Three paradigms, one professional
What planning history tells us about the transition we are actually in
This is Article 2 in a nine-part series: The Urban Professional in the Age of AI, based on The Urban Operating System.
1. The gap · 2. Three paradigms ← you are here · 3. The machine room · 4. Project memory · 5. Options and numbers · 6. The digital twin · 7. Participation · 8. Five levels · 9. The question behind the tools
A housing strategist sits across from her alderman. The programme they are reviewing has been in development for three years. The financial model is based on land values from 2022. The stakeholder analysis reflects a coalition that has partly dissolved. The scenario assumptions were built on a policy framework that was revised eight months ago. None of this is unusual. It is the normal condition of urban development work: by the time the analysis is complete, the world it was analysing has moved on.
She knows two things at once. First, that better tools now exist to reduce this lag: she has seen what AI-assisted scenario modelling can do with live transaction data. Second, that her organisation is nowhere near ready to use them in a way that she would trust in front of an alderman, a council committee, or a judicial review. The data is fragmented. The governance framework does not address AI-assisted outputs. Her team has no shared verification routine.
She is not caught between the past and the future. She is caught between three different institutional logics that her profession has accumulated over the past century, none of which has fully replaced the others, and all of which make demands on her at once.
That is what this article is about.
Planning has gone through several major reorientations in what it is for, what constrains it, and what counts as professional competence. The argument here is not that history moves in three neat stages. It is that three institutional logics now coexist inside most planning organisations, and that understanding how each emerged helps explain what the current AI moment is actually asking of professionals.
Understanding this pattern is not an academic exercise. Each major shift changed what the core constraint was, which changed what the most valuable skill was, which changed what it meant to be good at the job. Getting that analysis right is the difference between navigating the current transition deliberately and being shaped by it without realising it.
The first constraint: information lag
For most of the twentieth century, urban planning operated in what planning theorists call the rational-comprehensive mode. The logic was linear and optimistic: gather information, analyse it systematically, produce a plan that reflects the best available knowledge about how cities work and what they need. The ambition was scientific. The planner was an expert, applying rigorous methods to complex problems in the public interest.
The core constraint in this mode was information lag. By the time a plan reflected reality, reality had already moved on. Population projections went stale. Land markets shifted. Infrastructure decisions made on the basis of twenty-year forecasts collided with ten-year election cycles. The fundamental problem was not the quality of the analysis but the speed of the world relative to the planning apparatus.
The professional identity that emerged was the expert planner: trained in quantitative methods, versed in land use economics, authoritative on technical questions. The tools were projections, models, master plans. The legitimacy of the work rested on its analytical rigour.
The second constraint: coordination cost
The communicative turn of the 1980s and 1990s was a response to a double failure. First, the master plans of the post-war decades had produced environments that many communities experienced as alienating, unjust or simply wrong. Technical rationality had optimised for measurable variables and missed what actually mattered to people who lived in the places being planned. Second, the shift toward pluralistic, rights-conscious democracies made top-down expertise politically untenable in a new way.
Planning theorists, drawing on Habermasian communicative action theory, proposed a different account of what planning was actually for. Not the production of technically optimal plans, but the facilitation of legitimate collective decision-making. The planner was no longer primarily an expert who knew the right answer. The planner was a facilitator who could help diverse actors with conflicting interests navigate toward shared futures.1
There is an economic logic to this shift worth making explicit. In 1937, Ronald Coase asked why firms exist at all, if markets are as efficient as economists claim. His answer: markets are expensive to use. Finding counterparties, negotiating terms, monitoring outcomes: these transaction costs are real. Organisations exist to internalise them, replacing market friction with the cheaper mechanism of direction and authority.2 A similar logic applies to planning institutions. Where spatial development creates high coordination costs, institutions emerge to structure negotiation, authority and legitimacy in ways that markets cannot. The communicative paradigm did not abandon the first mode’s ambitions. It recognised that the primary cost preventing good outcomes had shifted from information gathering to coordination.
