Seminarium Fizyki Wielkich Energii

Most results in particle physics today come as much from software as from detectors: simulation, reconstruction and analysis code. Beyond programming lies software engineering — version control, code review, testing and packaging — the practices that make results reproducible. In industry these are standard, and engineers who master them are in higher demand than programmers alone.AI coding agents now write code in seconds, which makes these practices more important, not less. An AI agent produces code faster than anyone can read it, so tests, review and version control become the only things checking it. Crucially, passing tests are not enough in physics: code and its results must be validated against known results before it can be trusted. I will give an overview of the current GenAI landscape, from chat-based assistants to agentic coding tools, both paid and free, and show on concrete examples what changes when an agent writes the code.The talk includes a live agentic coding session: describing a task in plain language, letting an AI agent write code, run and fix it — with the human steering the workflow.