McDonald's runs AI pricing across nearly 14,000 restaurants
The pricing algorithm debate has been about airlines and ticket sites for years. It just arrived at the drive-through.
A Reuters investigation published Tuesday found McDonald's is using a machine-learning pricing engine across nearly 14,000 restaurants to generate what the company calls "the optimal price" for every menu item at every location — and that franchisees say they are being pushed to accept it. The reporting is built on screenshots of the pricing portal taken in August and interviews with nine people with first-hand knowledge of the strategy. The engine continuously analyzes millions of daily transactions and sets per-store recommendations; the interface shown to franchisees carries messages like "Your restaurant is showing medium sensitivity to price," derived in part from an estimate of "customer willingness to pay in your area." Reuters' own price check in September found a Big Mac at $5.69 in one company-run Fresno store and $6.89 at another company-run restaurant two miles away — a 21 percent gap the outlet could not attribute to the engine specifically.
The friction is not that prices move. It is who benefits when they move. Five franchisees told Reuters that corporate pressured them to adopt the tool, and a June franchisee document records deviations from the recommendations in detail; since January, McDonald's business standards have required franchisees to be "constructively engaging with McDonald's approved Pricing Consultant and Tools." McDonald's disputes the framing, calling the claims "speculative and uninformed" and describing the portal as "a tool, not a mandate." But the portal's own terms of service warn that franchisees "may be competitors of each other" and tell them to comply with antitrust law — language former FTC commissioner William Kovacic called "an acknowledgment there's a potential problem." The reason the two sides want different prices is structural: headquarters takes a percentage of franchisee revenue regardless of store margin, so a lower price that lifts volume helps corporate and hurts the operator. Recent months have seen the engine recommend more conservative pricing, including cuts, while U.S. foot traffic has fallen year-over-year every complete month since March, according to Placer.ai estimates.
DeepSeek shipped a desktop version of its agent harness, dropping the biggest barrier between its coding agent and ordinary PC work. DeepSeek Harness v0.2 arrived Tuesday as installers for macOS and Windows, with file, spreadsheet and PDF handling, side-by-side preview of files and code diffs, a plugin management page that installs plugins by package name instead of command line, and scheduled automation tasks that let the agent run on a timer. DeepSeek says roughly 60 percent of Harness users already run third-party plugins, and that by daily active users and sessions it is now the most popular coding agent among the company's own API customers. Eligible users can claim a limited six-yuan trial credit on login. The roadmap DeepSeek disclosed points past coding: long-term memory, browser and GUI automation, remote and mobile access, and multi-user collaboration, plus an official plugin marketplace.
Marissa Mayer's Dazzle launched out of invite-only beta with a thesis that your photo library, not your inbox, is the most useful context an AI assistant can have. Dazzle, which raised an $8 million seed round last December, scans a phone's camera roll to infer hobbies, travel history, food and style preferences, then splits its work between immediate tasks (turning an event flyer into a calendar entry) and suggestions (vacation ideas, birthday gifts). In her hands-on test, TechCrunch's reviewer found the suggestions notably personal — a pottery studio, a bioluminescent kayak tour — but also concrete blind spots, including an activity idea for a child who had already outgrown it. Mayer frames privacy as the differentiator, arguing users may be more willing to hand over photos than email, and says Dazzle discards anything it flags as sensitive. It is a narrower assistant than Meta's Muse, and the pitch is essentially that knowing who you are beats executing tasks quickly.
What to watch: whether any U.S. regulator opens a formal look at algorithmic menu pricing — the franchisee-as-competitor structure is exactly the fact pattern the FTC has been circling.
Would you eat somewhere that prices the same burger differently by neighbourhood — or does that cross a line? Tell us in the comments.
Sources: DeepSeek Harness (GitHub)
