Two rooms, two days, one dependency
On 1 September, the Facultad de Ciencias Económicas of the Universidad de Buenos Aires ran the fifth edition of its Congreso Internacional de Ciudades Inteligentes. The day opened with the inaugural lecture by Carlos Moreno, professor at Université Paris 1 Panthéon-Sorbonne and author of the fifteen-minute city, and with the conferral on him of the Doctorado Honoris Causa of the Universidad de Buenos Aires — a deliberate statement about which body of urban thought Argentina's largest public university wishes to be associated with. Three panels followed, with a meeting of the Argentine GovTech ecosystem. The speakers were mayors: Ezeiza, Tres de Febrero, Escobar, Vicente López, Lomas de Zamora, Tucumán, Córdoba.
On 2 September, the city government convened Urban Cities Buenos Aires 2026 at the Usina del Arte, billed as the first federal smart-cities congress: more than 300 cities called, 24 provinces represented, over 65 speakers, opened by the head of city government and closed by his chief of cabinet. Three of the mayors on Tuesday's stage were also on Wednesday's. Two registers in one week: the university's evaluative, the government's executive.
The Wednesday programme is the more revealing document, because it is longer and more specific. Twelve panels ran across three rooms, and classified by what they promise the split is uneven in an instructive way. Seven name a domain: urban security and emergency management, smart agriculture, real-time cities, climate resilience, mobility, local digital services, and a showcase of local experiences. Five name a constraint: urban data and interoperability — the programme calls it, precisely, the invisible infrastructure of the smart city — responsible governance and citizen trust, artificial intelligence in government from pilot to real impact, federalism between local governments, and urban cooperation across levels.
That classification is a judgement, worth stating plainly so it can be argued with. The dependency it exposes is not. A mobility programme that cannot reconcile a bus identifier across two jurisdictions is not a mobility programme, and a climate dashboard whose flood thresholds differ between two adjacent municipalities produces two incompatible warnings for one river. Every domain is downstream of every constraint, and a programme that gives them parallel slots has not yet said so out loud.
An agenda that lists the constraint beside the thing that depends on it has described the problem correctly and scheduled it incorrectly.
The Dependency Reading
Socradata reads any public-sector innovation agenda — a congress programme, a ministry work plan, a budget line — through three levels. The method is deliberately cheap: it needs no data, only the document and a willingness to sort its items into what a city wants and what it must first be able to do.
Seven describe what a city wants to do; five describe what has to be true first. Only the second group left the building without a named owner.
What the city wants to do. Mobility, security, waste, climate, licensing, lighting. Visible, fundable, with a ribbon-cutting and a constituency. They are also the only items a congress can reliably fill a room for, which is why every urban agenda over-weights them. The error is sequencing.
What has to be true first. Shared data definitions, an interoperability profile, decision rights over automated outputs, evidence that a pilot survived contact with production. Invisible and hard to fund, because the benefit accrues to the initiatives that come after rather than the one that pays. This is also where the Wednesday programme was most honest: naming the invisible infrastructure of the smart city, and the distance from pilot to real impact, is the correct diagnosis stated in public.
The document that converts a constraint into something a municipality can take home: a published specification with a version number, a named owner, a conformance test and a licence. Not a platform or a memorandum — a specification is the only artefact a city of three million and a town of fifteen thousand can both implement without one inheriting the other's operating cost. This level was absent from both programmes, and its absence is the week's real finding.
The spanning metric is the dependency ratio: funded domain initiatives divided by constraint items that carry a named owner and a delivery date. Above 1.0 and rising, an administration is accumulating stranded pilots faster than it is building the capability to run them; at or below 1.0, each new domain initiative arrives on ground that has already been prepared. Applied to the Wednesday programme as published, the numerator is seven and the denominator is zero: no constraint panel ended with a named owner. That is not a criticism of the organisers, who convened the right conversation. It describes what a convening can and cannot produce. Interoperability or it doesn't scale is easy to say from a stage and expensive to schedule.
