How it works
An agentic system usually runs on an LLM. It gets a goal and access to tools: a catalog search, an order tracking API, the site's cart.
The system then breaks the task down. It calls the right tool, reads the result, decides on the next step, and repeats until the goal is met. If it gets stuck, it stops or hands over to a human.
The level of autonomy varies a lot between systems. Some only suggest, others go all the way to purchase. The business sets the limits: which tools the AI can use, what it can do on its own, and when it has to ask for approval.
Why it matters in e-commerce
Agentic commerce rests on two building blocks: generative AI, which personalizes the answer, and agentic AI, which acts. On a brand's site, the second one changes what an assistant can do. Instead of pointing to a help page, it looks up the information in the product data, compares options and moves the purchase forward.
Shoppers are also starting to use their own agents, in ChatGPT or Gemini, to compare products before buying. Brands need to be readable by those agents, and to offer an AI on their own site that can act.
Further reading: Agentic Commerce: The New Era of AI-Powered Shopping Experiences
Real-life example
A shopper on a luggage brand's site types: "I'm looking for a light carry-on for a weekend trip, blue if possible." The assistant queries the catalog, keeps the carry-on models, checks which colors are in stock, then shows three suitcases and explains how they differ. The shopper picks one, and the assistant adds it to the cart without leaving the conversation.
Common confusion
People often mix up agentic AI and generative AI. Generative AI produces text, an image or an answer. Agentic AI uses that kind of model to act: search, compare, add to cart. The same assistant can do both in one conversation.