But this triggered a different thought for me. User alignment might be just as dangerous. We’ve been here before in design: user-centred design optimises for one person’s goals, which can lead us to dark places. We moved to human-centred, but even that feels too narrow now. What about the impact we have on nature, on our environment? What we really need is more-than-human alignment—or maybe ecosystem alignment.

This thought became more concrete last weekend at the Into Nature festival in Drenthe. Among the artworks, one piece stayed with me: Poly-Nation by Manjot Kaur. The artist created a kind of mimicked negotiation between humans and nature—birds, bees, the landscape itself—captured in a new visual language. It was addressing the question: how do we give nature a voice in governance?

The work was placed in the Onlanden, a former polder now returned to water and wild seeds. But “returned to nature” is misleading here. In the Netherlands, almost all nature is designed, managed, hybrid. And this is where my thinking about physical AI comes in. The Hugging Face incident happened in digital space. Data, models, infrastructure. But as AI becomes embodied, as it enters our physical world through robots and buildings and things, the stakes change. A rogue AI in digital space is a data problem. A rogue AI in physical space is a presence in the world.

What if we started thinking of physical AI as a force of nature? Not something to fully control, but something to live alongside, to negotiate with. This connects to the work we’ve been doing with ThingsCon on “things that care”. AI-built things that optimise for community rather than individual gain, that operate with humility rather than extraction.

How do we design these new hybrid ecosystems? The Dutch approach to nature—planned but diverse, managed but alive—might offer a model. We already know how to create governance structures where different interests negotiate shared space. The artist at Into Nature was prototyping exactly this kind of language.

As AI moves into our physical environment, we’ll need similar negotiations. Not user alignment, not even human alignment, but ecosystem alignment—where AI, humans, and more-than-humans find ways to coexist. What would that governance look like? What language do we need to capture it?

I suspect they won’t come from redesigning individual products or adding more restrictions. They’ll come from redesigning the system itself—how we value things, what we optimize for, who gets a voice in the negotiation.

That absence of negotiation — of anyone being asked before systems arrive — is precisely what Peter van Waart has been confronting for over fifteen years. As a researcher and design educator at Rotterdam University of Applied Sciences, and one of the 31 experts interviewed for the State of Cities of Things research*, he brings a disarmingly concrete observation to the column’s call for ecosystem alignment: we already have democratic processes for physical things, but not for digital ones.

When a building goes up or an area is developed, there are established procedures — notification, consultation, the right to object. When a digital system appears in the city — sensors scanning license plates, algorithms managing mobility, platforms allocating services — it simply arrives. No one is asked. No one objects, because no one is informed. Van Waart finds this remarkable: these systems are not less intrusive than physical infrastructure. They shape behaviour and experience at scales both granular and systemic. Yet they bypass every deliberative mechanism we’ve developed for governing shared space.

The column asks what governance for ecosystem alignment might look like. Van Waart’s answer starts with an inversion. Most cities deploy technology first and hope for benefits. His “Smart and Social City” framework, developed within Rotterdam’s policy context, insists on the opposite sequence: define what kind of city residents want to inhabit, then determine whether and how technology serves that vision. Not alignment to the user, not alignment to the model — alignment to a collectively articulated future. That is closer to the ecosystem thinking the column is reaching for.

This inversion matters especially as AI moves from digital space into physical presence. The column observes that a rogue AI in digital space is a data problem, while a rogue AI in physical space is a presence in the world. Van Waart sees the same stakes. The Internet of Things as a concept has faded, but its material reality — sensors, data collection, algorithmic management woven through urban infrastructure — has only intensified. What changed is not the technology but who controls it: proprietary, closed-loop systems optimising within their own domains, inaccessible to public scrutiny. The Dutch landscape may be planned but diverse, managed but alive. Its digital layer, so far, is neither.

The 2026 Manifesto for Immersive AI frames this as Democratic Precedence: governance infrastructure must precede technological infrastructure. Communities define the kind of environment they want before agentic systems are introduced. Van Waart’s work shows what that principle demands in practice — not referendums on every sensor, but legitimate processes through which the terms of engagement are surfaced, debated, and decided collectively. The negotiation the column imagines between AI, humans, and more-than-humans needs exactly this: a democratic floor beneath the ecosystem.

This weekly reflection connects the column to thinking from Peter van Waart, researcher and design educator at Rotterdam University of Applied Sciences, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas he shared in the original interview, not a new consultation. Claude made the first interpretation.

Taken from weeknotes 404 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week's and coming week's interesting events and a paper for the week.

