Robotaxis in the construction zone, and an ice-cream cartel: two stories about how markets actually work

On a Tuesday in mid-June, two unrelated wires landed within hours of each other and refused to sit apart. One described a robotaxi stalled in a construction zone on a US city street, passengers stranded while dispatch routed no human backup. The other described a small group of dairy executives indicted for quietly carving up the soft-serve market at summer camps and amusement parks: an ice-cream cartel, in the plainest possible terms. The conventional read treats both stories as consumer-protection vignettes. The structural read treats them as governance stories, and it is the more useful one.

The robotaxi story is the easier half to summarise. A vehicle operated by one of the major autonomous fleets entered a stretch of road undergoing repaving, lost its lane markings, and stopped. The dispatch system did not flag the vehicle for human rescue within a service-level window that passengers reasonably expect from a taxi. The story travelled because riders posted video, not because regulators opened a docket. The regulatory conversation that followed hinged on a narrow question: is this a product defect, an operator failure, or a permitting failure. None of those frames quite captures what is actually happening, which is the steady delegation of street-level discretion to software that cannot read a flagger's hand signal and does not yet know when to ask for help.

The ice-cream cartel is the harder half, and the more revealing. Federal prosecutors charged a group of executives and their companies with conspiring to allocate customers, fix prices, and rig bids in the institutional ice-cream market: the tubs and cones that end up at schools, amusement parks, and chain restaurants that nobody photographs for Instagram. The numbers, when they surface in court filings, are modest compared with the technology platforms above them on the antitrust docket. The conduct is not modest at all. Allocation of accounts and bid-rigging are the textbook predicate acts of cartel behaviour, and the case reads like a textbook. What makes the story more than a curiosity is that the market in question is one most people have never thought about as a market at all. Soft serve at a county fair is not a consumer choice; it is a concession contract. The discipline in that market is set by procurement officers, not by shoppers, which is exactly why a cartel can live there quietly for years.

Taken separately, each story is a routine item. Taken together, they sketch a picture of where market discipline in 2026 is actually being generated, and by whom. In neither case did the people most often described as the regulators of these markets do the catching. The robotaxi stall ended up in public view because riders had phones and the algorithm had no flagger. The ice-cream conspiracy ended up in court because a procurement officer, somewhere, decided that the bids had stopped making sense and picked up the phone. The through-line is mundane and worth stating plainly: in markets where consumers cannot easily observe the terms on which they are transacting, and where alternatives are bundled, opaque, or concession-locked, the only enforcement that scales is the kind that someone other than the buyer eventually chooses to perform.

The harder question is what this implies for the way these markets are governed. The default instinct in US policy circles is to reach for a familiar lever. In the case of autonomous vehicles, that lever is state-level permitting tightened until cities become, in effect, slow-rolling test tracks for software that the public has not consented to absorb. In the case of the ice-cream case, that lever is a steeper per-violation fine and a longer tail of corporate monitor obligations. Both levers do something. Neither lever addresses the underlying problem the two stories share, which is that the prices, the routes, and the choices that matter most in these markets are set by a small number of actors whose conduct is hard to monitor from the outside.

The political economy of that concentration is not mysterious. Autonomous fleets consolidate because the capital cost of a perception stack and a safety case is high, and because the marginal cost of adding a vehicle, once the stack exists, is low. Institutional ice-cream consolidates for older and duller reasons: refrigerated distribution is expensive, route density is everything, and a regional dairy that loses a school district cannot easily win it back the following season. Both markets tilt towards a small number of incumbents not because anyone planned it that way but because the cost curves reward scale and punish complaint. The consumer-protection frame treats this as a problem of bad actors; the structural frame treats it as the expected output of the underlying economics.

A governance frame that took the structural reading seriously would look different on both files. On robotaxis, the centre of gravity would move from vehicle-level certification to fleet-level service obligations: minimum human-handoff response times, audited publicly; a published incident ledger per operator per city; a meaningful municipal seat at the table when expansion permits are reviewed. None of that is exotic; all of it is currently optional in most jurisdictions. On the institutional ice-cream side, the centre of gravity would move from headline fines to procurement-side transparency: standardised bid templates for public-sector concessions, and rotation requirements long enough to keep a cartel honest without punishing efficient incumbents. The point is not to redesign these markets from a whiteboard. The point is to align the regulator's attention with the place where the discretion actually lives, which in both cases is upstream of the consumer, not at the counter.

The two stories also sit, more quietly, inside a longer argument about how antitrust and platform governance have drifted apart over the past decade. Antitrust rhetoric has spent its energy on the obvious targets: the consumer tech platforms, the app stores, the search defaults. The ice-cream case is a useful reminder that bid-rigging in a procurement-heavy commodity market is still bid-rigging, and that the agencies tasked with policing it can still win a case when they decide to bring one. The robotaxi story is a useful reminder that the platforms of the next decade will not all look like the platforms of the last one, and that the policy vocabulary built around app stores does not automatically transfer to a fleet of cars on a city street. What the two stories have in common is a buyer who cannot see the price being set and cannot easily switch. That is the condition under which markets stop disciplining themselves, and it is the condition any serious framework has to put at the centre.

For now, both files will move at their own pace. The robotaxi fleet will patch its dispatch logic, regulators in a handful of states will tighten their permitting language, and the conversation will tilt back to whether the technology is safe rather than whether the service is owed. The ice-cream defendants will plead, settle, or go to trial, and the procurement officers who quietly feed those cases will go back to their spreadsheets. The interesting question is whether, six months from now, anyone in either ecosystem will be asked the structural question: who, exactly, is supposed to be watching this, and on whose authority. The wires did not ask it this week. They rarely do.