Every week, we round up the most interesting AI and technology news that matters for independent cafe and coffee shop owners. No jargon, no hype - just what you need to know and why it matters for your business.

This week is a study in contrasts. Berry AI is being installed across 1,000 Culver's restaurants. Marc Lore wants anyone with a phone to spin up a restaurant brand from a single prompt. Meanwhile, in Stockholm, an AI-run cafe just ordered 120 eggs for a kitchen with no stove. The same week one report finds that 65% of operators think tech improves hospitality and only 41% of customers agree. Take it all together and the message is clear - AI is moving fast, but the experienced operator's judgement is still doing more heavy lifting than most of the press releases admit.

Berry AI Rolls Out Vision AI Across 1,000+ Culver's Restaurants

On 1 May, Berry AI announced a brand-wide partnership with Culver's - the US burger and frozen-custard chain - to deploy its real-time computer-vision platform across more than 1,000 locations. Cameras now sit over the drive-thru and key service points, watching how fast cars move, how long orders take to assemble, where bottlenecks form, and how consistently service standards are being met. The system plugs straight into the existing POS workflow, so managers don't have to learn a new tool.

The performance numbers Berry AI is quoting from earlier deployments are striking: up to 70% jumps in drive-thru comp sales, 20-40% reductions in service times, and up to 20% throughput improvements. Berry AI's CEO Eric Lam framed it bluntly - this scale of rollout signals that Vision AI is moving from "experiment" to "core infrastructure" in QSR.

Two weeks ago we covered Mill bringing Gemini-powered vision to the bin - watching what a kitchen throws away. This is the inverse: vision watching the front line, measuring how the operation actually performs against its own standards.

What this means for you: A 1,000-site rollout is a chain story, but the underlying lesson is general. The chains have just bought themselves the ability to see their own operation in real time - which corner is slow on Tuesdays, which station drops behind at 12:15pm, which shift loses ninety seconds per ticket. That's been the chain advantage all along: visibility at scale. The good news for indies is that you don't need a vision system to know your operation - you can stand in your own room. But you do need to know your numbers cold, because that's the bit a single-site cafe genuinely can't see by walking the floor. Knowing which items actually make money once cost and prep time are factored in is the indie equivalent of Berry AI's dashboard. You're already watching the floor. The job is to make sure you're also watching the margins.

Read the full story on Restaurant Dive ->

Marc Lore Launches "Wonder Create" - Anyone Can Spin Up a Restaurant Brand from One AI Prompt

Last week we covered Wonder hitting 100 locations and rolling out Spyce-powered Infinite Kitchens. This week Marc Lore went a step further. On 5 May, he unveiled Wonder Create - a tool that lets anyone, from a food entrepreneur to a TikTok influencer, type a description of a restaurant and have AI generate the entire brand in under a minute. Name. Logo. Menu. Recipes. Pricing. Allergen information. Photography. Lore described it as "a Shopify front end with an AI prompt" for restaurants.

The new brand goes live across Wonder's automated kitchen network - 120 sites today, a target of 400 next year, and 1,000 by 2035. There's no kitchen to fit out, no equipment to buy, no staff to hire. Just a brand on top of a robot kitchen. Lore is positioning Wonder for a 2028 IPO and was openly explicit about the scale he's chasing - "thousands of unique brands" running through the same physical infrastructure.

This is the consequence of last week's story playing forward. Once you have a robotic kitchen platform that can produce the food, the brand layer becomes software. And once it's software, anyone can generate it.

What this means for you: It would be easy to read this as terrifying. A single prompt, a brand on the app stores, food at the door inside thirty minutes - and a competitor that exists entirely on the phone of someone who has never run a kitchen. The honest truth is that some of these AI-generated brands will work, especially the ones with a creator's audience already attached. But here's the thing the slide deck doesn't say: the Wonder Create brand is a menu and a logo. It is not a place. It has no room, no smell, no familiar face behind the counter, no Sunday regulars, no birthday party last month, no kid who got their first hot chocolate at your bar. That stuff has never been more valuable than it is right now, because for the first time in a decade it can't be cloned by software. The play for a single-site cafe is to lean into the things a prompt can't generate - and to know your numbers well enough to keep the lights on while you do. Smaller, sharper menus you can actually evolve quickly help here too. The agility moat just got harder to defend on price. Defend it on character.

Read the full story on TechCrunch ->

A Real Cafe in Stockholm Is Being Run by an AI - And It's Going About How You'd Expect

This is the story of the week. San Francisco research outfit Andon Labs has opened a real, paying-customers cafe in Stockholm's Vasastan district that is managed end-to-end by an AI agent called Mona. Mona designed the menu, sourced the suppliers, posted the job listings, conducted interviews and made the hiring decisions. She places the orders, sets the prices, handles inventory. The cafe went live on 18 April, and Andon Labs published its first results post on 5 May. Two weeks of trading: 44,000 SEK in sales (a real cafe, real customers, real takings).

The honest part of the writeup is what makes it the most useful story we've covered all year. Mona ordered 120 eggs in week one for a cafe that has no stove. When staff pointed this out, she suggested using the high-speed oven - until they explained the eggs would explode. She tried to fix the problem of fresh tomatoes spoiling on sandwiches by ordering 22.5 kilos of canned tomatoes. She has, at various points, ordered absurd quantities of toilet paper and gloves. Andon Labs is running this experiment specifically to show what current-generation AI agents can and can't do when handed real operational responsibility - and they've been admirably blunt about the gap.

The team behind Mona aren't sceptics. They're true believers running a public stress test. The fact that they're open about the failure modes is the most credible thing in the AI-and-restaurants space we've read in months.

