those stubborn cases, there's image based recording, which is the fallback of last resort. Instead of reading the app's structure, the robot looks at pictures of the screen and uses OCR, which reads text out of an image to find what it needs. It's clever, and it's also the most fragile option because it depends on the screen looking the same every time. There's an honest caveat worth hearing now, and it's better you know it early. The recorder gives you the backbone of a flow, not a finished robot.

5:45 It can't record loops or conditional logic that do this again for every file. Or if this, then that steps. So you'll add those by hand. It also tends to capture a few redundant moves that you'll want to clean up, and that's normal. Think of the recording as a first draft and the flow designer as where you finish the job attended versus unattended. Now that you can build a flow, the next question is who presses go? And this is where the money lives. There are two modes, and the names tell you everything. If you listen closely, attended means a person is there. Unattended means nobody is unattended mode. You start the flow and it works alongside you on your own PC. Microsoft calls this a flow running with human supervision, and the thing doing the work is called an unattended bot. You hit go, you watch it fill in the form, and you stay in control.

6:26 There's even a version called picture in Picture, where the automation runs in its own little window that copies your desktop, so you can keep working on something else while it does its job beside you. You can try this mode today because the free app supports it. Unattended mode is the one people get excited about, and for good reason nobody is there. The robot signs into the machine on your behalf, does the work and signs out when it's finished, all while the screen stays locked so no one can see it running. This is the runs while You sleep version, the one that turns a nightly chore into something that simply happens. It's also the version with the stricter rules and they catch people off guard. The first rule is that unattended needs a paid plan, which we'll get to in a moment. The second rule is about the machine itself, and it's the one that trips up beginners on windows ten and windows eleven.

7:08 An unattended flow can't run if any user session is open, even a locked one. The machine has to be genuinely free. There's also a firm line on permissions, because unattended flows can't run with elevated administrator level rights. So anything that needs that kind of access has to be handled differently. There's one more quirk worth knowing, because it explains plenty of mysterious failures. When an unattended flow runs, it creates its own remote desktop session, and that session might have a different screen resolution than the one you had when you built the flow. If the screen is smaller than expected, an element the robot needs can fall off the edge and the flow fails. It's a small technical detail with big consequences, and knowing it saves you an afternoon of confusion. Connecting desktop flows to cloud flows.

7:49 Earlier I promised we'd come back to the clever part. So here it is. A cloud flow can start a desktop flow, and it does it with a single action called Runner Flow built with Power Automate for desktop. That one step is the bridge between the two halves of the product. That matters because it lets the cloud do what the cloud is good at, and the desktop do what only the desktop can. The cloud flow sits and waits for a trigger, whether that's a new email, a file dropped into SharePoint, or simply a certain time of day when the trigger fires, the cloud flow hands off to the desktop robot, which does the clicking that no connector could ever manage. The cloud watches the door, and the robot does the work inside. The two sides can also talk to each other while they run, which makes the whole thing far more useful than a simple hand off. A desktop flow can have input variables which let the cloud flow handle information like the name of the file to process.

8:33 It can also have output variables which let the robot hand a result back to the cloud so the cloud flow can pass. In today's report, the robot can fill it in, and the cloud flow can take the finished file and email it to the right person you choose attended or unattended right there when you set up that action. So the same cloud flow can drive a robot that works beside you, or one that works alone. A couple of practical numbers are worth keeping in your head. The connection can fire up to seventy desktop runs a minute, which is far more than a beginner will ever need. There's also a two megabyte limit on the information you can pass into a desktop flow, which is plenty for file names, IDs, and small bits of text, but not for shoving whole documents through what it costs. Let's talk money, because this is where a lot of people either relax or walk away. The good news comes first because the free tier is real.

9:14 The free license lets you build cloud flows with standard connectors and run desktop flows locally in attended mode on your own machine at no cost. If you want to automate a personal task and watch it happen, you can start today and pay nothing. There's a trap here, and it's worth pausing on if you already have a Microsoft three. Sixty five subscription, you might assume that covers everything, but it doesn't. The Microsoft three. Sixty five plans the ones most workplaces use include cloud flows with standard connectors only. They don't include desktop RPA bots, and they don't include premium connectors. So the moment you want the robot to run unattended or reach a premium system, you've crossed into paid territory before you worry about the price, there's a cheaper path than you'd expect. The paid tiers are simpler than they used to be, and the twenty twenty six list prices look roughly like this.