What caught my attention wasn’t the camera quality. It was how mundane this robot is. And how that might be exactly the point. Beni has no conversational AI. It doesn’t think. It doesn’t advise. It follows commands, moves in ways that feel alive, and gradually becomes something you care about. Not because it’s intelligent, but because it’s present, responsive, and a little bit cute. You don’t need an LLM to create attachment. You need behavior.

This is relational design without the conversation. Think about how TikTok or Spotify learn your preferences. Not by asking, but by watching how you respond. Beni could do the same. It doesn’t need to interview you about your filming preferences. It can tease interactions, observe your reactions, and quietly improve its understanding of what you want. Slower than a setup wizard, but far more natural. More like a pet learning your routines than an app onboarding you.

Here’s where I start wondering about Trojan horses. Beni itself might be perfectly fine. It isn’t free, it isn’t primarily a data play, and the company might have all the right intentions. But that’s beside the point. What Beni represents is a door opening. A near future where we welcome a flock of these mundane, charming, single-purpose helpers into our personal environments. The question isn’t whether this device is trustworthy. The question is whether we’re developing the awareness to evaluate the entire category. And demand the same principles from all of them. Devices like Beni pave the way for accepting them in our personal lives, without critically questioning how they work.

The honest version of this product runs entirely on Edge AI. Everything it learns stays local to your Beni. No footage uploaded, no behavioural data shared with a cloud. But here’s the interesting tension: Beni becomes more valuable if it can learn from the experiences of all Beni’s. How people move, what makes a good action shot, when to stay back. That collective learning is possible without sharing personal data. Behaviour patterns can be aggregated without attaching them to you. Whether companies will actually build it that way is a different question. These are the kind of design considerations I’ve been working through in the Cities of Things Manifesto with 10 principles for physical and embodied AI that should apply whether the device is charming or not.

Right now, Beni follows. It does what you ask. But what happens when it gets good enough to start directing? “Stand over there, the light is better.” “Do that jump again, I missed the angle.” What happens when Beni connects with other devices in your home? Your lights, your speaker, your other robots to orchestrate a moment you didn’t ask for? Or not even be connected but instruct you to change the settings of the house (indeed, the reversed centaur is a returning theme). This isn’t science fiction. It’s the natural trajectory of any device that learns your patterns and can act. The shift from tool to collaborator to director happens gradually, and we often don’t notice until we’re already comfortable.

I might order one and see how it behaves. And I’ll be paying attention. The future of physical AI might not arrive as a humanoid assistant that talks to you. It might arrive as a flock of single-purpose helpers, each one mundane, each one charming, each one learning. That’s exactly why we need to think about the category now, before the flock settles in.

That blurring — between robot and AI, between tool and companion, between something you use and something you relate to — is exactly where Iohanna Nicenboim’s research lives. Iohanna was interviewed as one of the 31 experts interviewed for the State of Cities of Things research*. As assistant professor of more-than-human design and regenerative AI at Linz’s Interdisciplinary Transformation University, she has spent years on the question Beni makes tangible: what happens when the categories become porous?

Her answer is that the significance lies not in settling whether something is “a robot” or “AI” but in what emerges when that distinction dissolves. Beni has no language model, no conversational intelligence. Yet it moves in ways that feel alive, and you start caring about it. Nicenboim would recognise this as precisely the productive blur she studies. A thing whose agency isn’t intellectual but relational, built through behaviour and proximity rather than conversation.

What makes her thinking particularly sharp for this column is a deeper move. By engaging with nonhumans, she argues, we simultaneously critique how we perceive ourselves. Beni creates attachment without intelligence, and that doesn’t just tell us something about the robot. It tells us something about our own relational wiring — how little it takes for presence and responsiveness to trigger care. That insight matters when we think about the flock. If attachment is this easy to produce, the design ethics around it become urgent.

Nicenboim traces a shift from questions of agency (”can things act?”) toward questions of representation and citizenship. The column follows the same arc: from Beni-follows to Beni-directs, from tool to collaborator to something that might start centering its own logic over yours. Her question — “who is centered in human-centered design?” — lands hard when the device starts saying “stand over there, the light is better.” At that moment, the design has quietly recentered.

She also brings a concept the column’s edge-AI tension needs: regenerative AI, drawn from principles of locality, reciprocity, and seasonal attunement. The honest version of collective learning — aggregating behavioural patterns without attaching them to individuals — is a regenerative logic: the system gives back to the network without extracting from the person. Whether companies will build it that way is the open question, but Nicenboim insists on holding that productive uncertainty rather than collapsing into either techno-optimism or rejection.