So what: a smart-city agenda that funds seven domains on top of five unowned constraints is not building a smart city. It is buying seven pilots that will each rediscover the same missing definition.
Three patterns from the region
Two of these are public and documented; the third is an anonymised composite from operating work, with figures stated as illustrative targets. Each names the system, the decision loop, the human override path and the artefact that made the outcome portable.
São Paulo — the constraint layer packaged for export. System: Prodam, the municipal ICT company founded in 1971, which runs the city's infrastructure, systems development and cybersecurity across more than 2,200 production systems. Decision loop: which technology a municipality procures, and on what legal path. In 2025 Prodam launched the Prodam Store, a hub through which any Brazilian city can acquire solutions already proven in São Paulo — consultancy, licences obtained through operational agreements, and complete systems including Smart Sampa — with simplified contracting that does not require opening a tender. Within two months managers from 150 municipalities had engaged. The exportable asset is not the software but the procurement path and the institutional form: a city that owns its technical capability can lend it to cities that cannot build one. Its president, Francisco Forbes, who spoke at the Buenos Aires congress, put it plainly: for the first time a municipal public company is taking the know-how of one of the world's largest metropolises to other cities.
San Miguel — the thing that actually travelled. System: a citizen alerting scheme running on WhatsApp, with a municipal operations centre on the receiving end. Decision loop: how a resident's report of a suspicious event reaches a dispatcher, and how quickly. Ojos en Alerta began in San Miguel in 2016 and now runs in more than a hundred Argentine cities; the province of Córdoba signed a collaboration agreement with the municipality to adopt it. Override path: a trained human operator triages every message, the design decision that keeps the scheme defensible. What made it portable was not sophistication but the opposite — no new hardware, no bespoke integration, no data-science team, and it moved between jurisdictions on a signed agreement rather than a procurement. A representative of the programme sat on Tuesday's panel at the UBA. The lesson is uncomfortable: the most replicated urban technology in the country is the one with the lowest technical ambition and the clearest institutional wrapper.
A Southern Cone metropolitan region — the domain funded ahead of its constraint. System: a real-time mobility dashboard drawing position and passenger-count feeds from four adjacent municipalities. Decision loop: frequency adjustment on shared corridors. The domain initiative was funded, delivered and demonstrated on schedule; the constraint went unaddressed because no one owned it. Each municipality defined a service interruption differently, so the same delayed vehicle appeared three times in one jurisdiction's counts and not at all in another's. The dashboard was accurate about four incompatible realities. Override path: staff learned to disregard the cross-boundary figures entirely — the moment a system stops being used without ever being switched off. The fix was administrative and cost a fraction of the build: one shared definition of interruption, headway and occupancy, versioned and agreed by four transport secretaries before further integration. Illustrative target once shared: cross-boundary reconciliation from a multi-week manual exercise to same-day, an industry estimate. This is the shape of POC theater in the public sector — not a failed pilot, but a successful one nobody can extend.
What a federal congress should leave behind
Argentina has more than 2,300 local governments in the federal registry maintained by the Ministry of the Interior. Urban Cities convened around 300 of them — a substantial achievement, and roughly one in eight. The other seven in eight are why the constraint layer matters more, and the regional evidence is blunt about who they are: CAF finds that roughly 74% of Latin Americans live in municipalities of fewer than 500,000 people, and that municipalities under 20,000 — about 19% of the regional population — are the weakest link in the service chain, held back by connectivity gaps and by capacity that cannot absorb platform costs.
That is the design constraint. A solution that requires a data team cannot be federal, whatever the banner says. The deliverable of a federal congress is therefore not a catalogue of what the largest cities have built; it is a small number of specifications the smallest city in the room can implement with the staff it already has, published with a version, an owner and a conformance test. The Prodam Store is the closest working example in the region, and its most transferable feature is the least technical one: a lawful contracting path that does not oblige every town to run its own tender. From pilot to policy is the difference between a congress that produces announcements and one that produces an artefact with a version number.
