# UiPath

UiPath combines robotic process automation with AI agents and orchestration to automate enterprise processes. Learn robots, agents, orchestration and when to use RPA.

## Overview

UiPath is an enterprise automation platform best known for **robotic process automation**, or RPA. Software robots built in UiPath operate applications the way an office worker does: they open programs, read screens, click, type, copy values and move files. Where systems offer APIs, robots call those instead. Robots follow defined steps exactly, many times a day, without getting tired.

Over the last few years UiPath has added AI to this foundation: document processing that reads invoices and forms, AI agents that can reason and choose actions, and an orchestration layer that coordinates robots, agents and people within one governed process. UiPath calls this **agentic automation**.

Think of a busy back office. Clerks follow precise procedures for data entry. A few analysts handle unusual cases that need judgement. A supervisor routes work between them and keeps track of everything. In UiPath, robots are the clerks, agents are the analysts, people still handle the real exceptions, and orchestration is the supervisor.

## What it is

UiPath is a platform of several products that work together. Names and packaging change over time, but the main roles are stable.

| Component | Purpose | Example |
|---|---|---|
| Studio | Design automations visually or in code | Build a workflow that logs in to an ERP and posts invoices |
| Robots | Run automations, attended or unattended | An unattended robot processing overnight batches |
| Orchestrator | Deploy, schedule, monitor and govern robots | Assign 500 queue items across ten robots |
| Queues | Hold work items for robots to process | One item per incoming invoice |
| Document processing | Classify documents and extract fields | Read supplier name, totals and PO number |
| Agent Builder | Build AI agents with prompts, tools and guardrails | An agent that checks invoice discrepancies |
| Maestro | Orchestrate processes across agents, robots, APIs and people | End-to-end invoice-to-payment process |
| Action Center | Hand tasks to people for review or approval | A clerk approves a mismatched invoice |
| Process and task mining | Discover how work actually flows | Find the steps that take longest |
| Test automation | Test applications, including with AI agents | Regression tests for a web portal |

## Where it came from

Daniel Dines and Marius Tîrcă founded the company in Bucharest, Romania, in 2005 under the name DeskOver, initially building automation libraries for other software firms. It was renamed UiPath in 2015 as it focused on robotic process automation, and moved its headquarters to New York in 2017. UiPath listed on the New York Stock Exchange in April 2021.

AI features arrived step by step. UiPath acquired Re:infer, a natural language processing company, in 2022. In July 2024 it released Autopilot, generative AI assistance across the platform, and in October 2024 it previewed Agent Builder for combining RPA with AI agents. In April 2025 it launched its platform for agentic automation, with Maestro as the orchestration layer for agents, robots and people. At that launch UiPath also announced work with partners on agent interoperability, including support for the [A2A](/tools/a2a) protocol and a collaboration that lets agents built with [LangGraph](/tools/langgraph) take part in UiPath automations.

## How it works

### Robots for fixed steps

A developer, or a trained business user, designs an automation in Studio as a sequence of activities: open an application, find a field, type a value, read a table, call an API, write to a file. Activities identify screen elements through selectors, which describe an element by its properties. Robots run these automations on desktops, servers or virtual machines.

### Orchestration of robots

Orchestrator manages which robot runs what and when. Work often arrives in queues: each queue item, such as one invoice, is processed by the next free robot. Orchestrator tracks success, failure and retries, stores logs and holds credentials centrally in assets so automations do not contain secrets.

### Documents and agents for judgement

Document processing classifies incoming files and extracts fields, with confidence scores. Agents built with Agent Builder use a model, a prompt, tools (which can be robots, APIs or other automations) and guardrails. They handle steps where the right action depends on reading and reasoning, such as explaining why an invoice does not match its purchase order.

### Process orchestration

Maestro coordinates the end-to-end process: which steps robots do, where agents decide, when a person must review in Action Center, and how long each step takes. Every step is logged, which supports audit and process improvement. See [human in the loop](/tools/human-in-the-loop) and [AI governance](/tools/ai-governance).

## Key concepts and terms

| Term | Meaning |
|---|---|
| RPA | Robotic process automation: software robots that follow defined steps |
| Attended robot | A robot that works alongside a person, triggered by them |
| Unattended robot | A robot that runs on its own, on schedules or queues |
| Selector | A description that identifies a user-interface element |
| Queue item | One unit of work waiting for a robot |
| Orchestrator | The control centre for deploying and monitoring robots |
| Document understanding | Classifying documents and extracting data from them |
| Agentic automation | Combining robots, AI agents and people in one process |
| Guardrail | A rule that limits what an agent may do |
| Human-in-the-loop task | A step where a person reviews or approves |
| Process mining | Analysing system logs to see how processes really run |

## A worked example

This is an illustrative scenario. A shared-services centre processes about 15,000 supplier invoices a month for several companies in a group.

