Mission

Build the mandate to mediate urban change.

Cities are changing their streets, squares and neighbourhoods to meet climate, housing and mobility goals. An institution decides each change, and people live with it. We help institutions build the mandate to work out that change in public: to put real options to the people affected, show what each would do, and decide with them while the choice is still open.

145 engagements92 cities46 countriessince 2021

We delivered 91 of these engagements, from 2021 to September 2025, under the brand name of UrbanistAI, in our previous collaboration with Toretei. The other 54 are CoPlanAI’s.

Delivering change takes plans, permits and schedules. Mediating it takes two more things: a process in which the people who will live with a decision can try the options and see what each would do, and an institution with the standing to act on what they choose. That standing is the mandate. The mandate to deliver change keeps growing; the mandate to imagine it has to grow with it.

A city with no mandate to imagine its future can only administer it.

Damiano Cerrone · CoPlanAI

On this page

How can a community be engaged in planning its own future? This page follows the shift from participatory planning to collaborative planning with AI: the ideas behind it, how it works, and what it takes to act on what people imagine.

01Bridging the medium imbalance
02From the ladder to the table
03Three kinds of knowledge
04Count, consent or contest
05Two questions, four tasks
06A phone, a street, a question
07Building the mandate
08AI and democracy
09The research behind the method

01 Who gets to make the picture

Bridging the medium imbalance.

Some people at the table bring renders. The rest bring post-its.

Who holds the render

The architect brings a photoreal render. Residents bring post-its and pins. Whoever holds the picture decides. Everyone else annotates it.

A render of their own

Generative AI bridges the gap: residents make their own image of the street and put it on the table beside the architect’s.

A visualisation of the planned Irmabyen quarter in Rødovre: a street with people and cafés

Sticky notes, small photographs and red dots on a shared workshop board.

Top-down and bottom-up, both struck through.

Neither bottom-up nor top-down: the medium itself creates the chance to collaborate.
Damiano Cerrone

Sources: Damiano Cerrone, doctoral manuscript (in progress), ch. 2 “Medium Imbalance”. Images: visualisation of Irmabyen, Rødovre Kommune, CC BY 3.0 (opens in a new tab), via Wikimedia Commons (opens in a new tab); the residents’ board and an image the residents generated, a workshop in Helsinki, January 2023 (photo Damiano Cerrone / SPIN Unit).

02 The ideas

From the ladder to the table.

For fifty years we measured how far up a ladder citizens got; the collaborative turn asked who sits at the table.

Most participation stays low

In 1969 Sherry Arnstein drew a ladder of eight rungs. Most participation lives in the middle, in tokenism: you are told, you are asked, you might even be placated.

From consultation to collaboration

Three things move. Remember legitimacy: “we held a consultation” only needs a ticked box; “we built this together” needs something genuinely co-produced.

A fourth chair at the table

The institution becomes one party among others. Now there is a new chair: a fourth form of expertise.

Next, it becomes the third circle of the Venn.

Sources: Arnstein, ‘A Ladder of Citizen Participation’ (1969); Healey, Collaborative Planning (1997); Innes & Booher, Planning with Complexity (2010); quotes from Damiano Cerrone’s doctoral manuscript (in progress), ch. 5 §4.1 and ch. 7 §7.3. Point at the rungs, the moves and the seats for more.

03 The ideas · the coplanning Venn

Three kinds of knowledge.

Coplanning is the discipline of holding all three together long enough to produce a decision none of them would produce alone.

01 · Who holds what

Stakeholders

hold the agency

Residents, businesses and visitors: the people who live with the outcome. They bring lived expertise, what they value and what a place is actually for.

02 · Who holds what

Institutions

hold the mandate

City offices, planners and funders: the bodies that must decide and deliver. They hold the mandate and set the guardrails every option has to respect.

03 · Who holds what

Agentic AI

never holds the pen alone

AI turns people’s own words into scenarios of the real place, guided by the rules the institution sets, and can evaluate each for its likely environmental and social impact. Scenarios, not forecasts: it shows what could happen; people decide what should.

04 · Where two meet

Public interest & governance
Seeing before choosing
Checking before committing

Point at an overlap, or tap it, to read what happens there.

05 · Where all three meet

Coplanning

One process rather than three: the institution holds the mandate and the guardrails, people bring what they know and value, and agentic AI turns both into scenarios everyone can see, weigh and agree on.

