Agentic AI builds on the capabilities of large language models (LLMs) and adopts a modular approach. At their core, AI agents perform three steps: first, they gather information – whether text, images, sensor data or datasets from databases. They then devise a plan by processing and analysing the input. Finally, they carry out actions, typically by invoking a set of tools and application programming interfaces (APIs) available to them.
The real strength of agent‑based systems lies in enabling multiple specialised agents within a multi‑agent system to work towards an overarching objective. For example, there may be research bots that source and pre‑process specific information, or quality‑assurance bots that validate the outputs of other bots. The target state for such orchestrated ensembles is often referred to as the Agentic Mesh – a network of interacting agents that, collectively, can take on highly complex tasks in dynamic environments.