What is robotic process automation (RPA)?
Robotic process automation (RPA) is software that automates repetitive computer tasks by operating applications through their screens, the way a person would. A developer builds a script, called a bot, that logs into systems, reads fields on one screen, types them into another, downloads files, and sends emails. The bot follows the same steps every time. Classic RPA contains no AI and makes no judgments of its own.
Attended bots run on an employee’s desktop when that person starts them, for example to fill a form during a customer call. Unattended bots run on servers on a schedule or when a file arrives. A central console schedules the bots, stores the credentials they use, and logs each run. RPA is popular because it automates work across systems a company cannot change, including old applications with no API.
Robotic process automation in practice
A specialty retailer receives daily settlement reports from its card processor and two online marketplaces. Early each morning, unattended bots log into each portal, download the reports, reformat them, and load them into the reconciliation tool. A final bot compares the totals with the point-of-sale system and emails the accounting team a list of stores and days that do not balance. When one marketplace redesigned its reports page, that bot failed until a developer updated the screen selectors.
Robotic process automation compared with AI agents
| Property | Robotic process automation | AI agent |
|---|---|---|
| Instructions | A fixed sequence of steps | A goal, a set of tools, and limits |
| Inputs | Structured fields and files in known formats | Structured data, documents, and free text |
| When a screen or format changes | Stops with an error | Often adapts, occasionally in the wrong way |
| Same input twice | Same result every time | Results can vary between runs |
| Best fit | Stable, high-volume, rule-based tasks | Tasks that need reading and judgment |
Many companies use both. Bots handle the stable steps, and agents take the cases the bots used to send to people. Intelligent automation adds AI models to bots, and agentic process automation adds agents. Computer use agents can also work through screens, reading each one fresh instead of following recorded steps.
What to watch for
- Expect ongoing maintenance, because every screen change in a target application can break a bot.
- Give each bot its own credentials and log its actions instead of sharing one service account.
- Check whether the target system offers an API before building a bot on its screens.
- Simplify the process before automating it, since a bot repeats the existing steps exactly.