The second week of October 2026 produced two stories about the future of prices, and almost everyone read them in separate tabs.

In the first tab: a federal antitrust lawsuit filed against McDonald’s on October 8, alleging the company uses an AI-powered “pricing engine” to set menu prices across its US locations and overcharge customers for Big Macs and fries. In the same tab, all month: electronic shelf labels spreading through Kroger, Walmart, and Amazon Fresh in the US and Tesco, Morrisons, and Asda in the UK, Sainsbury’s launching an AI feature that turns a photograph of a shopping list into a matched basket, and Revolut trialing checkout by selfie in coffee shops.

In the second tab: Stripe’s payment data showing agent-driven purchases through its Link wallet up 38x, with personal-agent platforms reportedly already processing an estimated $1 billion in autonomous purchases.

Seller-side AI and buyer-side AI, arriving in the same week, covered as unrelated news. They are not unrelated. They are the two halves of a single system: a marketplace where the price is set by a machine that estimates what you personally will pay, and increasingly paid by a machine that cross-compares what everyone else charges. Understanding what survives that collision, and what replaces comparison shopping once machines do it faster than humans can, is the real consumer protection question of the next few years. The short answer developed below: price becomes fluid, and the only fixed point left in a purchase decision is whether the product is any good.

The seller’s week: the price tag becomes a sensor

Start with the lawsuit, because it names the mechanism everyone else describes in softer language. As CNBC reported, the suit alleges McDonald’s operates an AI-powered “pricing engine” that sets menu prices across US locations, and that customers were overcharged as a result. McDonald’s has denied using AI to determine what individual customers are willing to pay, saying it provides franchisees with “tools, resources, research and recommendations to help them make informed decisions.” Whatever the court decides, the lawsuit landed on infrastructure that is visibly arriving.

That infrastructure has three layers, and each one is deployed in the name of convenience.

The first layer is the digital price display. Electronic shelf labels, which put prices on screens that can be changed instantly, are becoming standard at Kroger, Walmart, and Amazon Fresh in the US, and at Tesco, Morrisons, and Asda in the UK. Every retailer deploying them says the same thing: no dynamic pricing here. Whole Foods told CNBC it “does not use dynamic pricing” and that the labels “simplify the price tag change process for our employees [and] reduce paper waste from traditional tags.” The British Retail Consortium’s Andrew Opie said supermarkets have “no intention to use dynamic or surge pricing in their stores.” Those denials are worth having on record, and they are also worth noticing as a category: an industry installing, at scale, the exact hardware surge pricing requires, while insisting it will never press the button.

The second layer is markdown AI, the friendly face of algorithmic pricing. Kroger said in April it uses an AI platform called FlashFood to mark down perishables nearing the end of their shelf life and push them to shoppers through an app. Nobody objects to cheaper food that would otherwise be wasted. But the same system that learns when to discount is a system that learns demand curves.

The third layer is the data vacuum, and this is where Sainsbury’s SmartLists, launched October 7, deserves a careful read. The feature is genuinely useful: type “ingredients for spaghetti bolognese” or photograph a handwritten list, and the app matches items to real Sainsbury’s products, “including Nectar Prices,” so customers “can see their options and get the best value for their budgets.” Ben Edge, Sainsbury’s director of product, said SmartLists “is designed to take the effort out of creating a shopping list, whether customers are typing, speaking or taking a picture of what they need.” Add Revolut trialing facial recognition checkout in coffee shops, and the pattern across all three layers is identical: every deployment is framed as effort reduction, and every deployment also deepens the retailer’s record of who you are, what you buy, when, and at what price you kept scrolling.

Miroslava Marinova, a senior lecturer in commercial law at the University of East London, put the trajectory precisely to CNBC: “As retailers combine market-level information with increasingly detailed consumer data, the boundary between dynamic and personalised pricing becomes thinner.” Dynamic pricing responds to market conditions. Personalised pricing responds to you. The hardware erases the difference.