The professional identity this produced was not soft. It centred on listening carefully, building trust, managing conflict, and translating between technical and political registers. These are skills that take years to develop. They cannot be automated. That matters for what comes next.
A candidate third logic
A third institutional logic is taking shape, though it would be premature to declare it a settled paradigm. What The Urban Operating System called algorithmic urbanism, real-time data processing, iterative simulation, AI-assisted scenario development, represents a shift in the nature of the primary pressure on planning practice. Where the rational-comprehensive mode was under pressure from information lag and the communicative turn from coordination cost, the emerging pressure is data quality, data governance and the ability of planning institutions to work with tools that reason across complexity in ways that planners cannot replicate manually.3
One caution is essential here. AI reduces some information bottlenecks. It does not remove information lag as such. In public planning, lag sits in cadastral updates, fragmented administrative systems, missing local data, legal procedure and political timing. AI cannot solve what is not digitised, interoperable or lawfully shareable. And coordination cost has not disappeared. The hard part of contested urban development is still conflict, legitimacy and political consent. AI reduces some analytical friction. It does not dissolve distributive conflict.
What is changing is not that the older constraints no longer matter. It is that a third pressure is joining them, one that was not previously binding because the tools that expose it did not exist. Organisations that have not invested in clean, structured and searchable data cannot use the tools that depend on it. Governance frameworks that have not addressed AI-assisted decisions cannot legitimise the outputs of systems that produce them. Professional cultures that have not developed verification routines cannot reliably detect when those systems are wrong.
These three logics are better understood as overlapping conditions that coexist in most planning organisations today than as a clean historical sequence. Robert Goodspeed has argued that collaborative planning never achieved the uncontested paradigm status its proponents claimed, and the same will likely be true of algorithmic urbanism.4 The point is not that one logic replaces the next, but that each adds a pressure the previous one did not adequately address.
More than 90 percent of state chief information officers believe generative AI can enhance the citizen experience, yet only six percent report mature, scaled implementations today.5 That gap is not primarily a technology problem. It is a readiness problem. The tools exist. The institutional infrastructure to use them responsibly does not, in most places, yet.
Coase also observed that the right boundary between what is organised internally and what is delegated externally is not fixed. Businesspeople will be constantly experimenting, controlling more or less, and in this way equilibrium will be maintained. The same logic applies here. The right boundary between what a planning team handles internally and what it delegates to AI is not a one-time decision. It is a continuous experiment. Waiting is not neutral, because others are setting that boundary in the meantime.
A third time layer
Planning has always operated between two time tensions. Long infrastructure cycles, the road that takes a decade to plan and fifty years to use, sit uneasily beside short political cycles of four-year mandates and shifting priorities. Managing that tension has always been part of the work.
Current foresight work by the APA suggests how rapidly AI-related tools are entering planning practice, adding a much faster innovation cycle to the already difficult relationship between long infrastructure timelines and shorter political cycles.6 The tools available in 2023 were qualitatively different from those in 2021. Those available today are qualitatively different again. No infrastructure cycle and no political mandate adjusts at that pace.
This means that the planning apparatus, designed to manage long-term commitments under conditions of medium-term political change, is now also required to make durable institutional decisions about tools that may be superseded before those decisions are fully implemented. The organisations navigating this well are not necessarily those moving fastest. They are those that understand which decisions are reversible and which are not, and that protect reversibility wherever possible.
A less discussed knowledge shift
The shift in how planning is done is visible. The shift in how planning knowledge is accessed is less discussed but worth naming.