Diagram: the coplanning Venn, as on coplanai.com’s home page. The definition in the lede: Damiano Cerrone, doctoral manuscript (in progress), ch. 5 §5.4.

04 The ideas · three ideas of democracy

Count, consent, or contest?

Which idea of democracy you hold decides what AI does to it.

01 · Count

Majority wins.

The survey, the pin on the map, the vote. The decision most chosen is the legitimate one.

02 · Consent

Everyone agrees.

Habermas: we talk until everyone agrees.

03 · Contest

Every view stays at the table.

Chantal Mouffe argues that consensus is not democracy. The work is to see where people split, and where the bridges are.

04 · Coplanning

This is where coplanning sits.

Democracy is not only about choice, but about being able to change your choice.
Damiano Cerrone

Sources: Mouffe 2000 · Tessler et al. 2024 · Cerrone, doctoral manuscript (in progress), ch. 1, 5, 7, 9. The two lines on democracy are Damiano Cerrone’s own.

05 How we work · the framework

Two questions, four tasks.

Two questions place any engagement

Can it still change what will be done, or only how? Do people bring what we lack, or respond to what we give them?

02 · Open brief · respond

Scenario planning

What if this happened here?

Already decided: the question, not the answer.

03 · Open brief · propose

Speculative Futures

What could this place become?

Already decided: nothing yet.

04 · Defined brief · respond

Design review

What would make this better?

Already decided: the design.

05 · Defined brief · propose

Co-design

How should it look and work?

Already decided: what we will build, not how it looks.

Where each task has been used

The engagement framework, as on coplanai.com’s home page and in the use cases. Point at a task, or tap it, for what it is.

06 The participant app

A phone, a street, a question.

What a resident holds: their own idea coming back as a picture, with no account at any point.

Join with a QR

A poster QR or a short code opens the session. No sign-up, no email: a first name only, never a login.

Say it in your own words

One real street, one question. Type the idea, pick a theme, or draw on the photo.

See it as a picture, and change it

The idea comes back on the real street in seconds. Keep it, or change it.

Know the impact before you choose

Each picture can come with an evaluation of its likely social and environmental impact: who gains, who loses. The evaluation informs the choice. People make it.

Vote, and see where your ideas stand

At a table, each person picks up to three pictures they would defend, hidden until everyone has picked. Then you vote yes or no on other people’s ideas. Your words and pictures go into the engagement’s record.

To try the app, write to info@coplanai.com. Screen 01 is the app as it runs; the others are drawn from its steps, with a street and its impact evaluation taken from the platform’s own screens on this site. CoPlanAI.

07 Building the mandate

How do you build the mandate from scratch?

Sometimes the mandate exists in pieces.

Sometimes there is nothing to assemble. Then the mandate has to be built.

Four institutions · 2025–2026

Where we are building it

What each one lacked, and what we built with it. Point at a column, or tap it, for its dates.

What it takes

Governance first

Staff who can run it, a process redesigned around it, and support until the team leads.

Then the technology, alongside

Infrastructure, models trained on local rules, new features.

Not a tool delivered: a capability handed over

The mandate sits where all six tracks meet.

The rule we work by

Build the mandate before the pilot.

Sources: CoPlanAI project records; Connective Cities (opens in a new tab) (Oct 2026); TICO press release (opens in a new tab) (14 May 2026); City of Wiesbaden (opens in a new tab). The six-track figure is generic: no names, dates, budgets or places.

08 The ideas · AI and democracy

Replacing collaborative planning with synthetic deliberation programs

Can a synthetic public stand in for a real one? Three 2026 papers say no, each for a different reason.

01 · The public

Models made divided groups look uniform.

843 people and eight AI models answered the same US housing survey. All eight flattened the disagreement inside every group, three of them to nothing.

02 · The deliberation

The conversation changed nothing, the model is what matters.

Six AI citizens debated for three rounds. Sealed in separate rooms, unable to hear each other, they ended within one vote of the debate.

03 · The concession

Only people change their minds for each other.

Real assemblies started far apart and moved closer. AI groups started a third as far apart, and drifted further.

The ideas · the path between

The pass between them is narrow and the wind is strong.

When you bring AI into democratic planning, there is a cliff on each side.

01 · Two cliffs

Synthetic citizens, or a quiet handover.

Replacement: AI stands in for citizens. Abdication: we don’t build it at all, and the tools arrive anyway, sold by a vendor.

