An AI agent tried to guess what wine I was drinking based on my description — and the results were mixed to say the least

As AI continues to take over more and more everyday tasks, there's an increasing amount of people looking to enjoy experiences with more of a human touch.

AI can undoubtedly be useful great at sifting through large amounts of data, and taking away the drudgery of reporting and admin work - but what about areas which require the senses?

So can AI perform in matters of taste, smell and touch?

tAIster steps up

Salesforce has been a leader in pushing forward AI-enabled experiences in recent years, with its wide range of AI agents helping customers from all around the world.

At its recent Dreamforce 2026, I took part in an AI sommelier session powered by an AI agent to see if the technology could be applied to wine tasting.

In a plush executive lounge away from the madness of Dreamforce's campgrounds, we were given three glasses of mystery wines, and then connected to an AI agent by scanning a QR code with a smartphone.

Image 1 of 3

(Image credit: Future / Mike Moore)

(Image credit: Future / Mike Moore)

(Image credit: Future / Mike Moore)

This hooked us up with tAIster, an AI agent which had been educated on around 400,000 different types of wines, including their taste, feel, aroma and other attributes.

Before tasting a wine, the agent asked us to use a few keywords to describe how it looked, and how grippy it was in the glass. We were then asked to describe the wine's aroma, and finally, the taste - with the aim of being as descriptive as we could, so the agent could guess which wine we had been given.

So did it work? I'm not a massive wine buff, but I tried to be as descriptive as possible, with the hope if giving the agent what it needed. Ultimately, it guessed my first two samples correctly, but got the third hopelessly wrong - guessing it was Veuve Cliquot champagne rather than Chardonnay due to me mentioning the glass had some small bubbles in.

So all in all, a fun example of what AI agents could do in previously unexpected areas - and although it's just a proof of concept right now, I don't think any real-world sommeliers will be quaking in their boots just yet.

Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.