AI agents have leapt from research papers to real‑world job sites, but turning a clever model into a dependable digital coworker still feels mysterious to many practitioners. This 45‑minute session demystifies the process by walking through the end‑to‑end build of an AI safety assistant that spots a crack in a structural beam, prioritises the risk, and automatically alerts the right people — all before the morning coffee cools. Blending technical depth with plain‑language storytelling, we will:

- Explain how Large Language Models (LLMs) supply human‑level understanding and reasoning, forming the “brain” of modern agents.
- Break down the Perception → Decision → Action loop, showing how computer vision, rules and LLM reasoning interact in milliseconds.
- Expose the design anatomy of an agentic system — memory, planning, and tool invocation — using a clear architecture diagram rather than code.
- Highlight must‑have guardrails for security, monitoring, evaluation and human‑in‑the‑loop approval, ensuring safety in high‑stakes environments like construction.

Whether you architect AI solutions, part of the WHS team, or simply champion innovation, you’ll leave with three pragmatic takeaways and a mental blueprint for bringing agentic AI from concept sketches to concrete impact in any industry. A short live demo follows the talk for those who want to see the crack‑detecting agent in action.