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

At the recent Dreamforce 2026 conference, Salesforce showcased an experimental AI agent named 'tAIster' designed to function as a digital sommelier. The agent, trained on a database of approximately 400,000 wine varieties, attempted to identify mystery wines based on user-provided descriptions of appearance, aroma, and taste. During a hands-on demonstration, the AI successfully identified two of the three samples provided to the author. However, it struggled with the final glass, misidentifying a Chardonnay as a Veuve Clicquot champagne, likely due to the user's mention of small bubbles. While the experiment highlights the expanding capabilities of AI agents in sensory-based tasks, the author notes that the technology remains a proof of concept. For now, professional sommeliers do not face immediate displacement, as the system still lacks the nuanced precision required for consistent, high-level wine evaluation in real-world settings.