Power Automate vs AI Agents: Rules Where They Fit
Fixed Steps Stay as Rules, Judgement Steps Get an Agent
Use Power Automate for any step that has one right answer, and use an AI agent only for the steps that need judgement. A flow routes the approval, copies the field and files the document the same way every time, with a run history you can audit. An agent reads the messy email or picks the likely cost code, and a person approves its call until it has proven itself. Most processes need both, so the decision is made step by step.
The rest of this guide is for the person who has to make the call in a Microsoft 365 business: an IT manager, a business systems lead or an operations manager who already runs Power Automate flows and is now being asked about agents.
The pressure usually arrives from one of three places. A board member has seen a Copilot Studio demo. A vendor has offered to "replace your flows with agents". Or a flow that worked for years has started failing on inputs it was never built for, and someone suggests AI will cope better.
Working out whether this fits your business? Book a free 30-minute review
Why Rebuilding Every Flow as an Agent Goes Wrong
Gartner predicted in June 2025 that over 40% of agentic AI projects, its term for AI agent projects, will be cancelled by the end of 2027, citing escalating costs, unclear business value and inadequate risk controls (Gartner, June 2025). The same release warned about "agent washing", where existing assistants, chatbots and robotic process automation are rebadged as agents.
A wholesale rebuild runs into all three of Gartner's causes at once.
On cost, a flow that moves a form submission into a SharePoint list costs almost nothing to run. Putting a language model in the middle of that step adds a charge per call and gives you a slower, less predictable result for a job that never needed reasoning.
On value, if the flow already does the job, an agent cannot do it more correctly. It can only do it differently, and "differently" is a defect when the step is an approval limit or a payment run.
On risk, a flow either runs or fails, and the run history shows which. An agent can produce a plausible wrong answer that nobody notices until month end. For any step with a financial, safety or legal consequence, that is the wrong trade.
Microsoft's own documentation draws the same line inside its products. Copilot Studio describes its agent flows and topics as deterministic: given the same inputs, they produce the same output. The generative part of an agent decides which flow or tool to call. The flow itself still does the work by rules.
The Test: Does This Step Have One Right Answer?
Take a process you already run and break it into steps. For each one, ask a single question: if two careful people did this step with the same information, would they always get the same result?
Which tool runs this step?
The first row covers more of your processes than you might expect. Routing by dollar value, notifying the right manager, creating a folder, updating a status, posting to a channel, writing a row to a list: these have one right answer. They belong in Power Automate, where they run the same way every time.
The second row is where many "we need AI" requests actually sit. The step is still rules based, but the input is a document. Pulling the ABN, invoice number and total from a supplier PDF has one right answer; it just needs extraction first. AI Builder or Azure AI Document Intelligence reads the document, the flow checks the confidence score, and anything below your threshold goes to a person. No agent is needed.
The third row is the agent's territory. "What is this customer actually asking for?" "Which project does this vague invoice line belong to?" "Is this email a complaint, a quote request or an order change?" Two capable staff would usually agree, but they get there by reading and using context. A rule tree for these steps grows forever and still misses cases.
The fourth row stays with people. A credit decision, a safety call, a disciplinary matter, a response to a regulator. AI can draft or summarise, and a named person decides.
What a Mixed Design Looks Like in Practice
Consider a typical accounts payable inbox in a business that runs Microsoft 365 and a cloud accounting system. Invoices arrive by email from hundreds of suppliers, in every format. Some are clean PDFs. Some are photos. Some are an email body with no attachment at all.
Most of this process is rules. Only two steps need judgement.
Accounts payable inbox: flow with two agent steps
The flow owns the trigger, the extraction call, the approval routing by dollar limit and the final post into the accounting system. Those steps have one right answer, and they leave a run history.
The agent owns two steps. It decides what kind of document arrived, because suppliers send statements, remittances and complaints to the same inbox. It also suggests the coding, because a line reading "site works as discussed" needs history and context to place.
A person approves the agent's two calls in one screen. Their corrections are logged and become the measure of how well the agent is doing. If you want the accounting side of this in more depth, our guide to AI automation for accounting firms on Xero and MYOB covers the coding and reconciliation steps.
