Explainer
AI that does tasks on your computer while you watch: how it works and how to stay safe
Updated · By the Lida team
Short answer
An AI computer agent looks at your screen, decides the next click or keystroke, and carries it out, repeating until the task is done. The safest kind works where you can watch it, asks before risky steps like submitting or sending, stops the moment you touch the mouse, and keeps a log. Lida's Agent mode works this way on Windows, and it never types into password fields, enters card numbers, transfers money or changes security settings.
How a computer-use agent works
A computer-use agent runs a loop: look at the screen, decide the next action, act, look again to see what changed. Unlike a script, it does not need the screen to be exactly as expected; it reads what is there. That flexibility is the point, and also why safety rules matter: an agent that can click anything needs clear limits on what it should click.
Agents, macros and cloud agents compared
| Approach | Where it acts | Strengths | Trade-offs |
|---|---|---|---|
| Macros and recorded automations | Your PC, replaying fixed steps | Fast and predictable for identical, repeated jobs | Break when a window or layout changes; need setting up |
| Agents in a separate workspace or cloud browser | A separate desktop, virtual machine or remote browser | Can work in the background while you do something else | Harder to watch; may not reach the apps and files on your own desktop |
| On-screen agent you watch, such as Lida | The screen in front of you | Works in the apps you already use; you see every step and can take over | Uses your screen while it works |
A safety checklist before you let any AI act
- Approval for risky steps. Submitting a form, sending a message, or deleting something should need your yes.
- An instant stop. A shortcut, and ideally “my hands win”: any mouse or keyboard input from you stops it.
- Hard limits. Things it never does, whatever it is asked, such as entering passwords or payment details.
- A record. A log of what it did, where, and whether it worked.
- Visible state. You can always tell whether it is listening, thinking or acting.
- Privacy limits. It only sees your screen while working, and you can keep sensitive apps out of view.
How Lida's Agent mode handles each point
- Approval: safe steps run; risky ones wait behind an approval card, where you choose Allow or Skip.
- Stopping: touch the mouse or keyboard and Lida stops at once; say “carry on” to resume. Ctrl + Alt + Space silences Lida and stops anything it is doing.
- Never-list: Lida never types into password fields, never enters card numbers, never transfers money, and never changes security settings.
- Activity log: everything Lida does in Agent mode is logged: what it did, where, how risky it judged the step, and whether it worked. A short before and after replay of each step is coming soon.
- Visible state: a status pill shows whether Lida is guiding, teaching or acting, which step it is on, and a stop button. Its cursor shows each control it clicks or types into.
- Privacy: Lida sees your screen only during a session you start, and its view is blanked while a password manager, a banking app or any app you list is in front.
You can also switch modes mid-task. Say “show me how” and Lida stops acting and points at the next step for you to do yourself. The full list is on the security page, and how to let an AI control your computer safely compares these rules with Claude and ChatGPT computer use.
Good first tasks to hand to an agent
- Repetitive form filling where you check the result before it is submitted.
- Tidying, renaming or moving things across apps.
- Multi-step settings changes you understand but do not want to click through.
- Anything you have done with Guide mode once and now just want done.
What it costs
An agent task in Lida uses session time like anything else: about one credit a minute while Lida works. A cost quote before each agent task is coming soon. Plans start with a free tier of 30 credits a month; see pricing, where prices are marked provisional. Lida is in private preview on Windows; request early access.
Questions
Is there an AI that can use my computer for me?
Is it safe to let an AI click and type on my PC?
How do I stop an AI agent that is working on my screen?
How is this different from a macro?
Lida is a Windows desktop companion that sees your screen and talks with you. It is in private preview, and invites go out in waves.