Module 5.1 · Topic 2
Types of AI Agents
Bottom Line Up Front AI agents come in specialized types optimized for specific environments: browser agents for web-based tasks, desktop agents for operating system interaction, coding agents for software development,…
2.1 Browser and Web Automation Agents
Browser agents interact with web-based systems by reading HTML, interpreting user interfaces, and taking actions like filling forms, clicking buttons, and extracting data. They are the most mature and widely deployed agent type.
-
Form Filling and Data Entry
Fast/ChatA browser agent can populate complex forms where you supply structured input. The agent reads the form, maps your data to fields, handles dropdowns and multi-select options, and submits. Use this for single-session tasks with clear field mappings.
Filling out a vendor onboarding form: "Visit this URL: [form_URL]. Fill in the following vendor details: Company Name: Acme Corp, Address: 123 Oak St, Tax ID: 12-3456789, Primary Contact: Sarah Chen. Handle any required validation and submit the form. Report any fields that cannot be filled or errors encountered." -
Research and Information Extraction
Thinking/ReasonerA browser agent can visit multiple websites, search results, and compile findings. Use this when you need to cross-reference information across sites or extract structured data from unstructured pages.
Researching competitor pricing: "Search for and visit the pricing pages of these competitors: [list]. Extract current pricing for service tiers: Basic, Professional, Enterprise. Note any discounts, annual billing options, and add-ons. Compile results into a comparison table with source URLs." -
Account and Transaction Management
Thinking/ReasonerA browser agent can log into accounts, navigate dashboards, and execute transactions. Use this for routine account operations like generating reports, updating settings, or processing bulk actions.
Generating monthly reports: "Log into the analytics dashboard using provided credentials. Navigate to Reports > Monthly Summary. Export the past 12 months of summary data. Save as CSV. Verify all 12 months are included before closing." -
Real-Time Monitoring and Alerts
Deep ResearchA browser agent can periodically check websites, monitor status pages, or track changes. Use this for surveillance tasks where you need the agent to run repeatedly and alert on specific conditions.
Monitoring service status: "Every hour, visit [status_page_URL]. Check for any red or yellow incident indicators. If any are found, extract incident details and send an alert email to [recipient]. Continue until all indicators return to green."
2.2 Desktop and Operating System Agents
Desktop agents interact with your computer's operating system — opening applications, moving files, reading document contents, and executing system commands. They are less mature than browser agents but expanding rapidly.
Feasibility Checklist for Desktop Agents
- Technical Accessibility: Can the agent perceive your desktop environment? Desktop agents typically work through screen reading APIs or headless browser-like interfaces. Proprietary or specialized software may not be readable. Test first.
- Workflow Clarity: Is the workflow stable and rule-based? Desktop agents excel at consistent workflows — batch file organization, data migration between applications, routine report generation. Workflows with many decision branches or requiring subjective judgment are riskier.
- Error Recovery: What happens if the agent encounters an unexpected state? Desktop operations are harder to undo than web operations. Ensure your workflow can tolerate mistakes or includes rollback procedures.
Safety Cautions for Desktop Agents
File System Risk: Desktop agents can delete or overwrite files. Always test on non-critical data and maintain backups. Never grant desktop agents write access to directories containing irreplaceable data.
Application State Assumptions: Desktop agents assume applications are in a known state. If a window closes unexpectedly or an application crashes, the agent may not recover. Manual oversight is essential for high-stakes operations.
Unlock the Full AI Skill Building Experience
Visit LawQi for access to the full AI Skill Building experience.
Visit LawQi40% discount using code REASONABLE for personal subscriptions.
Free 48-hour preview access when investigating for teams.