For the first two modes, the primary information challenge was locating relevant knowledge and evaluating its quality. The signals of quality were institutional: peer review, organisational reputation, professional endorsement. That architecture is changing. Mike Caulfield, whose SIFT framework for source evaluation has become a standard in information literacy education, identified the structural shift in an interview published in August 2025: when AI generates extensive text that no human has fact-checked, and that text is shared and acted upon, evaluating the credibility of a source no longer suffices. The question becomes not “is this a credible institution?” but “is this specific claim accurate, and how would I know?” That requires checking the claim, not just the source. Tracing assertions back to original documents. Treating AI-generated information as a first pass that requires confirmation, not a final answer that requires only formatting.7
This shift extends beyond the professional context. The residents, elected officials and developers that planners work with are navigating the same changing information environment. As AI-generated content floods every channel at negligible production cost, the quality signals that previously indicated reliable information lose their discriminating power. Stakeholders arrive at participation processes having formed views through information streams that are increasingly AI-curated and inconsistently verified. Planning has always required building shared understanding across different perspectives. The candidate third logic does not make that task disappear. It makes the ground under it less stable. The full implications belong in article 7. For now: this shift runs in both directions, changing how planners access knowledge and changing the conditions under which shared understanding must be built.
What the third logic does not make obsolete
Here is the point most commonly missed in both the enthusiast and the alarmed versions of this conversation.
The skills that the communicative turn developed were a genuine advance. The ability to listen carefully to residents who are not using planning language, to identify whose interests are structurally underrepresented in a process, to build trust with communities that have experienced planning as something done to them rather than with them: these are not skills that survive by accident. They survive by deliberate protection.
Milad Malekzadeh, writing in Cities in 2026, argues that planners remain central as curators, applying AI tools critically and creatively rather than treating them as sources of objective answers.8 That framing is right, but a Coasean reading sharpens what it actually means. Coase distinguished between two fundamentally different roles. Initiative means forecasting and making new institutional arrangements. Management means reacting to existing conditions and rearranging what is already there. The curator is an initiator: deciding what matters, designing how it will be assessed, determining whose voice needs amplifying, shaping the governance architecture within which AI tools operate. The operator is a manager: rearranging inputs within parameters someone else has defined. What distinguishes an agent from a servant, Coase wrote, is not the wage structure but the freedom with which the agent carries out her work. The planner who can only enter inputs that vendor software permits, whose analytical output is bounded by external parameter choices, is not a curator. She is operating under direction, and the parameters determine who holds it.
There is a further tension worth naming, even if its full treatment belongs later in this series. The depth of professional judgment that good curation requires is built through doing the work that precedes it: the first syntheses, the initial comparisons, the early legal reads, the entry-level tasks that teach you where the sharp edges are. If those tasks are delegated to AI before that grounding has formed, the curator role may persist in title while the judgment it depends on quietly erodes. That is not an argument against using AI in those tasks. It is an argument for being deliberate about when and how that delegation happens.
The planner who understands what a neighbourhood organisation has been through over the previous decade, who can feel when a participation process is producing compliance rather than genuine engagement, who recognises that a technically optimal scenario violates something a community cares deeply about that no variable in the model measures: that knowledge is not in the data. It is in the professional. It becomes more valuable, not less, as AI handles more of the work that does not require it.
The issue is not whether planning becomes technical again. It is whether institutions confuse what is easier to compute with what is more important to govern. Traffic flows are easier to model than fear. Sensor-rich corridors are easier to optimise than dignity. Permitting backlogs are easier to automate than democratic conflict. AI accelerates the answer that an institution is already giving. If that answer has not been examined, the acceleration is not neutral.
Who benefits, and who pays
One dimension of the transition that deserves explicit attention is the distributional question.
The organisations best positioned to benefit are those already investing in data infrastructure, governance capacity and analytical skill. Larger municipalities with established data estates, teams already experienced in structured knowledge management, senior staff who can delegate first-pass analytical work: these are the early winners. The tools compound existing advantage.
The organisations that pay are those with fragmented legacy systems, limited IT capacity, constrained procurement frameworks and junior-heavy teams whose professional formation depends on doing the first-pass work that AI increasingly handles.
There is a further dimension the distribution argument tends to understate. The third pressure does not only shift from coordination cost to data quality. It also shifts from internal professional capacity to external platform dependency. Organisations that build on vendor-managed models, whose parameters are set outside the planning institution, do not simply gain analytical power. They partly transfer the definition of what counts as optimal to a party with different interests and limited accountability. Maroš Krivý’s critique of cybernetic urbanism goes further: digitalisation of planning is not only a readiness question but a question of control. When the logic of optimisation displaces the logic of deliberation, it is not governance that is being improved but governance that is being replaced by something that looks like governance while serving different ends.9
This is not an argument against the transition. It is an argument for being explicit about its distributional consequences, and for procuring accordingly. The same planning values that the communicative turn developed, attention to whose voice is underweighted and whose interests are structurally underrepresented, apply to the profession’s own internal transition. They are rarely applied there.