The 2026 Manifesto for Immersive AI frames this as Relational Integrity: the principle that the quality of a human-thing relation should emerge from the interaction itself, not be predetermined by external optimisation, and that it should protect, not erode, human-to-human relations. Beni earns its relationship through behaviour. The question is whether the flock that follows will do the same, or whether relational integrity becomes the first casualty of scale.

This weekly reflection connects the column to thinking from Iohanna Nicenboim, assistant professor at the Interdisciplinary Transformation University in Linz, drawing on her contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is an interpretation of ideas she shared in the original interview, not a new consultation. Claude made the first version of the comparison at my request.

An AI uses our formulations, mirrors our phrasing, recalls our preferences, answers in a recognisably human rhythm. It addresses you directly, and there’s a sense of understanding, of realness. But behind it, no one is home. No shared situation, body, history, or stake in the conversation. Presence without presence, the eerie in its purest form.

It deepens when you remember that AI is trained on us. It inherits not just our knowledge but our social patterns, our biases, our toxic habits, our collective quirks. And in a culture increasingly oriented toward instant, individualised gratification, AI doesn’t just reflect us. It reframes what we expect from connection, from thought, from each other. It becomes a warped mirror that also amplifies our displacement from collective experience.

Cory Doctorow’s idea of the reverse centaur keeps returning here: not a human steering a machine, but a person adapting themselves to the machine’s logic. Offloading thinking, memory, even social cues, because it’s convenient. In an agentic future, the question may not be whether an AI can act for us, but how quietly we start acting for it: structuring our lives, requests and choices to fit its assumptions. The cost is a slow erosion of the capacities that make us distinctly human. Context awareness, social intuition, even the friction of consciousness itself.

This becomes more tangible as AI moves into devices. A small pod for catching thoughts, a ring, a companion in glasses or earbuds. These promise something a phone cannot. But what exactly is their role? An extension of our own intentions, helping us retain and reflect on a thought? Or a conversational surface for a company with its own incentives: engagement, purchases, persuasion? The design challenge here isn’t technical; it’s relational: these systems need to feel like extensions of you, not conversations with an entity that might have its own motives, or its owner’s.

The same question sits inside the idea of AI as ghostwriter; an inverted ghostwriter, taking fragments or premises and spinning them into a story, even a personalised world to explore. Intoxicating, and genuinely interesting as an affordance. But stories shape how we see the world. At which point does a helpful simulation start to replace the more difficult reality it was meant to illuminate? What dark patterns might emerge, even unintentionally, when a machine starts telling us our own stories back — and how do we notice the patterns designed into it?

The current backlash against AI “slop” might be read as a clumsy but understandable response to exactly this. It isn’t only about bad images or generic text, and not only about quality. It’s about trust, about orientation: to know what was made, by whom, with what intent, what’s grounded and what’s simulated. Watermarks alone won’t solve that, but clearer signals of provenance may become part of keeping our bearings in a mixed reality.

Nate Jones’ plea for responsible agent ownership adds an important inversion. If we delegate action to agents, the question isn’t only what an agent is allowed to do. It’s what its owner remains responsible for noticing, prompting and preventing. It shifts the question from the machine’s behaviour to our own agency; the prompt becomes the new frontier of ethics.

So here we are, in the uncanny valley between tool and companion, author and audience, real and simulated. The eeriness won’t fade. But perhaps the question isn’t how to eliminate it, but how to live with it. And ensure that in this blurred landscape, agency remains ours to keep.

That eeriness is precisely what Vera van der Burg has been working with for the past five years. Vera was one of the 31 experts interviewed for the State of Cities of Things research. As a designer, researcher and Ph.D. candidate at TU Delft, she has developed a practice that offers an unexpected response to the column’s central unease: not to eliminate the eerie, but to sit with it, slowly, and see what it reveals.

Van der Burg works with ceramics and AI in a deliberate loop. She photographs her own sculptures, annotates them with personal conceptual labels, trains a model on this intimate dataset, prompts it with terms like “ceramic jealousy,” then realises the algorithmic proposals by hand in clay. Two full cycles in three months. The forms that emerge are, in her words, “always a bit strange, always a bit wrong.” That wrongness is the material. The eerie made tangible, something present that shouldn’t quite be there, something recognisable yet off.

Where the column describes AI as a warped mirror, Van der Burg makes that mirror productive. By annotating her own work, she must ask what she actually means by “rough” or “uncanny”. Making her own perceptual frameworks explicit. The technology becomes a mirror not into the system, but into the self. The eeriness doesn’t disappear; it becomes a working condition, the way a ceramicist lives with the kiln as black box. You prepare, you predict, you close the door, you wait. Sometimes it cracks. Sometimes the glaze runs differently. You think back, try again, or accept the surprise.

