How it works
Selector-based
DOM selectors, XPaths, accessibility IDs
Agentic automation
Computer vision understands the visible UI
Why agentic automation
Selectors automate implementation. Agents automate experience.
Traditional automation breaks when the UI changes. Computer-use agents operate the screen the way a person does: by seeing, deciding, acting, and verifying.
Comparison
Selector-based vs. agentic automation.
| Aspect | Selector-based | Agentic automation |
|---|---|---|
| How it works | DOM selectors, XPaths, accessibility IDs | Computer vision understands the visible UI |
| Task definition | Code-heavy scripts | Python, CSV, or natural language tasks |
| When UI changes | Fix selectors and rerun | Reason about intent and adapt to layout drift |
| Cross-platform | Different tools per platform | One runtime across desktop, mobile, web, HMI |
| Complex UIs | Limited or impossible | Works wherever a screen is visible |
Task definition
Selector-based
Code-heavy scripts
Agentic automation
Python, CSV, or natural language tasks
When UI changes
Selector-based
Fix selectors and rerun
Agentic automation
Reason about intent and adapt to layout drift
Cross-platform
Selector-based
Different tools per platform
Agentic automation
One runtime across desktop, mobile, web, HMI
Complex UIs
Selector-based
Limited or impossible
Agentic automation
Works wherever a screen is visible
The problem with selectors
Selector-based automation ties workflows to implementation details. A renamed button, refactored component, localized title, or canvas-based UI can break a test that still describes the correct user journey.
getElementById("submit_btn") // Web: ID changes
findElement("com.app:id/login") // Android: resource ID refactored
findWindow("MainForm") // Desktop: title localizedThe cross-platform reality
Real products span browsers, native mobile apps, desktop clients, Citrix, embedded HMIs, kiosks, and hardware benches. Traditional automation asks each platform for a different toolchain. AskUI keeps the operating model consistent at the screen layer.
How agentic automation works
You define the outcome, the agent observes the screen, chooses actions, verifies results, and records evidence for review.
tasks.csv
| LOGIN-001 | Enter user email | Email accepted |
| NAV-001 | Open settings | Settings panel visible |
| LANG-001 | Switch to German | Labels are localized |
The bottom line
Users do not interact with selectors. They interact with what they see. Agentic automation moves the contract from brittle implementation details to observable user experience.
Get started
Start free. Scale when ready.
Free includes a 14-day trial, a non-commercial AgentOS license, and 5,000 inference credits. Paid plans add commercial AgentOS and Hub token-based usage.