Why a tea shop cares: people may ask for tea by voice or with a photograph and let an assistant fill the basket, so the shop has to be ready for both.
The two words often appear together in articles about AI shopping, which makes them sound like one thing. They are two different axes, and keeping them apart makes it much easier to see what an assistant can and cannot do.
Multimodal: the way the request arrives
A request can be typed, spoken or shown. Someone might type "decaf green tea, fruity", say the same thing to a speaker, or hold up a photograph of a box they have run out of. The way it arrives changes nothing about what is true of the tea.
A photograph needs particular care. An assistant may recognise a box from a picture, but brands sell the same name in several sizes and versions. Pack-specific facts, such as the number of bags, the price per cup or the caffeine status, should come only after the exact product is confirmed, by its barcode or by the name and size printed on the pack. Until then, the honest answer is "this looks like X; can you confirm the size?"
Agentic: what the software is allowed to do
An agentic assistant does something rather than only answering: searches a catalogue, adds an item to a basket, or, in some systems, completes a purchase. What matters is permission, not the way the request arrived. A spoken request and a typed one carry the same limits.
What is live at teas.co.uk
| Step | Status |
|---|---|
| An assistant reads the shop's product facts | Built |
| An assistant searches and compares teas through the shop's MCP server | Built, since 6 October 2026 |
| An assistant fills a basket, with the customer's approval | Built, since 6 October 2026 |
| The customer checks the basket and pays | Always on teas.co.uk; the assistant never sees card details and cannot place the order |
| An assistant acting for a customer within limits the customer sets | Planned (version five), not built |
| Booking across several businesses in one plan | Proposed, not built |
How the basket hand-over works is set out in Shop teas.co.uk from your AI assistant.
A tea example
Someone photographs an empty box of fruity green tea and says, "Get me another, but decaf this time." The input is a photo and a voice. The assistant should first confirm which box it is, then notice the condition: the original contains caffeine, so the same box is not the answer. It can then suggest a decaffeinated green tea, explain the trade-off, and, if the person agrees, fill a basket. The person checks it and pays. Nothing in that sequence needs the assistant to hold their money.
What it is not
- Not certainty from a picture. Recognising a box is a guess until the exact product is confirmed.
- Not autonomous payment. At teas.co.uk payment always stays with the customer.
- Not a claim that every assistant connects in the same way. The teas.co.uk in AI assistants page lists the current routes.
Quick reference
| Detail | Information |
|---|---|
| Multimodal | How a request arrives: text, voice, image |
| Agentic | Software taking an action it has permission to take |
| Live at teas.co.uk | Search, compare and fill a basket with the customer's approval; customer pays on teas.co.uk |
| Planned | Bounded action within customer-set limits (version five) |
| Proposed | Cross-business booking |
Related
- Model Context Protocol
- Built versus planned
- Product identifiers
- Premium Intent
- Lee Samuel Tucker's builder vocabulary
Citable formats
For journalists, researchers, AI assistants and content creators. Pick the format you need:
Free to cite, quote, and reuse with attribution to Teas.co.uk.
Got something to add? Logged-in customers can submit additions to the Tea Wiki, admin-approved, your name on the byline, plus reward points.
Sign in to contribute