1. **Intake.** A robot monitors a mailbox and saves each PDF invoice as a queue item.
2. **Extract.** Document processing reads supplier, invoice number, date, line items and totals, each with a confidence score.
3. **Low confidence.** Fields below the confidence threshold, about 8 percent of invoices in this scenario, go to a person in Action Center for correction.
4. **Match.** An agent compares each invoice with the purchase order and goods receipt in the ERP, using a robot as a tool to look them up. It classifies the result as matched, price difference, quantity difference or missing PO, and writes a short explanation.
5. **Post.** Robots enter matched invoices into the ERP.
6. **Exceptions.** Mismatches go to an accounts payable clerk with the agent's explanation, which shortens their review.
7. **Monitor.** Maestro shows the time spent in each step. The team sees that most delay is in exception review and adjusts the tolerance rules.

Average processing time falls from several days to the same day for most invoices, and every step for every invoice is traceable.

## Strengths and limitations

**Strengths:**

- Mature enterprise governance, security and central management.
- Works with legacy and desktop systems that have no APIs.
- Strong document processing for invoices, forms and contracts.
- A clear model for combining robots, agents and people in one process.
- Large ecosystem of activities, templates and trained practitioners.

**Limitations:**

- Screen-based robots break when applications change their layout.
- Licensing and infrastructure costs can be significant.
- Many robots across many processes need active maintenance.
- Automating a poor process only makes a poor process run faster.
- AI agent steps add variability that must be tested and monitored.

## How it compares with alternatives

| Option | Choose it when | Trade-off |
|---|---|---|
| UiPath | Large organisations automating many processes across legacy apps | Cost and maintenance of many robots |
| [Computer use](/tools/computer-use) agents | Flexible, low-volume tasks without stable scripts | Slower and less predictable per step |
| APIs and integration platforms | Systems offer good APIs | Need engineering, no screen automation |
| [n8n](/tools/n8n) or similar | Teams want lighter app-to-app automation | Weaker for desktop and legacy systems |
| [Camunda](/tools/camunda) | Processes are modelled in BPMN with strong audit | Needs separate tools for screen automation |

## Practical guidance and best practices

1. Fix and standardise the process before automating it.
2. Prefer APIs over screen automation wherever they exist.
3. Use robots for fixed, repeatable steps and agents only where judgement is needed.
4. Route low-confidence extractions and agent decisions to people.
5. Keep credentials in central stores, never inside automations.
6. Use queues so work is tracked item by item and can be retried.
7. Monitor robot failures and selector breakages, and plan for application updates.
8. Test agent steps on a fixed set of real cases before and after every change.
9. Measure the whole process, not just robot run times, to find where time is lost.

## Common mistakes

- Automating a broken or undocumented process as it is.
- Building fragile selectors that break with every small screen change.
- Putting an AI agent where a simple rule or lookup would do.
- Letting agents post financial transactions without a review threshold.
- Running hundreds of robots without ownership or monitoring.
- Measuring success by number of robots instead of business outcomes.

## Frequently asked questions

### What is RPA?

Robotic process automation is the use of software robots that follow defined steps to operate applications, often through their user interfaces. It is most useful for repetitive, rule-based work across systems that lack good APIs.

### What is agentic automation?

It is UiPath's term for combining robots, AI agents and people in one orchestrated process. Robots handle fixed steps, agents handle steps that need reading and judgement, and people handle exceptions and approvals.

### Is RPA still relevant with AI agents?

Yes. Robots are fast, cheap per run and predictable for fixed, high-volume steps. Agents are better at variation but slower and less predictable. Most real processes benefit from both, each where it fits.

### What is the difference between attended and unattended robots?

An attended robot runs on a person's machine and is started by them to help with part of their work. An unattended robot runs on its own on a server or virtual machine, driven by schedules or queues.

### How does UiPath compare with computer-use agents?

Classic UiPath robots follow designed steps and are reliable when screens are stable. Computer-use agents decide each action from screenshots, which copes better with change but is slower and less predictable. UiPath has also added its own agents for interface work.

### Who founded UiPath?

Daniel Dines and Marius Tîrcă founded the company in Bucharest in 2005, originally under the name DeskOver. It became UiPath in 2015 and is now headquartered in New York.

## Glossary

- Robot: software that runs an automation.
- Studio: the design tool for automations.
- Orchestrator: the control centre for robots.
- Queue: a list of work items waiting for robots.
- Selector: an identifier for a screen element.
- Document processing: classifying and extracting data from documents.
- Agent: an AI component that reasons and uses tools.
- Maestro: UiPath's orchestration layer for agents, robots and people.
- Action Center: where people review and approve tasks.
- Process mining: analysis of event logs to map real processes.

## Summary

UiPath grew from a Bucharest start-up founded in 2005 into a major enterprise automation platform built on RPA robots, central orchestration and document processing. Since 2024 it has added AI agents and, from April 2025, an agentic automation platform with Maestro to coordinate robots, agents and people. Robots handle fixed steps, agents handle judgement, and people handle exceptions, all with audit trails. Standardise processes first, prefer APIs, use agents only where they add value, route uncertain results to people and keep robots monitored and maintained.