There is precedent for how fast this can move, and how fast it can retreat. As TheStreet recounted, Wendy’s floated a dynamic pricing pilot in 2024, and “48 hours later, that got canceled because of the backlash.” Public tolerance for openly personalized prices is low: a national poll of 2,200 US adults conducted in October 2025, reported by Analytics Insight, found 59% see price gouging as a major worry with AI-driven dynamic pricing, only 27% support the method, and more than half oppose it. Which is exactly why the deployment pattern above matters: the credible path to personalized pricing is not an announcement. It is convenience features that collect the data, on hardware that can change prices instantly, optimized quietly.

What the Bank of England actually measured

The best data on how far this has already gone comes from an institution with no product to sell: the Bank of England. In April, BoE economists Clare Lombardelli and Rupal Patel published a Bank Insights analysis on the rise of dynamic, personalised pricing. Four findings stand out.

First, the speed. Digitalization has collapsed what economists call menu costs, the expense of changing a listed price. The share of UK hotel room rates that change at least once a month has risen from around 15% in 2005 to roughly 80% today. A price that used to be a fact is now a stream.

Second, the breadth. The Bank’s Decision Maker Panel survey asked more than 1,600 firms about their pricing practices over the three months to January 2026. Market-responsive pricing, where firms “use algorithms, including artificial intelligence (AI), to adjust prices in response to data on demand, capacity or competitors’ prices,” sits at 21% of firms now and is expected to reach 31% within a year. Every sector surveyed plans to increase its use. The direction firms are drifting toward is, in the authors’ words, the textbook view of “perfect price discrimination”: “charging as close to the maximum price a consumer is willing to pay for a good or service.”

Third, the restraint is reputational, not technical. “Adoption in the UK may be limited more by reputational considerations than by technological ones,” the authors write, noting that international studies find UK consumers more likely than consumers elsewhere to consider dynamic pricing unfair. The Wendy’s episode and the 59% gouging-worry poll number explain the gap between 21% adoption and a hardware install base that could support much more.

Fourth, and most unsettling: the collusion experiments. “In some unsettling experiments, algorithms have learnt over time to charge above competitive and static prices and drifted into tacit, collusive behaviour,” the Bank notes, adding that algorithmic collusion is still rare but regulators are alert. Two pricing algorithms that never communicate can still discover that undercutting each other is bad for both. That is not a hypothetical consumer harm; it is the end state of leaving the disciplining to the machines on one side only.

The Bank also flags what this does to measurement. A consumer price index built on a representative monthly sample strains when “prices shift continually, and differently for each shopper,” so the UK’s Office for National Statistics started ingesting bulk weekly grocery scanner data in March 2026, capturing loyalty prices and person-specific discounts from retailers representing around half the grocery market. The estimated effect so far is small, around 0.03 percentage points on headline CPI between January 2022 and June 2025, and up to about 0.1 percentage point during the months food inflation ran hottest. But the Bank’s Inflation Attitudes Survey contains the psychological finding that should outlive all the others: when prices fluctuate because of AI-enabled dynamic pricing, people perceive that prices are higher on average, whether or not they are.

The buyer’s week: 38x

Now the other tab. Ability AI’s October analysis of Stripe’s data reports a 38x surge in agent-driven purchases made through Stripe’s Link wallet, transactions where an autonomous agent researches, selects, and settles payment without a human completing checkout. Personal-agent platforms are scaling with it: Instinct, a prominent personal-agent platform, is reportedly processing an estimated $1 billion in gross merchandise value, and consumer agents like Meta’s Muse now execute multi-step tasks that end in a transaction.

The Bank of England, writing from the other side of the same phenomenon, spotted this too: “AI shopping assistants already help people hunt for bargains, though benefits accrue mostly to digitally savvy consumers who use AI in their search for the lowest prices.” Read that sentence with the seller-side data and the full picture appears. One-fifth of firms already reprice by algorithm, heading for nearly a third. The defense available to consumers is an agent of their own. The consumers who have one are the digitally savvy and the affluent. The consumers being profiled by pricing engines without an agent to haggle back are everyone else, precisely the households a Consumer Federation analysis found tend to pay more under dynamic pricing: those with fewer options and lower digital literacy.

