Hey,
I’m looking for some technical feedback on a workflow I’ve been developing to solve a common pain point: automating ERP and CRM layouts without burning through LLM tokens or dealing with brittle selectors.
The idea is to avoid constant LLM prompting for every action. Instead, I'm using AI (Gemini/Groq/OpenAI) for an initial analysis of the UI. Once the AI "understands" the scenario and creates a "Blueprint," the execution runs autonomously for subsequent runs.
I’ve integrated a few things to make this work:
- Token Efficiency: AI is only used during the initial learning phase, not for every click during playback.
- Manual Blueprint Control: Support for manual editing if the AI struggles with a specific element.
- CI/CD Ready: Built a CLI for automated pipelines.
I’ve put the code and a demo GIF on GitHub (it's a project called Aetherium). I'm curious to hear from other QA engineers: does this model of "Initial AI analysis vs. Autonomous playback" feel like a viable solution for large-scale enterprise apps?