The Weekly Grind: AI & Tech News for Cafe Owners - 10 August 2026
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.
Two results statements landed in the past fortnight, and read side by side they settle an argument this column has been circling all year. Sweetgreen, fast casual's robot-kitchen poster child, reported falling same-store sales and a $26m loss. Greggs, whose headline technology is a demand forecast and a loyalty app, reported a 20% profit jump in a market that shrank. One automated the making. The other automated the knowing. Around those two sit the rest of the week: Google quietly moved the food order inside the map, the government's rewrite of the zero-hours rulebook entered its final fortnight, and a drought repriced the second most important liquid in your building, effective September. Machinery is arriving at every layer of this trade. What this week's numbers say, unusually clearly, is which kind pays.
Ask Google Maps for Lunch and It Will Now Order It - the Shopfront Just Moved Inside the Map
On Thursday 6 August Google switched on food ordering inside Ask Maps, the AI assistant built into Google Maps. Ask it for a dish - Google's own example is spicy pad kee mao - and it finds nearby restaurants that serve it, then adds the dish straight to your cart, "so you can easily review and complete your order". Launch partners are Square and Toast, with Uber Eats "coming soon". For Square sellers the checkout runs through Order by Cash App and the order lands in the existing Square POS and kitchen display exactly like any other online order - with, in Square's words, "no added setup, contracts, or fees required". Menus, hours and location details sync automatically from the Square Dashboard; the only thing a seller needs is an active Google Business Profile. Toast's orders carry no third-party commission either. Ordering is US-only at launch, "with more countries to come over time" - the assistant itself is rolling out to over 150 countries, but not yet the transactional part.
Regular readers have seen both halves of this story before, separately. In April, when Google launched its Universal Commerce Protocol with Shopify, Stripe, Visa and Mastercard, the note attached was that Google had explicitly said small businesses were next. In July, Square wired ChatGPT and Claude into the till and we called it the first mainstream case of AI agents routing around the delivery marketplaces. This week the two merged: Google, Square and Toast are extending UCP to food as an open spec for how an AI agent reads a menu and fills a cart, and the order now starts in the map - the place local customers were already looking - rather than in a chatbot they may not use. A Brooklyn roaster quoted in Square's announcement put the operator's side of it in one line: "Increasingly, that connection begins before someone ever walks through our doors."
What this means for you: Nothing switches on in the UK this week, and no date has been given. But notice what just happened to two pieces of admin you already do. Your Google Business Profile - the hours, the menu, the photos you update when you remember - has quietly become a shopfront an AI can walk into and buy from, which turns the weekly Google routine we've recommended for reviews into commercial infrastructure rather than reputation management. And your menu data has become the stock on its shelves: an assistant can only sell what it can read, so wrong prices, dead items and a PDF menu are invisible to the entire channel. The commission arithmetic is still the quiet headline - orders arriving through your POS at no marketplace commission, against the 25-35% a delivery platform really takes, is exactly why Square and Toast are fighting to be the rails here. When this crosses the Atlantic it will favour, as ever, the operators whose data was already clean. The hour spent getting your profile and menu current is the cheapest bet in this week's paper.
Read the announcement on Google's blog ->
Sweetgreen Has 35 Robot Kitchens and a Demand Problem
On Thursday 6 August, Sweetgreen - the 287-site US salad chain whose robotic Infinite Kitchen this column has tracked all year - reported a second quarter that reads like a controlled experiment in what automation can and cannot fix. Revenue rose 3.8% to $192.7m, but same-store sales fell 6.2%, restaurant-level margin dropped from 18.9% to 13.1%, and the net loss came to $26.3m. Shares fell around 15% in after-hours trading. All of this with 35 restaurants now running the robotic makeline that assembles up to 500 bowls an hour against a human's 30 to 45, and with about half of this year's planned openings set to feature it. Chief executive Jonathan Neman conceded that "our results are not where they need to be", while insisting the plan is working.
The second half of the story is crueller, and it is the half with the lesson in it. Management told analysts that a multistate cyclospora outbreak - one not linked to Sweetgreen's own supply chain, a distinction that has done the company no good at all - has cut consumer demand for fresh prepared food since mid-July, and is expected to take 200 to 300 basis points off full-year comparable sales and $7m to $10m off adjusted EBITDA. Sit with that for a moment. The most advanced kitchen automation in fast casual, and the shape of its whole year now turns on whether people feel like salad. The robots make the bowls faster than any human ever will. They cannot make anyone want one, and they cannot shield a fresh-food brand from a fresh-food headline that was never about it in the first place.
