AI Glossary / Agents and automation / Beginner
Robotic desktop automation
Definition
Robotic desktop automation, or RDA, is software that automates tasks on one person's own computer: it watches the desktop, helps with or performs repetitive actions like copying between applications, filling forms and compiling files.
What it means
RDA is the personal-scale cousin of robotic process automation. Where RPA runs unattended on servers processing thousands of transactions, RDA sits on an individual's machine and works alongside them, sometimes called attended automation.
A typical use looks modest and saves real time: the call-center agent whose screen pulls up the customer's records automatically, or the clerk whose weekly file-merging ritual becomes one click.
Because it shares the desktop with a person, RDA's design questions are human ones: when may it act alone, when does it suggest, and how does the person override it. The technology records or reads screen elements and replays actions, so it depends on the applications staying visually stable.
That dependence is the known fragility: when a supplier portal or spreadsheet layout changes, desktop scripts break silently or click the wrong thing. Maintenance belongs in the budget from day one.
For a small business, RDA is often the cheapest first step into automation: no new systems, just the removal of swivel-chair work between the applications already in use.
How it works
RDA tools record or script interactions with the desktop: read this cell, open that application, paste into this field, click this button. Modern versions use accessibility APIs and element identifiers rather than raw screen coordinates, which makes them less brittle.
Attended mode is the defining feature: the automation triggers from the person's workflow, pauses for their input, and hands control back. Some run as helpers that assemble information rather than complete transactions.
AI components increasingly slot in: reading a scanned document, classifying an email, or drafting a reply the person then edits. The desktop script carries the result between applications.
Safety on a shared machine means clear rules: which actions need the person's confirmation, what the script may never touch, and a log of what it did, because on a personal machine the person answers for its clicks.
Three practical examples
- Customer calls: the caller's number triggers a lookup, and their account, orders and open tickets assemble on one screen before the agent says hello.
- Data re-entry: a script copies confirmed figures from emails into the accounting package, with the clerk verifying each batch.
- Reporting: Friday's files are gathered, renamed and merged into the standard pack while the analyst checks the exceptions it flags.
Think of it this way
Think of an efficient assistant who sits at your elbow: fetching the right folder as each caller comes through, pre-filling the routine parts of forms, and waiting for your nod before anything is filed.
The limit: the assistant understands the work and would notice nonsense. The script replays patterns, so the person at the desk remains the judgment layer, especially when screens change.
Worked example
killing the swivel-chair task
Suppose an administrator spends an hour daily copying appointment details from emails into a booking system, fifty entries with the same five fields.
An RDA script reads each email, fills the booking form, and holds it for a one-glance confirmation. The hour becomes ten minutes of checking, and errors fall because nothing is retyped. The figures are invented; the review-before-commit pattern is the point.
When the booking system updates its layout, the script pauses safely on the unfamiliar screen instead of misfiling, because it was built to verify each field's anchor before typing.
The administrator now owns the script: a small maintenance task that replaced a large repetitive one.
Illustrative case study
A fictional accounts team automates its invoice filing on one desktop. It works for months, then a software update moves one button, and the script files a week of invoices into the wrong folder before anyone notices.
The recovery is tedious but instructive: every click since the update is in the log, so the misfiled items are found and corrected in an afternoon.
The invented lesson: desktop automation fails quietly, so build the tripwires first: verify anchors, log actions, and sample-check the output folder like you would a junior's filing.
Common misconceptions and mistakes
- It is a robot on the floor. It is software on one computer; nothing physical is involved.
- It removes the person. RDA is attended by design: it assists and accelerates the person, who stays in charge.
- Set it and forget it. Desktop scripts depend on application layouts, so plan for maintenance after every software update.
Business use
Best first targets are repetitive cross-application chores on one desk: re-keying data, assembling files, pre-filling forms and pulling context for calls.
Keep the person as the commit step on anything irreversible, log every action, and re-test the scripts after any update to the applications they touch.
People also ask
How is RDA different from RPA?
RDA is attended and personal, running on someone's desktop; RPA usually runs unattended on servers at process scale.
Does it need AI?
Not necessarily. Classic RDA replays recorded actions; AI features like document reading can be added where inputs vary.
What breaks it?
Application updates that change screens, and unusual inputs it was never built to recognize.
Check your understanding
A desktop script misfiles a week of invoices after a software update. Which two design features would have limited the damage?
Answer: Anchor verification before each action (pausing on unfamiliar screens) and an action log, so misfiled items can be found and corrected quickly.
Related terms
Learn next
AI Agents and Automation for Small Business covers attended automations with human approvals: https://aiforbusinessandlife.com/academy/ai-agents-automation-small-business. [C4]
More in Agents and automation
- Problem-solving agent - A problem-solving agent is an agent that works toward a goal by searching: it imagines sequences of actions, evaluates where they lead, and chooses a...
- Proto-agent - A proto-agent is one of two semantic proto-roles from linguistics: the participant in an event who acts, causes, moves or experiences, as opposed to the...
- Rational agent - A rational agent is an agent that, given what it has perceived, chooses the action expected to achieve the best outcome according to its performance measure.
- Router agent - A router agent is an agent whose job is to decide where work goes: it reads an incoming request and directs it to the right specialist agent, tool, model or...
- Simple reflex agent - A simple reflex agent is the most basic kind of agent: it chooses its action from the current percept alone, using condition-action rules, with no memory of...
- State machine agent - A state machine agent is an agent built as an explicit set of states and transitions: it is always in one known state, such as waiting, collecting details...
Sources
[S3] NIST, AI glossary (AIRC). Full URLs are in the source ledger. Verified 11 October 2026.
Previous: Robot tax Next: Robotic governance