Self-Healing AI Browser Agent for Data Extraction
Pratik Patil
The problem
The client needed to pull records out of many different web-based SaaS tools, most of which offer no API or integration. Hand-written scrapers break every time a vendor changes its UI, and writing a new one for every tool does not scale.
The approach
An AI agent explores a new web app, discovers which data fields are available, and the team picks the ones they need
The agent generates Playwright scripts that then run headless and in parallel for speed
When a page changes and a script breaks, the agent repairs it, with human review reserved for critical changes
Read-only safeguards so the agent can never change data in the source system
API-first by design, so browser automation is only used where no API access exists
Tech stack
Claude (via Amazon Bedrock and the Claude Agent SDK), Playwright MCP, AWS and Terraform.
Status
In active development. Designed from day one for sensitive data. Client details are kept confidential under NDA.
Like this project
Posted Sep 28, 2026
An AI agent that learns any web app with no API, writes its own Playwright scripts, and repairs them when the UI changes.
Likes
0
Views
0
