By Xavier Sala, 15 June 2025

AI shopping agents | shopping assistants | conversational commerce | personal shoppers | agentic commerce | agent-readable | agentic AI

From traditional ecommerce with search bar, category browsers, recommendations, etc.

To interacting with AI agents / shopping assistants in a natural language and acting on behalf of the user. s

How do AI shopping agents work ?

The customer interacts with the agent in natural language and the agent understands, process and returns the right answer.

Common interactions could be,

Product Discovery and recommendations

• I need a comfortable red dress for a summer wedding under 100€.
• What tools do I need to reform my kitchen ?
• What gears would you recommend to go camping on the Pyrinees ?

Product Comparison & Details

• What are the key differences between [product A] and [product B]?
• Is this item available in other colors or sizes?
• I’m looking for eco-friendly alternatives — what do you recommend?
• Can you show me the best recommendations for this product ?

Bundling & Inspiration

• What else do I need to complete this look?

Shopping Process & Logistics

• What’s the estimated delivery time for this item?
• Can I return this if it doesn’t fit ?

Placing orders

• Add this item to my cart.
• Place the order to my home address.
• Where and when can I have it be delivered ?

It is worth to notice that AI commerce agents goes beyond what chatbots do.

Challenges and limitations

• Retail websites should implement internal AI shopping agents - AI agents are a new channel inside the omnnichannel.
• Generic AI platforms building shopping agents - they interact with the different reteailers platforms. Customers could go less to retailers websites.
• Who will own the customer relationaship ? - It seems retailers could have less interactions if customers goes via generic shopping agents.
• Making ecommerce websites should be agent friendly - otherwise be exposed to lose sales.
• Agent-to-agent integration / communication.
• We are still in early stages - results might not be yet acurate enough.
• Which role will paid ads and sponsored listings play ?
• Computing power and costs - providing these platforms can be considerable.
• Shopping agents are not just for B2C nor physical products - they apply to B2B and services.
• AI-generated results can sometimes be influenced by biases, inaccurated or wrong.
• Privacy and data security.
• Still technological limitations could make agents slow and being not savy as expected.

Technical high level details

The user interaction, usually starts with a natural lanauge interaction. The request goes to the LLM (Large Language Model) that along with an orchestrator decides what the user is asking for, for example if asking for a product search, a recommendation, asking for a clarification question, etc. so that the orchestrator can decide which tool to use to get the right answer and render back to the customer. Each tool does it own simple work; ie. talking to the products database, use an API, place an order, etc. The agent gets the answer back to the customer.

Examples

Retailers providing their own AI agents

• Zalando - AI-powered fashion assistant

AI platforms building shopping agents

• Perplexity Shop Like a Pro (I have not tried it yet)
• OpenAI Operator (I have not tried it yet)

Current status and my personal opinion

The AI race is there; companies investing milions to get better and better and lead the race. Every two months new releasea just surpass what was there just few months before.

To me, all past predictions about AI failed; as it was supposed that AI would do repetitive and those tasks that humans do not wan to do but the reality is that AI is taking already the most creative tasks.

So, I do not think anyone can predict what we will have in three or five years time.

Unfortunately, all AI commerce agents I tested and read about did not convinced me yet, the results were not that good. So I believe we are still a bit far from getting good results.