Power Platform has been around for quite some time now and Power Automate has become a powerful and popular tool for automation, used for a wide variety of processes. But after all the Artificial Intelligence (AI) hype started a few years ago and with all the improvements to agents in the last couple of months, many times we have heard the sentence “let’s build an agent for this” or the question “can we use Copilot for this process?”. But it’s important to know that Copilot / AI is not a replacement for Power Automate or the other way around. Choosing the right approach is key in the digital transformation journey of all organizations.
In this article, we’ll try to answer the so-called “an Agent or a Flow” question. We will explore the differences between agents and flows, common use cases for each, and how they can work together to create end-to-end automation.
Power Automate and flows
Power Automate is Microsoft’s official tool for Process Automation and it’s part of Power Platform. It can be used to automate both simple tasks or end-to-end processes. As a first step in the Agent vs Flow question, let’s define what Power Automate and flows are.
What is Power Automate?
Microsoft’s official definition for Power Automate is as follows:
“Power Automate allows you to optimize your business processes across your organization and automate repetitive tasks. This service helps you create automated workflows between your favorite apps and services to synchronize files, get notifications, collect data, and more.”
Some key points can be derived from above definition:
- Power Automate is a tool to create automated workflows.
- Power Automate acts as a process orchestrator and connects multiple steps.
- Power Automate can connect to systems and apps that can reside inside the Microsoft 365 and Azure ecosystem (SharePoint, Outlook, Teams, Blob Storage) or completely exist outside of it (Hubspot, Gmail or SAP).
We have a complete blog on Power Automate if you’re interesting in learning more on Power Automate’s capabilities and features.
Flows: Logic and Structure
A flow is an automated process that performs steps or actions, it’s based on a trigger (i.e., something needs to happen for the process to start) and it can include conditions, loops, variables and all the “traditional” concepts of programming. It is usually designed to handle repeatable, rule-based tasks with little or no human intervention.
Flows usually:
- Run based on structured logic (inputs, pre-defined conditionals, set of steps)
- Uses rules
- Deterministic outcome (e.g. fill out a template, create a new client, notify someone)
As a summary, flows are a series of steps that follow logic and structure and usually have deterministic inputs and outputs.
Typical use cases
Based on above, Power Automate is ideal for automating structured, repeatable processes that follow clear rules. Some examples are:
- Sequential approvals. Power Automate is the perfect tool to handle approval requests when new projects are created. The approval assignment can be based on rules such as: Company, Cost Center and Project Amount.
- Customer Onboarding automation. Customer onboarding is usually a process with many steps and requirements and it’s usually rule-based: creating new customers in 3 different systems, creating a contract template based on conditions.
- Data entry automation. Tax forms are standard every month and there are specific rules to follow to fill out, so Power Automate comes in really handy as the perfect tool for this process.
It’s key to mention that Power Automate can also integrate Artificial Intelligence (AI) features inside workflows. For example, we can create a workflow that starts every time a new client signs a contract in DocuSign and use an AI service to read the PDF content and based on the results of this action, send an approval request (rule-based) or fill out an Excel template (deterministic outcome). This doesn’t mean the workflow is an agent, it just means it’s using AI features to perform some parts of the automation.
Copilot Studio and Agents
Now that we know what Power Automate is, let’s talk about the other side of the equation: Copilot Studio and Agents, but before getting into deep waters, let’s clarify that Copilot Studio and Copilot are different products.
Copilot (without “Studio” in the name) is the ready-to-use AI assistant built by Microsoft, that can be used in all the apps we already know: Outlook, Excel or PowerPoint. The little icon bugging us everywhere, offering help. Copilot Studio is the tool to build our own AI agents. In this blog, we’re referring to Copilot Studio.
What is Copilot Studio?
Microsoft’s official definition for Copilot Studio is as follows:
“Copilot Studio is the AI agent platform for Copilot, where you can build agents that help do work. Effortlessly connect agents to your organization’s data, tools, and systems to multiply capacity and accelerate outcomes. Build agents for any scenario—from simple to sophisticated—using natural language and publishing to where you need them.”
