Why Browser Automation Needs Intent-Based Skills

#ai #automation #showdev #webdev

Browser automation has a fundamental problem. We've been teaching computers how to do things instead of teaching them what we want to achieve. This approach creates brittle systems that break at the slightest change.

The Selector Trap

Traditional browser automation relies on CSS selectors, XPath queries, and coordinate-based interactions. These are implementation details that have nothing to do with the actual task being performed.

When a developer writes:

await page.click('#submit-btn-3');
await page.fill('.form-input[name="email"]', 'user@example.com');

They're not saying "submit the form" or "enter the email address." They're saying "click the element with ID submit-btn-3" and "fill the input with name email."

This is fragile. When the website updates and the ID changes to submit-button, the automation breaks. Not because the task changed, but because the implementation details changed.

Intent Over Implementation

Intent-based automation flips this model. Instead of specifying exactly how to do something, you describe what you want to achieve:

## Goal
Submit the contact form with user information

## Workflow
1. Navigate to the contact page
2. Fill in the name field
3. Fill in the email field
4. Fill in the message field
5. Submit the form

Notice: no selectors, no coordinates, no implementation details. Just intent.

Why Intent Survives Change

When you describe intent, the AI figures out the implementation. If the website updates:

The AI adapts to new element locations
The AI recognizes equivalent functionality
The skill continues working

This is how humans operate. When a website redesigns, you don't forget how to use it. You look for the same functionality in new places. AI agents should do the same.

SKILL.md: Capturing Intent

The SKILL.md format enables intent-based browser automation:

# Book a Meeting

## Goal
Schedule a 30-minute team sync for next Tuesday

## Workflow
1. Open the calendar application
2. Create a new event
3. Set the duration to 30 minutes
4. Select next Tuesday as the date
5. Add team members as attendees
6. Send the invitation

## Context
- Calendar app is Google Calendar
- Team member list is in the company directory
- Preferred meeting times are 10 AM - 4 PM

This description tells the AI what to accomplish without prescribing exactly how to do it.

How SkillForge Enables Intent-Based Automation

SkillForge generates SKILL.md files from screen recordings:

Record — Show the AI what you want to do
Extract — AI understands your intent from the demonstration
Structure — Intent is captured in SKILL.md format
Deploy — Any compatible agent executes based on intent

The AI observes your actions and infers the underlying goals. It doesn't just copy your clicks—it understands your purpose.

Real-World Impact

Before: A developer spends hours writing and debugging selectors. The automation works until the next website update, then breaks mysteriously.

After: A domain expert records themselves once. The AI extracts intent and generates a skill. The automation adapts to website changes automatically.