02 · The path

AI proposes. Humans authorize. No hand hidden.

A narrow path, braided from people, institutions and AI.

Sources: Cai et al. 2026, arXiv 2607.27100 (opens in a new tab) · Jang et al. 2026, arXiv 2609.07573 (opens in a new tab) · Flechtner 2026, arXiv 2608.10186 (opens in a new tab), charts redrawn from their tables · The path: Peng, Lu, Liu, Hou & Zhang 2026, JAPA (“AI proposes, humans authorize”) and Cerrone, doctoral manuscript (in progress), ch. 7 · The cliffs: Cerrone, ch. 9.

09 The research

The research behind the method.

We study our own engagements against the literature on planning, participation and generative AI. The research asks who gains power when AI enters planning, what has to stay with people, and what the AI cannot see.

Operationalising academic literature.

Four shelves, one Venn.

Collective intelligence, the commons, the human in the loop, and imagination at the centre.

02Stakeholders

Not one big brain. Many small rooms, linked.

“…a central intelligence simply can’t know enough… Widely distributed networks offer an alternative.”

Geoff Mulgan

03Institutions

Written with people, or filed and forgotten.

Rules people help write are kept and lived. Rules written for them end up in a drawer.

“…must recognize … their right to help make the rules.”

Ostrom’s principle, via Guy Standing

04Agentic AI

AI mediates the change.

Scenarios, impact evaluations and the institution’s rules made visible give participants and the institution an informed discussion, and a shared decision.

“Agentic AI provides support to mediate urban change.”

Damiano Cerrone

05Coplanning

An image none of us held alone.

Bisociation: frames that never meet, colliding.

“The logical pattern of the creative process… consists in the discovery of hidden similarities.”

Arthur Koestler · The Act of Creation · 1964

Sources: Mulgan, Big Mind (2017) · Ostrom’s principles, in Standing, Plunder of the Commons (2019) · Damiano Cerrone · Koestler, The Act of Creation (1964) · Jacobs, The Death and Life of Great American Cities (1961). Covers: the publishers’ cover art, shown small as a reading list.

205 dots10 groups241 referencesfrom 1959 to 2026

How to read itEach dot is an idea, a work, a person or one of our engagements: the oldest sit on the rim, the newest near the centre. Click a dot to read its note and references, drag to move the graph, and hold Ctrl or ⌘ while you scroll to zoom.Tap a dot to read its note and references, and use two fingers to move and zoom.

Every dot, in orderThe same graph as a list: all 205 dots, group by group, each with the same note.

The centre and three questions

4 dots

Collaborative AI planning, and the three questions the research keeps asking of it.

Power & participation

20 dots

This group asks who holds power when a public is invited to take part. Bachrach and Baratz (1962) and Lukes (1974) show power working through agendas and preferences; Gaventa (2006) adds who creates the space for participation. Arnstein (1969), Hart (1992) and Fung (2006) judge how much authority participation carries, and Cooke and Kothari (2001) examine from development practice how it can entrench power. Winner (1980), Kalluri (2020) and D'Ignazio and Klein (2020) carry the question into technology and data. The test for collaborative AI planning: do generative tools move choices towards residents, or concentrate framing with whoever configures them?

Participatory design

21 dots

This group follows the traditions that brought the people who live with a result into its design. De Carlo (1970) named the public beside the client, Habraken (1972) separated the collective support from each household's decisions, and Negroponte (1975), with Yona Friedman, imagined computers helping residents design their own homes. Scandinavian participatory design, from Ehn (1988) to Björgvinsson, Ehn and Hillgren (2010), added mock-ups, mutual learning and agonistic public spaces. Manzini (2015) separates diffuse design by everyone from expert design. Recent work asks how much room AI leaves for this. Participation counts when people shape the thing itself.

Communicative & collaborative planning

21 dots

This group follows planning theory from Lindblom's (1959) incrementalism and Rittel and Webber's (1973) wicked problems to the communicative turn, when Healey (1992; 1997) and Innes (1995) recast planning as argument among those who share a place. Power stays in view: Forester (1989) on attention, and the critics Flyvbjerg (1998), Fainstein (2010) and Hillier (2003) on rationality, justice and conflict. A parallel line runs from Lee's (1973) critique of large urban models through planning support systems (Klosterman 1997) to digital tools and AI (Goodspeed 2016; Peng et al. 2026). Collaborative AI planning inherits the method, its tools and their critique.