A lot to process from the official definition! But here is a list of key points:
- Copilot Studio is the tool to build our own “Copilots” or our own “ChatGPTs”
- With Copilot Studio we can build agents that can do work for us
- We can connect agents to data and multiple systems
Copilot Studio (and any other tool for building agents) leverage large language models (LLMs) to receive an input or a question and derive an answer or plan actions based on the input.
You can also read our full blog on Copilot Studio fundamentals.
Agents: Context and decision-making
Since agents leverage LLMs, an agent can interpret natural language and choose an appropriate (or better said, what the model thinks is appropriate) next step based on the information available.
When we build agents in Copilot Studio, we usually provide knowledge sources (e.g. an HR Policy document or a SharePoint site with standard procedures for expense reports), tools (e.g. Send emails) and data (e.g. access to a Database). All this information provided helps the agent understand context, make decisions, and act on behalf of a user.
Based on above, agents usually:
- Work based on unstructured logic (natural language, prompts, instructions)
- Use context to decide, plan or answer
- Acts based on Intent-based and may have dynamic outcome
Typical use cases
Copilot Studio is ideal for scenarios that require context and decision-making. Some examples are:
- Employee self-service for HR processes. Answering questions about maternity leave or vacation policies or guiding through the process of submission and approval of days off.
- Expense reports. Agents can guide employees through submitting expense reports, validate required information, answer policy questions, and trigger approvals or reimbursements automatically.
- Categorize support requests and route accordingly. An agent can have a conversation with a user that is having a technical issue and provide quick fixes based on general guidelines but based on the type, urgency, or content of the request, the agent can direct and escalate it to the appropriate team and create support tickets.
AI Models vs Agents
It’s very important to also learn the difference between AI Models and Agents. An AI Model can perform a wide range of tasks, but these are usually isolated tasks. For example, an AI model can read a PDF document, summarize text or classify a text as positive or negative.
AI Models do not have memory and do not take any action. Agents, on the other hand, as we have discussed, use AI as part of a broader system that includes context and memory. AI Models can be used as part of a Power Automate workflow or as part of Agents.
Copilot Studio + Power Automate
The features and capabilities of Copilot Studio agents and Power Automate flows can be combined and Microsoft has been releasing a lot of new features that leverage the two tools at the same time. And in fact, many of the Power Automate options we see from the portal can be used from Copilot Studio.
Autonomous agents
We think of agents as a conversational tool: someone has a question or needs to submit a support request, so they open Teams, SharePoint or whatever channel has been selected for the agent and start a new chat. But did you know agents can also “start” actions on their own?
Using Copilot Studio triggers, we can set up different events that will automatically activate an agent when the event happens.
For example, let’s say we have an agent that receives IT support tickets from users in a chat, the agent has been trained with guidelines and policies and works just fine. But let’s assume that we also have an email address (servicedesk@yourcompany.com) where users can send emails asking for help.
We can connect the service desk mailbox to the very same agent that was trained for the chat, and it will use the very same instructions, tools, data and knowledge and simply act whenever a new request arrives through email.
This feature converts an agent that is only conversational to an agent that is also autonomous, and behind the scenes is using Power Automate workflows and triggers but we don’t need to create separate automations, it’s all done automatically inside Copilot Studio.
Agent flows
Agent flows are just flows that are created from Copilot Studio and they are used to connect agents with automated actions. Agent flows are just the Power Automate interface used inside Copilot Studio.
Agent flows can receive input data from, perform one or many actions and return a response to the agent.
Below is an example of a workflow built for a Copilot Studio agent that receives an Order ID from the customer, then queries this order ID in a Dataverse database and updates its status.
How our Power Automate and Agenting automation services can help
Using our Power Platform consulting services, we have designed and implemented many solutions combining Power Platform, Copilot Studio and AI features, covering many scenarios and business cases.
Whether you have some ideas in mind or if you’d just like to have a conversation to learn what we’ve done on intelligent automation, contact us, we’ll be more than happy to have a conversation and see how we can be of help for your team.