What this means for you: Read this carefully, because it's the most important framing of the year. AI is going to be in your business, in some form, within the next 24 months - through your POS, your invoice processing, your suppliers, your bookings system. That is happening whether or not you opt in. What Mona shows is that AI is a tool, not a replacement for judgement. Every single one of Mona's mistakes was the kind of thing an experienced operator would catch in five seconds: the eggs (we have no stove), the canned tomatoes (those are not fresh tomatoes), the gloves (we ordered some on Monday). The reason Andon Labs has staff "standing by to intervene" is that without operator judgement, the cafe doesn't function. The operators who win the next decade aren't the ones who hand the keys to the AI - they're the ones who use AI to remove the boring bits (transcribing invoices, drafting rotas, flagging price changes) while they spend more time doing the things AI demonstrably cannot do: read the room, read the team, read the customer, decide. The right way to think about AI in a cafe is as a fast, tireless junior who needs a senior operator to sense-check it. Take the senior operator away and you get 22.5 kilos of canned tomatoes.

Read Andon Labs' own writeup ->

Arabica Coffee Falls to an 18-Month Low - Rabobank Tips a Third Off by Year-End

Some genuinely good news for once. Arabica coffee futures (ICE Coffee C) have fallen to around $2.71 per pound this week - the lowest level since November 2024 - as the market reprices on expectations of higher near-term supply. Rabobank now forecasts Arabica futures could fall by roughly a third by Q4 2026 as the next harvest lands. This is a real shift. We've spent the last two years watching coffee prices break record after record on the back of Brazilian drought, Vietnamese supply issues, and shipping costs. That pressure is starting to ease.

The wider picture matches. UK wholesale butter has fallen to its lowest level since September 2021 (£3,540/t), with cream now 53% cheaper than a year ago and butter 41% cheaper, according to AHDB data following last week's Global Dairy Trade auction. After a brutal eighteen months on inputs, the curve has bent.

What this means for you: Resist the temptation to do nothing. Falling commodity prices are easier to misread than rising ones - because the gap between wholesale and what you pay can take weeks or months to close, and only if you're paying attention. Your roaster's price doesn't drop the day futures do. Your dairy supplier may absorb some of the relief into their margin. Now is the moment to look at every line on your last few invoices and challenge the ones that haven't moved - especially on coffee, milk, butter, and cheese. The operators who'll capture this shift aren't the ones who notice it on a news site. They're the ones tracking supplier prices through their invoices week by week and pushing back the moment they see a supplier sitting on the old price after the market has moved. With wage and rates pressure baked in for the rest of 2026, recovering 30-50p per litre on milk or a couple of pounds per kilo on coffee is the difference between a tight summer and a comfortable one.

Operators Love the Tech. Customers, Not So Much.

A new piece of research from Nation's Restaurant News and Datassential found a 24-point gap between operators and customers on whether technology actually improves hospitality. 65% of operators say it does. Only 41% of customers agree. The gap is widest on AI ordering - just 39% of customers said they'd be comfortable placing an order with an AI-generated persona, and the discomfort climbed sharply for diners over 35.

The disconnect is fascinating because the operators aren't wrong - the right tech genuinely does make their lives easier. Faster checkouts. Cleaner data. Fewer manual errors. The customers also aren't wrong - the wrong tech genuinely makes hospitality worse. Self-service that breaks. Voice agents that mishear. QR codes that lead to a forty-step checkout. Both sides are responding to a different version of the same word.

What this means for you: This is the most important data point in this edition for an indie operator. The chains will keep pushing customer-facing AI because their economics demand it. Your economics don't demand it. You can choose to keep the human at the till and use AI only on the back office - to process invoices, draft rotas, flag price changes, surface margin alerts. That keeps you on the right side of the customer gap and lets you compete on the thing chains genuinely can't replicate: a person who knows their regulars by name. The 39% comfort number isn't going to flip to 90% overnight. There's a real, durable advantage for operators who keep humans on the customer-facing side and put AI in service of those humans, rather than in their place. The cafes that run smoothly without burning out the owner tend to do exactly that already - automate the dull stuff, protect the human bits.

Read the full story on Nation's Restaurant News ->

The Brikly Take

Five stories, one quiet thread. AI is moving fast, the chains are moving with it, and the press release cycle is on full blast. But the more interesting stories this week aren't the rollouts - they're the gaps. The gap between what Berry AI's vision system can see and what an experienced manager already knows by walking the floor. The gap between what Wonder Create can generate and what a cafe actually is. The gap between what Mona ordered and what an operator with five years of experience would have caught in three seconds. And the 24-point gap between operators who think tech improves hospitality and customers who don't.

Those gaps are the indie advantage. The chains will close some of them with capital. They will not close all of them. And the operators who'll come through the next two years strongest are the ones who use AI as a tool - on invoices, on price alerts, on the boring data work - while keeping their judgement, their relationships, and their atmosphere firmly out of automation's reach. Mona is going to keep ordering 120 eggs for kitchens with no stove for a while yet. The lesson isn't that AI is useless. The lesson is that experience is still the most valuable thing in the room.

The chains have vision systems, AI brand factories, and POS agents that draft purchase orders. You have a room, regulars, and a head full of context that no model has ever been trained on. Use the tools to free up time. Spend that time on the floor. That's the play.

The Weekly Grind is published every Monday by Brikly - modular intelligence tools for independent cafe and coffee shop owners. Got a story we should cover? Get in touch ->

Ed O'Brien has run Hunters Cake Company for 17 years across cafes in Witney, Burford, and a bakery in Carterton, Oxfordshire. He's building Brikly - modular tools that give independent cafe owners the same data the big chains have, without the big chain price tag.