Deliberation & the public

22 dots

This group follows how publics reason together, from Habermas's public sphere to the deliberative polls, mini-publics and citizens' assemblies that governments convene. Its critics ask who is left out of the room (Fraser 1990; Young 2000), whether consensus hides conflict (Mouffe 2000), and whether a small sample may decide for everyone (Lafont 2020). Goodin and Dryzek (2006) trace how a forum's results reach decisions, and Fung and Wright (2001) tie deliberation to real authority. Recent studies test AI as mediator, drafting aid and stand-in respondent. Technology can widen and inform deliberation; the reasoning and the choosing stay with people.

Collective intelligence & the commons

17 dots

This group asks how people think and govern together, and how AI changes that. Ostrom (1990) showed communities governing shared resources through rules they make themselves; Hess and Ostrom (2007) and Foster and Iaione (2016) carried the argument to knowledge and to the city. Lévy (1997), Woolley et al. (2010), Landemore (2012) and Mulgan (2017) treat intelligence as a property of groups. Couldry and Mejias (2019) show everyday life captured as data; Burton et al. (2024) and Lovett (2026) ask what AI does to shared knowledge. For collaborative AI planning, residents' knowledge is a commons to govern with them.

Imagination & the generative image

18 dots

This group asks how images of places not yet built change what people imagine, say and choose. Appleyard et al. (1973) asked whether simulated views support valid judgement, and Sheppard (2001) proposed ethics for such images. Al-Kodmany (1999) matched sketches, maps and edited photographs to phases of participation. Koestler (1964) treats creation as joining frames that rarely meet; Jansson and Smith (1991) and Doshi and Hauser (2024) show examples and AI ideas narrowing what people produce. Field studies, including Ruoppila and Anastasiou (2026), ask when generated images open a conversation, when they close it, and how to label them.

Futures & utopia

18 dots

This group holds methods that let people picture a future before choosing one. Scenario planning, from Wack (1985) to Goodspeed (2020), explores several futures instead of one forecast; Myers and Kitsuse (2000) survey how planning represents futures to reach agreement. Future workshops (Jungk and Müllert 1987) and speculative design (Dunne and Raby 2013) hand imagining to residents and designers, and Appadurai (2004) finds the capacity to aspire unevenly shared. Levitas (2013) and Davoudi et al. (2018) show an imagined future can open a debate or close it. Generated scenarios should be plural, made with the people affected and held open.

Place, perception & situated knowledge

21 dots

This group gathers work on how places are seen: by residents, by institutions and by image models. Lynch (1960), Jacobs (1961) and Whyte (1980) taught planning to learn from residents' images of a city and what they do in it. Scott (1998) shows states seeing through simplified schemes that miss local knowledge; Corburn (2005) shows residents' knowledge joining professional methods. Berger (1972) and Haraway (1988) add that every view comes from a position. Image models carry the positions of their training data. The group asks of every generated scenario: whose view does it carry, and what does it leave out?

Our engagements

19 dots

Public cases where the ideas in this graph meet practice, from a street charrette in Jyväskylä in 2021 to Bologna, Vienna and Dubai in 2026. They involve city departments, universities, United Nations and German development agencies, a museum and civic labs. Each note draws on the live case page or a published report, and on an independent study where one exists. Ruoppila and Anastasiou (2026) studied processes in Helsinki, Pristina, Lahti and Dubai: images of real places helped people say what they wanted, and uptake depended on an institution able to act on the result.

Institutions & the human role

24 dots

This group asks what institutions must hold when AI enters public decisions, and what stays with people. Weizenbaum (1976) separates calculation from judgement; Lipsky (1980) and Bovens and Zouridis (2002) trace discretion from front-line officials into information systems. Bainbridge (1983), Elish (2019), Green (2022) and Schroeder, Roy and Kabbara (2025) show a person placed beside automation losing skill, absorbing blame or following the model. Jasanoff (2003) asks experts to admit uncertainty and bring citizens in. Lowndes, Pratchett and Stoker (2006), the OECD (2026) and Ruoppila and Anastasiou (2026) tie participation's effect to an institution able to act.

Sampo Ruoppila

Co-founder · research and policy lead

PhD and docent in urban studies. Professor of Practice, University of Vaasa; senior researcher, University of Helsinki.

Damiano Cerrone

Co-founder · product design and innovation lead

Doctoral dissertation on collaborative planning with generative AI, 2026.