What this asks of professionals trained in the second logic
Return to the housing strategist at the start of this article. She has the judgment to know when a technically clean answer is politically or socially wrong. That is the hard-won competence of the communicative turn, and it does not become less valuable in the third logic. It becomes more so, because the analytical machinery that might otherwise absorb her attention is increasingly handled elsewhere.
What she needs to add is the capacity to work in a mode that her training did not anticipate. That requires learning how data pipelines work without becoming a data engineer. It requires understanding what AI tools can and cannot do without uncritical trust in either direction. It requires developing verification routines without becoming paralysed by doubt. And it requires carrying forward the communicative turn’s central insight, that legitimacy is not a byproduct of good analysis but a condition for decisions to hold, into an environment where the analysis has become much more powerful.
Urban professionals who understand all three logics, and who can work in the third without losing sight of what the second taught about inclusion and legitimacy, are the ones who will shape how this transition actually goes. Those who treat it as a purely technical upgrade will miss its institutional dimensions. Those who resist it as an existential threat will find the transition happening around them rather than through them.
The tools are available. The institutional work is not yet done. And the professional judgment that makes both useful has to be brought deliberately into the new environment, because it will not carry itself.
Next in this series: The machine room. The operational layer of urban professional work is unglamorous, load-bearing, and the place where AI is likely to deliver the fastest and most immediate returns. It is also where most teams have not yet looked.
Notes
The communicative turn in planning theory is developed most fully in: Healey, P. (1997). Collaborative Planning: Shaping Places in Fragmented Societies. Macmillan. The original theoretical grounding draws on Habermas, J. (1984). The Theory of Communicative Action. Beacon Press.
Coase, R. (1937). The nature of the firm. Economica, 4(16), 386-405. The transaction cost argument, the initiative/management distinction, and the direction/freedom definition of the employment relationship are all drawn from this source. The application to planning institutions is an analytical analogy, not a claim Coase himself makes.
The three-mode framework and the shift toward algorithmic urbanism are developed in: Deelstra, R. (2026). The Urban Operating System. remcodeelstra.substack.com. This is the founding essay for the current series, cited here for the reader’s orientation rather than as independent external evidence.
Goodspeed, R. (2016). The death and life of collaborative planning theory. Planning Theory & Practice, 17(2), 275-281.
NASCIO & Accenture (2025). Harnessing GenAI to Elevate the Citizen Experience. September 2025. nascio.org. The survey covers US state chief information officers, not planning organisations specifically. It is used here as an indicator of institutional readiness patterns in the public sector more broadly.
APA (2026). 2026 APA Foresight Trend Report for Planners. American Planning Association. The “third time layer” framing is the author’s synthesis of current AI tool cycles and planning practice timelines, informed by the report’s treatment of AI as a fast-moving planning-relevant trend.
Caulfield, M. (2017). Web Literacy for Student Fact-Checkers. The SIFT framework is the foundational work. His adaptation for AI-generated content is discussed in: Panke, S. (2025). Interview with Mike Caulfield about Deep Background, AI literacy and future skills. AACE Review, August 2025. aace.org/review/caulfield-ai. The “epistemological shift” framing in this section is the author’s interpretation of Caulfield’s argument applied to planning practice; it should not be read as Caulfield’s own claim about planning specifically.
Malekzadeh, M. (2026). Urban planners should not be afraid of AI: the planner as curator. Cities, Elsevier, 168, article 106497. doi.org/10.1016/j.cities.2025.106497
Krivý, M. (2018). Towards a critique of cybernetic urbanism: the smart city and the society of control. Planning Theory, 17(1), 8-30.