What this means for you: In May, when Wonder was putting this same robotic line behind 100 locations, this column's position was that robotic kitchens compete on price and indies compete on reason-to-come-in. This is the first results day to test that position with real numbers, and the position held - file it alongside the 60% of small firms who told Capify that AI hasn't moved their needle as evidence that the machinery is not the moat. But do not read it smugly, because the transferable lesson cuts both ways. The robots' genuine virtues - identical portions every time, a process that never gets sloppy on a busy Saturday - are available to you for pennies: portion control done properly is most of what the arm does. And the cyclospora detail deserves respect rather than schadenfreude: a food-safety scare you had nothing to do with can still empty your counter, and the operator whose hygiene rating and records are visibly excellent is the one customers still trust when a category headline turns ugly. Copy the discipline. Skip the robot arm.
Read the results on StockTitan ->
Read the earnings call coverage on Investing.com ->
A Fortnight Left on the Zero-Hours Rules - and the Numbers Being Argued Over Are Your Rota's
The "ending one-sided flexibility" consultation - the piece of the Employment Rights Act that we broke down in June, covering guaranteed-hours offers, shift notice and cancellation pay - closes at 11.59pm on Tuesday 25 August. On Friday UKHospitality published its formal response, and the numbers in it are worth a cafe owner's attention because they are the numbers that decide whether these rules apply to your team at all. The trade body wants the low-hours eligibility threshold set at 8 hours a week or fewer; the government's consultation document offers options from 8 to 48 hours and states a preference for somewhere between 8 and 20. It wants the reference period - the window of actual worked hours a guaranteed-hours offer would be calculated from - set at 26 or 52 weeks rather than the government's preferred 12. It wants a regularity test, so the duty only applies to someone working at least 10 of every 12 weeks and 20% above their contracted hours. And it wants commencement pushed to October 2028, against a current expectation of 2027. Chief executive Allen Simpson's warning: "If Government does not get this right, it risks creating barriers to employment, particularly for young people, and restricting the ability of businesses to adapt to their customers' needs."
Translate those abstractions onto a rota and you can see what is actually being fought over. At a 20-hour threshold, most part-time baristas in the country are in scope; at 8 hours, most are out. A 12-week reference period that happens to land across your summer captures peak-season hours as someone's "regular" pattern and obliges you to offer them in November - which is precisely the seasonal-trade problem a 26 or 52-week window would smooth out. Worth holding onto two calmer facts as well: the duty being consulted on is a duty to offer a contract reflecting real hours, which a worker is free to decline, and the fine detail here comes from an 84-page consultation document, so the 8-to-20 preference and 2027 timing are the legal profession's reading of it rather than settled law. Nothing is settled. That is rather the point of the next fortnight.
What this means for you: Two moves before the 25th, and neither is difficult. First, respond - not through gritted teeth, but because this is the rare consultation where the government is explicitly choosing between numbers, and a paragraph from a real cafe describing a real week ("I employ four students who work 6 to 12 hours depending on their timetables; here is what a 12-week summer reference period would do") is exactly the evidence the process is short of. UKHospitality is urging individual operators to reply directly, and the GOV.UK page takes responses by online survey or email. Second, keep building the record, because whatever thresholds land, guaranteed-hours offers will be built from evidence of actual hours worked - the audit of real patterns against contracts that the Employment Rights Act guide walks through is the preparation that survives every possible outcome of this consultation. And if your rota still runs on last week's habit rather than what the till says you need, fixing that got more valuable again this week: hours you commit to in writing are hours you want to be certain you can fill profitably.
Read UKHospitality's response ->
Respond to the consultation on GOV.UK ->
The Drought Has Repriced Milk - and September's Invoices Are Already Written
Here is a price rise with a date on it. Farmers Weekly reported on 31 July that the drought now covering much of England and all of Wales - seven Environment Agency areas are in official drought status, England took roughly 7% of its usual July rainfall, and Wales has had its driest July on record - has hit milk production hard enough that processors are repricing in a wave. NFU dairy board chairman Ian Harvey: "It has severely restricted grass growth and diminished pasture quality, with 18.5m litres wiped off milk yields in recent weeks." The farmgate moves are specific and mostly dated 1 September: Freshways, a major liquid-milk processor for the foodservice trade, is raising its price by a minimum of 3p a litre; First Milk adds 2p to 34.85p; Muller adds 0.5p to 35p; Arla already lifted August by 0.88p to 38.15p; and Barber's is adding what Farmers Weekly calculates as a combined 4.14p across August and September. AHDB's price board logged a further round on 6 August, with cheese contracts up as much as 3.63p. One honesty note: these are farmgate prices on different contracts and different months, so resist comparing them line to line - the signal is the direction, and the direction is unanimous.
If that direction feels like whiplash, it should. In May this column reported cream down 53% and butter down 41% on the year, and suggested the input curve had finally bent in your favour. It has now bent back, on weather, in about ten weeks - the second summer running that drought has done this. And the way it will reach you is the way milk pricing always reaches you, which is to say silently: not a letter announcing 10%, but a case price that ticks up a few pence somewhere between the dairy's September review and your October statement. Farmgate is not your price - you buy finished, delivered milk at a wholesale rate - but farmgate is the cost underneath that rate, and a 2-3p farmgate rise on the liquid processors that supply hospitality is how a 2p step on your invoice gets scheduled. The exact lag varies by supplier and contract. The arrival rarely gets announced.