So what: the measure of a federal smart-cities congress is not how many cities attended. It is how many left with something a town of fifteen thousand can implement on Monday without hiring anyone.
Governance
One control, owned above the municipality. No domain initiative enters the budget without its upstream constraint carrying a named owner, a version and a delivery date — in the funding decision itself, not an annex. The owner sits at the provincial secretariat or inter-municipal body, never with the vendor, because a specification owned by a supplier is lock-in with a schema attached. Each specification ships with a conformance test a municipality can run itself, a reuse licence and a published change log. Where an automated output affects a citizen — a fine, a benefit, a priority in a queue — decision rights and the human override path are part of the specification, not a separate ethics document. Interoperability or it doesn't scale, and at federal level that is a procurement clause, not an aspiration.
KPIs
Dependency ratio: funded domain initiatives divided by constraint items with a named owner and a date. Baseline is usually undefined because the denominator has never been counted; target at or below 1.0 within two budget cycles. Second-adopter cost: what the second municipality spends for the same outcome, as a share of the first. Target below 40% by the third adopter; if it is not falling, a project was funded rather than a capability. Specification coverage: funded initiatives whose data definitions, schema and metrics are published and versioned. Baseline typically zero; target 100% on anything inter-municipal. Adoption floor: network municipalities under 20,000 inhabitants able to implement with existing staff, against the CAF finding that this cohort is the weakest link. Time to conformance: days from adoption to passing the published test, reported by municipality.
12-month roadmap
0–90: sort the urban technology portfolio into domain and constraint items and compute the dependency ratio; pick the two constraints blocking the most funded work; name an owner and a date for each, and publish both names. 90–180: publish version 1.0 of the two specifications with a conformance test and a reuse licence; recruit three pilot municipalities of different sizes, one under twenty thousand inhabitants; measure second-adopter cost against the first implementation. 180–360: make conformance a condition of provincial co-funding for any new domain initiative in scope; publish time-to-conformance by municipality; open a shared contracting path on the Prodam model so a small town can acquire a proven implementation without its own tender.
The agenda was right. The sequencing was not.
It is worth being fair to both convenings, because being clever about conferences is a cheap trade. The week did something difficult: it put three hundred cities and twenty-four provinces in one building, and named its own binding constraints from the stage in language an operator would recognise. The invisible infrastructure of the smart city. From pilot to real impact. Nobody was pretending the hard part was the sensor.
What neither room produced — and what no convening can produce on its own — is the artefact that survives the week. A university confers a doctorate and generates evaluation that no one is obliged to operationalise. A government convenes three hundred cities and generates commitments no one is obliged to measure. The two failure modes are symmetrical and share one unglamorous remedy: a published, versioned specification with a named owner, sitting between the evaluation and the commitment, cheap enough for the smallest municipality to adopt and precise enough that conformance can be tested rather than asserted.
There is an irony worth naming, and it is not at the university's expense. The idea the UBA honoured on Tuesday travelled further than almost any urban concept of the last decade for a reason unrelated to its elegance. The fifteen-minute city is portable because it is specified: a threshold, a mode of travel, a unit of territory, and a test any mayor can run against their own map without commissioning anything. It is closer in kind to a conformance test than to a manifesto. A congress whose commitments are to travel as far would do well to copy the form, not only the vocabulary.
This is the layer Socradata works in. Socradata transforms ERP, WMS and supply-chain data into predictive intelligence and governed operational decision systems, and a metropolitan region is the same problem at another scale: many operators, incompatible definitions, decisions that have to be made anyway. The useful first step before the next urban technology initiative is not a platform evaluation but a dependency reading of the portfolio you already fund — which we run as an operational diagnostic, at sergio@socradata.com.
Count your constraints before your domains
Every Wednesday, The Operational AI Dispatch takes one consequential AI signal and translates it into an operating model, a KPI set and a practical action plan for leaders running enterprise operations, ERP, WMS, supply chains and public systems. Published weekly by Socradata from Buenos Aires and New York.