For nearly a century autonomy has been just around the corner, and the professionals who design, power, regulate and invest in the built environment have, reasonably, stopped listening. Yet the authors finished this paper in the back seat of a Waymo, watching a delivery robot push its way down a crowded sidewalk. By March 2026 that service was selling 500,000 paid rides a week across more than ten US cities. Two clocks are running at incompatible speeds: fleets scale in months, while the institutions that govern cities reform in years.
Technologists have spent years defining the conditions under which a vehicle is safe to operate, its Operational Design Domain. No one has defined the institutional conditions. We call those the institutional ODD — whether a city has the authority, the information and the oversight to govern what arrives on its streets. Federal safety rules are silent on curb access, depot siting, grid load and data access; and where a state preempts, as Texas has with SB 2807, a city loses the power to set conditions while keeping the obligation to absorb the results. The decisions that matter are being taken this year under other names: a zoning change, a sidewalk permit, a grid connection.
The window is closing. Conditions set before operators arrive are operating requirements; the same conditions imposed once they have scaled are regulatory burdens, fought over in court. The terms of the first operating licences become the baseline for every negotiation after them, the first depot sites fix where fleets run and where the grid gets built, and the information asymmetry between operator and city compounds daily.
What is at stake falls across five levers — fiscal, energy, real estate, spatial allocation and legitimacy — and on every one of them the same technology can produce a gain or a loss. None of the instruments is complex. We estimate that early governance is worth $2.5–5bn a year to a global city and $40–125m to a mid-size one; with no governance, both go negative.
The paper is organised as one diagnostic, one toolkit and one action plan. The diagnostic scores a city's institutional ODD across four domains (Flows, Space, Power, Data), and the toolkit, the Governability Stack, sets out the four kinds of instrument that work across them (Sensing, Rules, Economics, Design). The action plan opens in the first 90 days with a Minimum Viable Operating Picture — the smallest set of capabilities that lets the critical instruments function, built on data frameworks cities already run for micromobility and ride-hailing.
At this stage regret is asymmetric. A city that acts and turns out to be wrong has spent a few staff-weeks building capacity it needed anyway. A city that waits and turns out to be wrong will retrofit governance from a weak position, against operators already embedded on terms set somewhere else. Cities learned this with Uber and Airbnb: what would have cost a few staff-weeks in 2015 is still being settled in court a decade later. The remaining work is institutional, not technological, and it can begin under existing authorities.
