Welcome to the M365 FM podcast, where we explore the people, ideas and technologies shaping the future of work. What Copilot and Power Automate really is copilot in Power Automate isn't the tool most beginners think it is. You don't have to build flows by hand or write any code before it starts helping you. You describe the automation you want in a normal sentence, the way you'd ask a coworker for a favor, and it drafts the trigger and the steps for you. It lives in four places inside the product, and it has real limits that nobody should hide from you. By the end, you'll know exactly where it helps and where it stops. So let's start with the plain version. Copilot in Power Automate is the AI helper built into Microsoft's automation tool, and its whole job is turning your words into automation. You don't buy it separately and you don't install anything new. It sits inside Power Automate in a chat pane, waiting for you to describe what you want to happen. A flow, in plain terms, is a set of steps that runs on its own when something happens, like a new email arriving or a file being uploaded, you can still build one by hand if you enjoy that. Copilot is simply there for the moments when you'd rather describe the outcome and let it handle the first draft, and nothing forces you to use it. The engine behind it comes from Azure Open AI service, which is Microsoft's AI service in its Azure cloud. That same foundation runs under other Microsoft Copilot experiences. So this isn't some side experiment. It's the same family of technology pointed at a different job. One thing to know upfront is that copilot is built for English first. Support for other languages is still limited, so parts of it may answer in English even when your own language is something else. Now it shows up in four places and knowing them saves you confusion. Later you'll find copilot in cloud flows, which is where you build automations that run online. There's also one in desktop flows for automations that click through apps on your own computer. The other two places are less obvious. There's copilot in process mining, which helps you understand how a business process actually runs. And one sits in the automation center answering questions about the automations you already have. The one most beginners meet first is the cloud flow designer, and that's where we'll spend most of our time. Open a flow and copilot appears as a chat pain on the right side of the screen. From there, it can create a brand new flow from scratch. Edit an existing one. When you ask, answer questions about the flow in front of you and even write a description of what the flow does. Think of it less like a robot that replaces you, and more like a coworker who has already read the manual. It builds the first rough draft while you watch, and you stay in charge of whether that draft is right. How you talk to it and what happens next. The way you talk to copilot matters more than almost anything else, and Microsoft's own guidance gives you the pattern that works best. Say what should happen and then say what you want it to do. When this happens, do this. And that's the whole shape of a good prompt. A weak prompt sounds like I want to process an email. Copilot can't do much with that because it doesn't know what processing means to you. A strong prompt sounds like when an email arrives, post the subject of the email to the Contoso teams channel. Now it knows the trigger, the action, and the destination all at once. Here's what actually happens after you type that, you start on the Power Automate Home page. You pick the option to create with copilot and you describe your flow in the box. Copilot reads your sentence and proposes a structure which is a trigger plus one or more actions. A trigger is the event that starts the flow and actions are the steps that run after it. If the draft looks right, you keep it and move on. If it doesn't, you add more detail in plain words and it rebuilds around what you said. Then it checks the connections for you. A green checkmark means a connection is ready, and a red exclamation mark means it needs your attention before the flow can run. This is the part that surprises people. Copilot sets up those connections automatically where it can. So you often get a working automation faster than you'd expect. You press create, the designer opens and the chat pane is right there with you. Once you're inside, you can keep talking to it. You can say send the email subject to teams instead, or explain what this flow does or delete that action. That explained feature is quietly powerful because it will walk you through a flow that someone else built and left behind, step by step in words you can follow. It also answers product questions like how to reach a child flow, or where a particular setting lives so you can get unstuck without leaving the screen you're already working on. If you'd rather not invent a prompt from nothing, Microsoft keeps a library of ready made prompts. You can borrow and adapt, and that's only one side of it. The same assistant can also look at your automations after they run, and plenty of people never discover that part at all. One real beginner task, from a blank page to a working flow. Let's follow one real task from a blank page to a finished flow, because this is where it stops being abstract. Imagine you handle onboarding, and every Monday a batch of new hires should get a welcome email. The names and email addresses already live in a SharePoint list, which is a shared table of information inside Microsoft three hundred sixty five. You open copilot and you type something like every week, send an email for each item in a SharePoint list. Copilot comes back with a draft. The trigger is a weekly schedule, say Monday at nine in the morning. The first action reads the items from your SharePoint list, and the second action sends a personalized