Flows Fail Loudly, Agents Fail Plausibly
The strongest reason to keep fixed steps in a flow is how each one breaks.
A Power Automate flow that meets an input it was not built for stops. The run history shows a failed action, the owner gets an email and somebody fixes it. That is annoying, and it is also safe: the error is visible the same day.
An agent that meets an input it was not built for usually still answers. The answer reads well. It may be wrong. Unless a person checks it, or a later control catches it, that wrong answer moves downstream into a bill, a customer reply or a report.
One process, rebuilt as a single agent vs split by step
| Metric | Whole process as one agent | Flow with agent steps | Improvement |
|---|---|---|---|
| Approval routing by dollar limit | Model decides each time | Fixed rule, same result every run | Predictable |
| When an input is unexpected | Plausible answer, may be wrong | Flow stops and alerts the owner | Visible |
| Audit trail | Conversation log, hard to replay | Run history per step, plus approvals | Auditable |
| Running cost per item | Model call on every step | Model call on judgement steps only | Lower |
| Who can maintain it | Specialist who built the prompts | Any Power Automate maker, plus the agent owner | Shared |
Agents can be run safely, provided each agent step has a control around it: a person approving, a confidence threshold, or a later check against a system of record. A flow gives you that control by default for the steps that do not need judgement.
Where Agents Earn Their Place
Agents are worth their cost on steps with three features. The input is unstructured. The decision depends on context that a rule cannot hold. And a person can check the answer quickly.
Good candidates in a typical Microsoft 365 business:
- Reading a customer or supplier email and working out what is being asked, then routing it to the right queue.
- Suggesting a cost code or project for an invoice line from the history of how your team coded similar lines.
- Answering staff questions from approved policy documents, with the clause shown. Our HR policy assistant guide covers that build in detail.
- Drafting a reply to a routine enquiry for a person to edit and send.
- Summarising a long thread or a set of meeting notes into the actions someone has to take.
Each of those has a person approving the output, at least until the agent has met a standard your team sets. That is shadow mode: the agent runs beside your staff, its answers are compared with theirs every week, and nothing it produces reaches a customer or a ledger until it meets your threshold.
Copilot Studio Agent, Agent Flow or Cloud Flow
Microsoft now gives you three overlapping places to build, and the names do not help.
Power Automate cloud flows
These are the familiar ones. They trigger on an event or a schedule and run a fixed sequence of actions. Licensing sits with Power Automate or the Microsoft 365 plan that includes it.
Copilot Studio agent flows
Agent flows are built in Copilot Studio with the same designer and connectors. They are still deterministic. The difference is that an agent can call them as a tool, and they are billed through Copilot Studio capacity rather than a Power Automate licence. Microsoft has changed these terms more than once, so confirm the current billing with Microsoft or your licensing partner before you design around it.
Copilot Studio agents
Agents hold the conversation or the instructions and decide which tool to call. The judgement lives here.
A practical rule
If a person or an event starts the process and every step has one right answer, build a cloud flow. If an agent needs to trigger a fixed sequence, build an agent flow. Put only the judgement in the agent itself.
If your staff already have Microsoft 365 Copilot, check your permissions before any agent can read SharePoint and Teams. An agent sees what its user can open, so an over-shared site becomes an agent problem on day one. Our Microsoft 365 Copilot readiness and permissions audit explains what to clean up first.
When a Custom Agent Beats Copilot Studio
Copilot Studio fits well when the data lives in Microsoft 365 and Dataverse and the users already have the right licences. Some processes need more than that.
A custom agent built in your own Azure tenant usually suits better when:
- The judgement step needs data from a system Copilot Studio cannot reach cleanly, such as an on-premises database, a line-of-business app or an industry system.
- You need processing pinned to an Australian Azure region for a regulatory or contractual reason.
- The step runs at a volume where per-message billing becomes harder to predict than a compute cost you size yourself.
- Staff without Copilot licences need to use it.
In those cases the flow still owns the fixed steps. It calls your custom agent through an HTTP action or a custom connector for the one step that needs reasoning, gets an answer and a confidence score back, and carries on. Our custom AI development service builds that layer inside your tenant, so the code and credentials stay yours.
