Why I Stopped Writing Browser Automation Scripts
After years of building browser automation, I had a realization: we're solving the wrong problem. We've been writing increasingly complex code to interact with interfaces that were never designed for machines. And then we wonder why everything breaks when a website updates.
The Maintenance Nightmare
If you've built browser automation, you know the cycle:
- Spend hours crafting the perfect selectors
- Test thoroughly on staging
- Deploy to production
- Wake up to alerts because the target site changed their CSS
- Repeat
This isn't sustainable. Every website update becomes a crisis. Every new feature requires rewriting automation code. We're spending more time maintaining scripts than benefiting from them.
The Real Problem
The issue isn't that websites change—it's that we're asking the wrong question. Instead of "how do I click this specific element?" we should be asking "what am I trying to accomplish?"
When a human uses a website, they don't think in terms of CSS selectors. They think: "I need to book a meeting" or "I need to submit this form." The intent remains constant even as the implementation changes.
A Better Way: Intent-Based Skills
What if we could capture workflows as intent rather than implementation? Instead of:
await page.click('#booking-btn');
await page.fill('#email-field', 'user@example.com');
We describe:
## Goal
Book a meeting
## Workflow
1. Navigate to booking page
2. Find and click booking button
3. Fill contact information
4. Submit form
This description survives UI updates because it captures what matters—the goal and the workflow—not fragile implementation details.
How SKILL.md Makes This Possible
SKILL.md files capture agent capabilities in a format that's both human-readable and machine-executable. The key insight is that modern AI can bridge the gap between intent and implementation.
Given a skill description like the one above, an AI agent can:
- Navigate to the right page
- Identify UI elements by context and appearance
- Execute the workflow
- Handle variations and errors
All without brittle selectors.
Creating Skills from Demonstrations
The real breakthrough is that you don't write these skills by hand. You record yourself performing the task, and AI generates the structured skill file automatically.
Five minutes of recording becomes a reusable, portable skill that works across different agent frameworks.
The Impact
For developers, this means:
- No more 3am pages because a website updated
- Skills that work across different sites with similar functionality
- Focus on high-level orchestration instead of DOM manipulation
For organizations, this means:
- Domain experts can create automation without coding
- Institutional knowledge captured in executable form
- Reduced maintenance burden
Try It Yourself
🚀 Check out SkillForge — record your screen, get a SKILL.md file
🔥 Support our Product Hunt launch
What workflows would you automate if you never had to worry about selectors breaking?
Top comments (0)