Computer-Use Agents and AI Web Scraping: What They Can and Cannot Do

Two different ideas

• Computer use: an agent controls a browser or desktop app to complete a task, for example filling in a form or exporting a report.
• Scraping: a program collects information from web pages at scale, which an agent can then analyse.

Good uses for a business

• Moving data between two tools that have no integration.
• Checking your own site for broken forms or missing pages on a schedule.
• Gathering public information, such as published prices, for research.
• Testing your own checkout or booking flow the way a customer would.

What can go wrong

Agents that click are slower and less reliable than a proper integration, and one wrong click can have real effects, such as sending a message or deleting a record. Pages can also contain hidden text that tries to give the agent new instructions, known as prompt injection. Give such an agent the least access possible and require approval for irreversible actions.

Scraping responsibly

• Check the site's terms and robots.txt, and respect them.
• Do not collect personal data without a lawful reason.
• Keep request rates gentle so you do not harm the site.
• Prefer an official API when one exists. It is more stable and clearly permitted.

Our advice

Use an API or a proper integration first, a scripted browser second and a free-roaming agent last. When a client asks us to automate a task, we choose the most predictable approach that works and keep a person in the loop for anything that cannot be undone.