For the running cost of that model, our Azure AI services pricing breakdown for Australia covers what to budget for.
How to Move an Existing Process Across
You do not need to touch a working flow to add an agent step. The safest path keeps the flow running as it does today and adds the agent beside it.
Adding an agent step to a process that already runs
Start with the baseline. Before any AI runs, measure how long the judgement step takes today and how often staff correct each other. Without that number you cannot tell whether the agent helped.
Then choose one step. The best first candidate is the one where skilled people spend the most time reading and deciding, with work queuing behind them. In accounts payable that is usually coding. In a service team it is usually triage.
Shadow mode comes next. The agent answers every item, staff keep working as before, and each week someone compares the two. You set the accuracy threshold in advance. Only when the agent meets it does the step move to co-pilot mode, where the agent suggests and a person approves.
At every stage the flow keeps running the fixed steps. If the agent step is switched off, the process falls back to a person and keeps going.
Questions to Ask a Vendor Who Wants to Rebuild Everything
If someone proposes replacing your flows with agents, ask these before you sign anything:
- Which steps in this process have one right answer, and why would an agent do them better than the current flow?
- What happens when the agent is wrong, and how will we know the same day?
- What does each run cost, per item, at our volume?
- Who maintains the agent's instructions after you leave, and can our Power Automate makers read them?
- Where does our data go during processing, and in which region?
- What is the accuracy threshold before anything goes live, and who sets it?
A good answer to question 1 names specific steps. "Agents are more flexible" is not an answer. If you want an independent view on a proposal, that is the kind of review an AI consultant should be able to do in one conversation, without selling you a platform.
Common Questions
Is Power Automate being replaced by Copilot Studio?
There is no sign of that in Microsoft's current product direction. Copilot Studio agent flows use the same designer and connectors as Power Automate, and agents call flows as tools. Fixed, rules-based automation still runs as flows; the agent decides which flow to call.
Can an AI agent run an approval?
It can route one, but the rule should stay fixed. Approval limits, delegations and segregation of duties have one right answer and belong in a flow. An agent can prepare the approval request, summarise the documents and flag anything unusual for the approver.
What is the difference between AI Builder and an AI agent?
AI Builder adds models such as document extraction and classification inside a Power Automate flow. The flow stays in control and checks a confidence score. An agent holds instructions and decides which steps or tools to use, which suits work that needs reading and context.
Are agents more expensive to run than flows?
Usually, per item, because each agent step calls a language model. Flows that only move data cost very little to run. That is why the agent should handle only the judgement steps and leave the fixed steps to the flow.
How do we know an agent step is accurate enough?
Run it in shadow mode against your own staff for several weeks. Compare its answers with theirs each week and track the correction rate. Set the threshold before you start, and move the step to co-pilot mode only when it is met.
Do we need Microsoft 365 Copilot licences to use agents?
Not always. Copilot Studio agents can be licensed through Copilot Studio capacity, and a custom agent in your Azure tenant needs no Copilot licence. Confirm current Microsoft licensing terms with Microsoft or your licensing partner, because they have changed several times.
Find Out Which Steps Need an Agent
Bring one process you run in Power Automate today, or one you are being asked to rebuild as an agent. In a free 30-minute Microsoft Teams call we will go through its steps with you. You leave knowing which steps stay as rules, which one is worth an agent, and what to measure first.
Book your free 30-minute process review
For the wider service, see our process automation page.
Related Reading
- Microsoft 365 Copilot Readiness: Audit Permissions First - The SharePoint and Teams clean-up to do before any agent reads your files.
- AI agents for Australian businesses: why DIY goes wrong - The common failure modes when an agent is built without understanding the process.
- AI agent governance: data access, privacy and human override - Who controls an agent's access, and how human override is designed.
- Shadow AI inventory and policy - What to do when staff are already using AI tools outside your approved flows.
- AI accounting automation on Xero and MYOB - The coding and reconciliation steps in the accounts payable example above.
Sources: Gartner, "Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", press release, 25 June 2025; Microsoft Learn, Copilot Studio agent flows FAQ; Microsoft Learn, AI Builder in Power Automate; Microsoft Learn, Data, Privacy, and Security for Microsoft 365 Copilot.