What this means for you: Last Tuesday we published a full guide to the milk bill - the second-biggest ingredient line nobody negotiates - and this news gives its five-minute discipline a deadline. Work out your true delivered cost per litre this week, before the rise lands, because a baseline taken now is what makes September's move visible instead of deniable; on the guide's 8,300-litre example cafe, every 2p step is £166 a year, and this wave looks like more than one step. Then watch the September and October invoices line by line, and if a rise letter does arrive, treat it as the opening of the negotiation the guide walks through - your annual volume in litres, a written review schedule, plant milks in the same conversation - rather than a fact to absorb. The one place this must not end up is in stale menu maths: milk is in almost everything a coffee-led site sells, and a per-litre rise you never recost is a margin cut you chose not to see. Our free recipe costing tool will do the per-drink arithmetic if you don't already have it pinned down.
Read the full story on Farmers Weekly ->
Track the processor moves on AHDB's milk price board ->
Greggs Grew Profit 20% in a Market That Shrank - the Interesting Part Is How
Greggs' interim results, published 29 July, were the good-news story of the sector's summer, and the market treated them accordingly: first-half sales up 7.2% to £1,101.5m, like-for-like sales in company-managed shops up 2.1%, operating profit up 22.9% to £86.5m, profit before tax up 19.7% to £76.0m, and the shares closed up around 18% on the day - a move flattered by a heavily shorted register, with around 14% of the shares out on loan, and one Greggs itself tempered by crediting a soft 2025 comparator and warning that second-half profits will dip without a consumer recovery. The estate reached 2,773 shops at 27 June, with 100 to 110 net openings planned this year plus ten trials of a self-service "Greggs Express" format. But the statistic that should stop a cafe owner mid-scroll is the market one: across the whole UK food-to-go market, visits fell 1.9% in the twelve months to June. Greggs grew its share of a shrinking market, taking it to 8.7% of every food-to-go visit in the country.
How, is the useful question, and the results statement answers it in the least glamorous terms imaginable. The Greggs App was scanned in 31% of company-managed transactions, up from 25.7% a year ago - meaning Greggs now has customer-level data on nearly a third of everything its own shops sell. Its new forecasting and replenishment system "utilises machine learning and is supporting a reduction in waste costs". Robotic picking of frozen goods starts rolling out at its Derby distribution centre in the coming months, and "our use of agentic AI has accelerated" in its engineering and support teams. Nothing in that list is customer-facing wizardry. It is measurement and logistics, pointed at demand Greggs has confirmed it already has - the same statement notes customers shifting to made-to-order drinks and protein-led choices, hence the matcha lattes. Set it against the first story of the week and the contrast does the arguing: Sweetgreen automated the assembly and lost money; Greggs automated the forecasting and made it.
What this means for you: The comparison that matters is not 2,773 shops to your one - it is that every tool on Greggs' list has a cafe-sized version that costs nearly nothing. Their machine-learning demand forecast is your production plan that decides how many brownies to bake so the bin stays empty at close. Their 31% app-scan rate is them paying handsomely to learn their sales mix - yours is already sitting in your till, free, waiting to be read properly. And their share-of-a-shrinking-market number is the honest frame for autumn: when visits fall, trade does not vanish evenly, it consolidates towards operators who know their numbers and give people a reason to come in. You will not beat Greggs on price, and should not try - a £4 flat white next to their coffee deal wins on the things a 2,773-shop chain structurally cannot do: the room, the welcome, the food with a story. What you can copy, this week, is the discipline of measuring - because that, not the sausage roll, is what their results actually reward.
Read the interim results on Investegate ->
The Brikly Take
Put the week's five stories in a row and every one of them lands on the same square: whose numbers are current when it matters. Google's assistant can only order from you if your menu data is clean. The zero-hours rules will be applied to whatever your rota records show. The milk rise will be caught, or missed, on a September invoice line. And the fortnight's two results statements make the sharpest version of the point this column has ever had to hand: the chain that automated its kitchen lost money on falling demand, while the chain that automated its forecasting, its replenishment and its knowledge of what sells grew profit by a fifth in a shrinking market. The machinery that paid was not the arm that makes the food. It was the measurement that knows what to make, what it costs, and what people are actually buying.
That kind of measurement has always been available to the chains, at chain prices - Greggs' version involves an SAP migration and a head-office data operation. The point of this column, most weeks, is that the independent needs the same knowing at a cafe-sized price, and this was a week that showed exactly which pieces matter.
The machinery will keep coming - to the map, the kitchen, the statute book and the dairy. Let the chains argue about which of it to buy. Your edge is the same as it was last week, only better evidenced: know your own numbers, and know them first.
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.