What is RPA, and how does it compare with AI agents?

What is RPA?

UiPath, one of the best-known RPA vendors, describes RPA as software robots that automate repetitive, rule-based tasks like data entry and system integration. The "robot" is a program that runs on a computer. It works through the same screens a person would use.

UiPath gives examples such as entering data, moving files and processing transactions. In one example, a robot signs in to a business system, creates sales orders, makes invoices and updates stock levels.

What is attended and unattended RPA?

• Attended RPA: The robot runs on a person's computer while they work. The person starts it, often with a click, and it handles a task for them.
• Unattended RPA: The robot runs on its own, on a separate machine, on a schedule or when something triggers it.

Microsoft says unattended desktop flows suit tasks that don't need human supervision. Its unattended flows sign in to a machine, run the task, sign out and report what they did. They need a machine where all users are signed out.

What does RPA look like in a small business?

Here are some common examples.

• Copying orders: Take each new web order and type it into an older accounting system that has no connection to your store.
• Downloading reports: Sign in to a supplier portal every morning, download a report and save it to a shared folder.
• Updating records: Read rows from a spreadsheet and enter each one into a government or insurer portal.
• Moving data between old systems: Copy customer details from one desktop app to another.

RPA is most useful when a system has no API. An API lets other apps read data from a system and send data to it. If two apps have APIs, a tool such as Zapier or Make can usually connect them more simply.

What does RPA cost?

Here is what two large RPA vendors list on their own pricing pages as of October 2026.

Tool	Plan	Price
Microsoft Power Automate	Premium, includes attended desktop flows (RPA)	$15 per user a month, paid yearly
Microsoft Power Automate	Process, for unattended bots	$150 per bot a month, paid yearly
Microsoft Power Automate	Hosted Process	$215 per bot a month, paid yearly
UiPath	Community	Free
UiPath	Basic, up to 2 robots	Starting at $25 a month
UiPath	Standard and Enterprise	Contact sales

Microsoft notes its listed prices are for marketing and may not match the actual list price. UiPath notes that users and robots may need extra licenses.

The license is only part of the cost. Someone has to build each robot, test it and fix it when something changes.

Enterprise deployments cost far more than the entry prices suggest. Independent trackers put a full UiPath rollout, with Orchestrator, AI Center and process mining, at $50,000 to $200,000 or more a year once you add 10 or more robots.

How are AI agents different from RPA?

An AI agent uses an AI model to understand inputs and decide what to do next. Anthropic describes agents as systems where the AI directs its own process and tool use. RPA follows fixed steps that a person designed.

Question	RPA	AI agent
How does it decide what to do?	Follows fixed rules and steps	The AI model chooses its steps
What inputs can it handle?	Neat, structured data in the same format each time	Messy inputs such as emails, PDFs and chat messages
What happens when a screen changes?	It can stop working until someone updates it	It can sometimes adapt, but may also make mistakes
How predictable is it?	Very predictable	Less predictable, so it needs checks
Running cost per task	Low once built	Higher, because each step uses the AI model
Main risk	Breaks when systems change	Confident wrong answers and prompt injection

Where does RPA win?

RPA wins when the task is the same every time, the inputs are neat and you need the exact same result each run. Payroll uploads, report downloads and data copying between old systems are good examples. You can test RPA fully, and it does what you told it.

Where do AI agents win?

AI agents win when the input is messy or needs judgment. For example, an agent can read a customer email, work out what they want and draft a reply. It can pull key details from invoices that all look different. Rule-based RPA can't do that well on its own.

What are the risks with AI agents?

AI agents can make mistakes that look correct. They can also be tricked. OWASP, a security group, warns about prompt injection, where hidden text in a website or file changes what the AI does. OWASP advises giving the AI only the access it needs and having a person approve risky actions.

Anthropic adds that the freedom agents have brings higher costs and a chance that errors build on each other. It recommends the simplest solution that works.

Can RPA and AI agents work together?

Yes, and that mix is often the best setup. The AI handles the step that needs judgment. RPA or a fixed workflow handles the steps that must be exact.

Here is an example for an insurance broker.

1. An AI step reads a client email with an attached claim form and pulls out the key details.
2. A person checks those details on a review screen.
3. An RPA robot types the approved details into the insurer's portal, which has no API.
4. The system saves a confirmation and emails the client.

The AI does the reading. The robot does the typing. The person checks the one step where a mistake would cost money.

Is "agentic AI" just RPA with a new name?

Sometimes it is. In June 2025, Gartner warned about "agent washing," where vendors rebrand older tools such as chatbots and RPA as agentic AI. Gartner estimated that only about 130 of the thousands of agentic AI vendors are real. It also predicted that over 40% of agentic AI projects will be canceled by the end of 2027, due to costs, unclear value or weak risk controls.

UiPath itself is a live example of this shift. Through 2026, UiPath's own marketing moved from calling itself an RPA company to calling itself an "agentic automation" company, folding AI agents, document processing and its older robot-based tools into the same Basic, Standard and Enterprise pricing structure. That is not necessarily dishonest, UiPath genuinely has added AI agent features, but it does mean the vendor that popularized the term "RPA" no longer wants to be known for it, which tells you something about where the market is pointing.

So ask any vendor to show you the product working on your own examples. Ask what decisions the AI makes and what happens when it gets one wrong.

RPA, AI agent or custom software: which should you choose?

• Choose RPA if: The task is stable and rule-based, and one of the systems has no API.
• Choose an AI agent if: The task needs reading, sorting or judgment, and a person can check the risky steps.
• Choose a no-code workflow if: Your apps all have APIs and the steps are simple. Tools such as Zapier and Make often cost less than RPA for this. See my Zapier vs Make guide.
• Choose custom software if: You have many robots and workflows that break often, or you are paying high per-bot fees. Sometimes the better fix is one small internal tool that replaces the old screens the robots click through.

For a step-by-step look at building a simple agent yourself, read my guide on how to build an AI agent. For wider options, see my guide to business process automation.

How Blur Studio approaches RPA and AI agents

I start by mapping the task. I look at each step, which systems it touches and where a person makes a call. Then I pick the simplest tool for each step.

Often that means a direct API connection where one exists, an AI step for the messy parts and a person approving the risky parts. Where an old system blocks progress, I can build a small custom tool to replace it. Your business owns the finished code and all the accounts. I can also look after it after launch, so it keeps working when your systems change.

Need something completely custom?

Everything above is a path to get it running quick and cheap. But sometimes the honest answer is that none of those quite fit, and what you actually need is something custom that you own, built uniquely for your business from the ground up, with AI thoughtfully designed into the parts of it that your processes actually benefit from. An asset that grows in value, not a generic, one-size-fits-all boilerplate that's an expense with a lock-in. This is a complete bespoke build, and it's something I do very well. Get in touch to know more.