Quality & Reliability

Beyond LLM Agents: A World Model for Visual Software Testing

with Manuel Weichselbaum

Thursday 9 July 16:15 – 16:20 Airstream 1

About This Session

What if a model could learn software the way humans do, through pixels, mouse, and keyboard alone? This talk introduces the Vision Action Model, a foundation model that understands UI concepts, not coordinates, and executes them deterministically. Purpose-built for software testing. Trained purely on interaction data.

Topics

AI Models
Agents
Agentic AI
Automation Testing
CI/CD
Compliance
Cross-Platform
Deep Learning
E2E Testing
HTML
Model Training
Neural Networks
Playwright
Reinforcement Learning
RPA
Selenium
Testing

← Back to Schedule

Speaker

Manuel Weichselbaum

CTO · anyconcept

CTO of anyconcept

Read bio Hide bio

Manuel Weichselbaum is co-founder and CTO of anyconcept, a research company building a world model for software interfaces. His background spans physics, telematics, and computer science, and he has spent over a decade doing AI research before that was the obvious thing to do. In 2020 he began exploring a question that the industry had largely ignored: what if a model could learn to understand software purely through visual interaction data, the way a human does, without ever reading a line of code. That question became the Vision Action Model, a domain-specific foundation model trained on pixels, mouse, and keyboard data alone. Manuel leads all VAM research and architecture at anyconcept. GHOSTS, the first product built on the VAM, is already running in production for enterprise customers. He is based in Graz, Austria.