This is the two-tier market taking shape. Marinova’s warning to CNBC describes the mechanism: “If consumers cannot understand why they received a particular price, cannot compare it with prices offered to others, and cannot effectively switch to another supplier, the normal disciplining effect of consumer choice becomes weaker.” Personalized pricing dissolves comparison. Agents rebuild it, at machine speed, for the people who have them.

What happens when the machines meet

Project the arms race forward one step. Sellers deploy willingness-to-pay inference. Buyers deploy cross-comparison agents. Sellers respond with personalized offers agents cannot see across sessions. Buyers respond with agents that rotate identity, pool price observations, and share them. Price stops being a label and becomes a negotiation, conducted between algorithms, in milliseconds, over a corpus of behavioral data neither party can fully audit.

In that world, what does a shopping decision even rest on? Not the price, which is now fluid by design and personal by default. The anchor that remains, the thing a purchase decision can still stand on when every other input is adversarial, is the quality of the product: whether it works, whether it lasts, whether the five thousand people who bought it before you were real and honest. And that anchor is under active attack from the same economy that is monetizing everything else. This column has spent the season documenting it: machine-written five-star reviews sold at S$5 a post with a rating calculator and a replacement warranty, Amazon productizing review volume behind media budget, brand-authored answers inside the shopping assistant. A pricing algorithm can move what you pay in real time. A review-burst campaign can move what you believe in weeks. The shopping agent movement, 38x and climbing, inherits both distortions as structured input.

An agent can defeat a personalized price. It cannot defeat a contaminated corpus, unless something upstream of it filters the corpus first.

Regulators are drawing lines, slowly

The legal system is starting to catch the seller’s side of this. New York now requires most businesses using consumers’ personal data to set prices to disclose it clearly. Maryland has restricted food retailers and delivery services from using personalized, data-driven pricing to charge higher prices for certain food. New Jersey and Connecticut have enacted measures aimed at “surveillance pricing.” The UK’s Competition and Markets Authority has issued transparency guidance, and the FTC’s fake reviews rule has been in force since October 2024 for the quality-signal side of the ledger. The McDonald’s suit will test whether antitrust law reaches algorithmic price-setting at all.

Regulation is necessary and it is slow. The hardware is already on the shelf, the loyalty data is already collected, and the reputational constraint the Bank of England identified only binds while prices look the same to everyone. The moment they can differ quietly, the constraint quietly stops working.

The only fixed point left

Strip the week’s news to its structure and one requirement survives every scenario: the transaction needs at least one input that neither the seller’s pricing algorithm nor the review economy can manufacture. That is the design brief GoBuy was built to, and this week made it concrete:

  • Filter before aggregating. Fake, incentivized, and low-information reviews are removed before any score exists, so a burst campaign changes the count, never the verdict.
  • Score quality, not quantity. The Smart Score, 0 to 100, is computed from the information content of authentic reviews, because volume is precisely what the S$5-per-post economy and the media-funded review machine manufacture.
  • Require persistence. The GoBuy Verified badge demands a filtered score above 80 held across 90 days. A pricing algorithm can move a number in milliseconds and a review farm in a fortnight; neither can manufacture three months of clean history. Time is the one input that cannot be compressed.
  • Shortlist, don’t shelf. Seven products per category instead of forty thousand listings, because when an agent or a human compares, the comparison should be between survivors, not noise.
  • Serve both sides of the machine divide. The Chrome extension injects the trust panel directly onto the Amazon product page, for the shopper without an agent. Agents consult the same filtered evidence in one structured call to the MCP server at gobuy.ai/api/mcp, for the shopper with one.

The price tag learned your name this month. Your agent will learn to haggle with it. Make sure something in the transaction still tells the truth about the product. Check what you are about to buy at gobuy.ai, and if you build shopping agents, wire them to independent product evidence at gobuy.ai/agent-docs.