email to each person on that list. You point it at your actual SharePoint site, you pick the right list, and you adjust the wording of the email. Then you save it and test it once before you trust it. That's a working onboarding automation, and you built it by describing it in a sentence. If one step looks off, you fix that single step and you can switch the whole flow off later with one toggle if you ever need to. Now let's do a second one, because the pattern repeats. And that's the point. Suppose you want your team to know? When someone drops a new file into a project folder in OneDrive? You type send an email to the team whenever a new file is added to the Project Documents folder in OneDrive, including the file name and a link. Copilot builds that, and then you refine it with one small follow up. You say instead of email, post a message in teams with the file link, it updates the flow. And now your team gets a heads up in the channel they already live in without you touching a single setting by hand. That's the payoff. And it only took two sentences from you. Notice what you didn't do. You didn't describe the entire process in one giant paragraph, and you didn't try to get it perfect on the first try. You gave it one clear step. Check the result and then give it the next step. That habit is the difference between a flow that quietly works and a mess you have to untangle later if you want to try this yourself. Start with the smallest possible thing. Pick a task you repeat every week. Describe just that one task and let copilot show you the shape of it. You'll learn more from one small flow than from an hour of reading about big ones. The other half copilot that reads your automation for you, building flows is only half of what this assistant does, and the other half lives in the automation center. This is where copilot stops helping you create and starts helping you understand. You ask questions in plain language and it answers them. It handles four kinds of questions. It can tell you about your cloud flow runs, which are the online automations. The same goes for desktop flow runs, which are the ones that act on a computer. The last two are your work queues, which are lists of tasks waiting to be processed, and general questions about how Power Automate works. So you might ask how many flows failed yesterday, or show me the most frequent run errors last month? Or which machines had the most failures today? Instead of clicking through reports and filters, you just ask and it answers with a number or a chart. Under the hood, something interesting happens. Copilot turns your question into a database query, and then it chooses the best way to show you the result, like a table or a bar chart or a pie chart. You can even open that query, edit it, and run it again if you want to change what it looks for. If you already know how you like to read an answer, you can ask for it plainly. Like show me that as a bar chart and it will shape the result to match. There are two honest catches here. This side of. Copilot currently needs a work or school account with a power automate environment in the United States, and it only works in environments based there. And in a chat. It only remembers the last ten questions you asked. So a long conversation slowly drifts away from where you started. It can also return data that's incomplete or plain wrong. So when a number really matters, open the query it used and check it yourself. Still, for anyone who has stared at a wall of run history and felt lost, being able to ask what went wrong last month in a sentence is a real shift. It's the difference between digging through logs and simply asking a question, and that shift points at something bigger. The same AI that helps you build a flow also has to live inside Microsoft's wider world of tools, licenses, and cloud services. And that's where the last piece of the picture sits. What copilot can't do yet. Most explanations skip this next part, and it's the one that decides whether you waste an afternoon, Microsoft says. Being upfront about the limits saves you time. So here they are. Straight copilot can't debug a runtime error. It has no access to your run history, your error logs, or the context of what happened when a flow ran. So it can't tell you why a specific run failed. That's a hard wall, not a small gap. It also can't read your data. It doesn't look inside your SharePoint lists, your Dataverse tables, or your SQL databases. It can't confirm that a column name exists or that a value is correct. So if you ask it what's inside your data, it has no way to know. No matter how clearly you ask, it can't fix your connections either. Expired passwords, permission problems, and gateway issues all sit outside what it can touch. And those need a person or sometimes an administrator to sort out. To be fair to it, copilot does keep itself on topic. It filters out questions that have nothing to do with Power automate or the flow you're editing, and the experience warns you when you're close to something, it can't safely change. There's more to the list than that. Copilot struggles with deeply nested logic, which is when one condition sits inside, another condition inside another one. It also can't see across flows. So if your flow calls a child flow, it treats each one as if the other didn't exist. It can't reliably write complex expressions, especially the kind that stack several functions together or reach back to data from many steps ago. And it only works with a subset of the connectors that Power automate supports? One more practical limit is worth remembering. Copilot is only available in the new designer, not the older classic one. So if your screen looks different from a tutorial, that mismatch is usually the reason. Now some of that is already changing and that should make you feel better rather than worse. Microsoft's own advice