9:56 Microsoft changes them, though, so check the live pricing page before you buy. Power Automate premium costs about fifteen dollars per user per month, and it adds premium connectors plus attended desktop automation. Power Automate process costs about one hundred and fifty dollars per bot per month, and it gives you one unattended bot. Hosted process costs about two hundred and fifteen dollars per bot per month, and Microsoft hosts the machine for you so you don't manage any hardware. If those numbers feel like a lot for a personal project, there's a middle path. Microsoft offers a ninety day trial that includes unattended runs, which means you can test the sleep through it version before you commit a single dollar. That's the sensible way for a beginner to find out whether unattended automation is worth it for their own work.

10:36 The rule of thumb to carry away is short attended is the cheap on ramp, and you can start it right now. Unattended is where the real bill lives, because someone has to pay for the machine and the bot that works without you. Neither is wrong. They're just built for different jobs. What people actually use it for costs only make sense once you can picture the payoff. So let's make it concrete. The examples that come up most often are all boring. And that's the point. Boring is exactly what a robot is good at. Picture the person who pulls data off a website every morning and drops it into Excel. Maybe it's competitor prices. Maybe it's a supplier's stock levels. With a desktop flow, the robot opens the site, reads the numbers, and writes them into the spreadsheet, and you get your mornings back. Microsoft names this exact case, extracting data from websites and storing it in Excel files as one of the core jobs for desktop flows.

11:20 Then there's the invoice crowd. A small business owner gets a stack of supplier invoices as PDFs, and has to type the details into a finance system that has no connector and never will. That's slow. It's error prone, and it's the same every time a desktop flow reads the invoice and types the details in the way a patient assistant would. All day long, there's the quiet, everyday housekeeping to a flow can tidy up your folders on a schedule, moving documents into the right places, and renaming files so you can actually find them later. And for the enterprise crowd, there's the classic example of typing data into an old ERP system one screen at a time, because that system will outlive us all and has no other way in. The advice that matters more than any of those examples is to start small. Pick one task you do every day.

12:02 Something that takes five minutes and never changes, and automate that. First. Don't begin with a company wide process. The people who fall in love with this tool are the ones who save themselves a daily annoyance, not the ones who tried to boil the ocean on day one. There's one weakness that takes down more automations than anything else, and it's coming up next where it breaks. Now let's look at the honest part, because no tool is magic and pretending otherwise would waste your time. This one has a specific weakness, and if you understand it, you'll be ahead of almost everyone who tries it. The number one killer is the selector, the bit of information the robot uses to find a specific element on screen. When the app you're automating gets an update and a button moves or its name changes, the selector can stop matching and the robot can no longer find what it's looking for. It's the most common reason unattended flows fail overnight, and it has nothing to do with your skills.

12:47 The fix is to build selectors that are tolerant rather than precise, and to use proper wait conditions instead of fixed pauses. So the robot waits for the screen to actually be ready instead of guessing. The unattended mode has its own set of traps, and they all come back to that locked down machine. No signed in user can be present, no administrator rights are allowed, and the resolution has to line up with what you built against. None of these are deal breakers, but each one is a box you have to tick before an overnight run will work. The recorder has blind spots too, and they're worth knowing before you rely on it. It can miss steps that happen in the windows start menu or the system tray, so those may need to be added by hand. And when you use image based recording. A click can sometimes land in the wrong spot if the screen doesn't match the picture closely enough, which you fix by loosening the match tolerance in the settings.

13:32 When something does go wrong, you're not left guessing. For unattended runs, power Automate can capture video logs, which record the last 60s of screen activity before the failure. That means you can actually watch what happened. Instead of reading a cryptic error and hoping for the best. Remember this one thing because this tool automates the screen and screens change. So expect to maintain your flows, not just build them once. That sounds like a downside, and for some tasks it is, but for the repetitive, unchanging jobs that eat your week, a little upkeep buys back real hours. The recap. So that's the whole map. Power Automate desktop is the free robot on your windows PC. Desktop flows handle the screen while cloud flows handle the apps the recorder builds the first draft attended or unattended, decides who's there and the licensing decides the bill.

14:17 If you want more of this, subscribe to M365 FM, where we make the tools and the mindset simple. If you could handle one task on your own computer to a robot tomorrow, which task would you choose?