is that for some jobs, like reading a confusing error message or building a complicated expression, you should paste it into a general AI tool today and bring the answer back into Power Automate. That isn't a knock on Copilot's quality. It's about scope. Copilot is designed for editing the flow you're working on, and general tools are built for broad knowledge, and these limits are described as temporary. Microsoft says the team is working on giving copilot access to run history and error details, which would close the biggest gap on this list. The prompts that waste your time. Most of the frustration people hit with copilot comes down to how they ask, not what the tool can do. So let's walk through the prompts that quietly waste your time and then at what to say instead. The first one is simply fix my flow. It feels natural, but copilot can't see the error and doesn't know what fixing means to you. It will guess, and the guess is usually wrong. The second is, why did my flow fail at three in the afternoon yesterday? That question needs run history and inside the designer copilot doesn't have it. The third mistake is a long run on instruction. You type. Add a trigger, then get the item details. Check if the status is approved. Send an email to the manager. Wait for approval. Update the item and post to teams. It sounds efficient, but copilot handles single steps well and loses accuracy across a long chain. Break it apart and the results get noticeably better. The fourth mistake is asking about your own data. Like what columns does my SharePoint list have? It can't look. Check the list yourself. Then tell copilot the exact field names to use and it will build around them. So let's turn that around and look at what actually works. Be specific and give it one request at a time. Start your sentence with a verb like add, change, remove, explain, or move. Because those map onto things it can really do. Name the fields and the values so it doesn't have to guess what you meant. Then look at what it built before you save, Dave, especially the parts that pull in data from earlier steps. One habit. Ties all of this together. Save your flow before you let copilot make big changes so you always have a version to fall back to if the result isn't what you wanted, treat saving like a checkpoint and you'll never lose good work to a bad edit. Get this habit right and copilot starts to feel like a shortcut. Get it wrong and it feels like it's fighting you the whole way. And if it gets something wrong, use the feedback buttons under its answer. That feedback goes back to the team, so a bad result isn't wasted. It's one small way you help make the next version a little sharper. Where it fits in Microsoft three hundred sixty five and Azure. Let's place this in the bigger picture, because that's where a lot of beginners get lost. The AI running inside Copilot and Power Automate is the same Azure OpenAI service that sits under Microsoft's other copilot tools, and it runs on Azure. Microsoft's cloud platform. Power automate itself is part of the power platform, a family that also includes power apps power BI, and Copilot Studio. One confusion is worth clearing up early because these are two different assistants. Microsoft three hundred sixty five copilot lives inside word, outlook and teams where it helps you write and summarize. The one inside power Automate is for building an. Understanding flows. They share the same Azure foundation, but they live in different rooms and they do different work. They do connect, though. Flows can be used as plug ins inside Microsoft three. Sixty five Copilot in preview. So an automation you build can show up inside the assistant you already use every day. Then there's cost. And honest detail matters more here than any slogan. Power automate premium lists at fifteen dollars per user per month, and it covers cloud flows, attended desktop flows, and the basics of process mining power Apps premium lists at twenty dollars per user per month, premium connectors and unattended bots cost more on top of that, and that's usually where surprise bills come from. One time sensitive change is worth knowing if you're planning ahead. The AI builder credits that used to come bundled with premium licenses end on the first of November in twenty twenty six. After that, AI heavy automations move onto copilot credits, which you buy separately. If your automation leans on AI features, that's a budget line to keep an eye on. For a beginner, though, the reassuring part is simple the copilot experience rides on the Power Automate license you already have, so it isn't a separate purchase that makes it worth checking which plan your organization already owns before you assume you need something new. Start with a free trial or the developer plan. Build something small and see how it feels before you spend anything. The short recap and one question for you. So here's where we've landed. Copilot in Power Automate turns plain sentences into working automations. It turns up in four places across the product, and the cloud flow designer is where most people start. It can build a flow from a description, explain a flow. You didn't write and answer questions about your runs in the automation center. The honest edge to carry with you is simple. It can't see your data, your errors, or across your flows. So you stay the builder and it stays the first draft helper. That's not a weakness to fear. It's just where the line sits. And knowing where the line sits is what stops you from blaming yourself when a prompt comes back wrong. Now, I'd like to leave you with one question. What's the one task in your week that you repeat by hand every single time? And what would it look like if you simply described it out loud? If this made the topic feel less intimidating? Subscribe to M365 FM to master the tools and the mindset shaping the future of work.