This is where her practice speaks directly to the design consideration “Temporal Honesty”. The column asks when helpful simulation starts replacing difficult reality. Van der Burg’s answer is methodological: by insisting on slowness, on hand-making, on loops that take months rather than seconds, she refuses the temporal collapse that makes AI eerie in the first place. A system that behaves as if it knows you when you have never met it creates exactly the uncanniness Fisher describes. Van der Burg’s practice does the opposite: it earns its depth through sustained interaction, encounter by encounter.

The 2026 Manifesto for Immersive AI frames this as Incorporate Temporal Honesty: the principle that agentic environments should not simulate relational depth that has not been earned, should respect the difference between a new encounter and an established relation, and should acknowledge the incompatible temporalities of their own layers. Van der Burg’s ceramics embody this. Clay operates on the timescale of the kiln, algorithms update in seconds, and the maker moves between both, slowly, refusing to let either collapse the other.

This weekly reflection connects the column to thinking from Vera van der Burg, designer and Ph.D. candidate at TU Delft, drawing on her contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas she shared in the original interview, not a new consultation. Claude Opus 4.6 made the first version of this reflection on our suggestion.

The triggered thought is taken from weeknotes 402 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week’s and the coming week’s interesting events, plus a paper for the week.

Are you interested in diving deeper into a question you or your organisation has? I am available for shorter or longer reflections, deep research, inspiration sessions, guest lectures, or workshops. Or organising this as an event.

The increasing use of “teams of agents” by independent workers raises questions about whether these individuals still qualify as independent workers or if they transition into SMEs. This raises legal and structural considerations about what defines an SME, particularly the number of people involved versus distributed work. Cory Doctorow’s observation in Reverse Centaurs triggers: “Copyright inheres at the moment of fixation of a new work of human creativity.” It’s a legal principle, but it triggered a thought: what happens when the fixation isn’t just human? When the work emerges from a collaboration between one person and a constellation of AI agents, who — or what — owns the creative act? And if a single human can now orchestrate what once required a team, are they still an independent worker, or have they quietly become a new kind of SME?

When these communities evolve into more legal entities that stand on their own, it might invite other functions than working together. Shaping a shift in our digital autonomy and sovereignty, an often-discussed topic at the moment. Like for the management of our data. In the current discussion on how to value data centres, we might rethink these too. The EU’s own communications on digital sovereignty and data spaces underscore the tension: we want control over our data, but the infrastructure to achieve it comes with its own contradictions. Yet sovereignty doesn’t have to mean centralised fortresses. What if it looked more like a network? Every building, every block, a node in a distributed data infrastructure. The peer-to-peer dream never died; it just waited for the right moment. If we’re serious about civic economies, shouldn’t our digital backbone reflect that too? Not just technically decentralised, but socially embedded, governed by the communities it serves.

Perhaps the imagined SME of the near future might indeed be one human and a constellation of agents. Not a team, but a new kind of entity. And if the legal system will catch up, it opens up other forms and uses for sovereignty. The cultural shift is already here. We’re not just building tools. We’re designing new forms of coexistence. And the most interesting question isn’t whether we can, but what kind of world we create when we do. One where the line between worker and tool, between human and system, between individual and community, is not just blurred but redrawn entirely.

Simone Rebaudengo is already designing these constellations. As director of oio.studio and one of the 31 experts I spoke with for the State of Cities of Things research, he has spent the past years moving from single-object intelligence toward what he calls multi-agent constellations embedded in physical things. “An object no longer contains one algorithm,” he observes. “It contains a potential ecosystem of agents operating at different purposes, scales, and timescales.” A product isn’t smart, it’s inhabited. The design question shifts from “how do we make this smarter?” to “what is the world inside this object?”

That reframing matters for the column’s argument. The one-human-plus-constellation scenario isn’t speculative, it’s already design practice. Where a hardware project once required a team and months of development, Rebaudengo notes that a single designer can now build functional interactive objects in weeks using off-the-shelf components and locally-run language models. The tools for becoming an SME of one already exist. What hasn’t caught up is the vocabulary — legal, cultural, organisational — for understanding what that entity is.

Rebaudengo also names the design challenge the column leaves open: topology. Not the shape of the device but the topology of its intelligence. The interfaces, knobs, and mechanisms by which humans maintain agency when intelligence is distributed across a constellation. When team culture is coded into agents, someone needs to design how that culture is visible, negotiable, and governable. Without that, you have a black box that happens to collaborate. With it, you have something closer to the transparent assemblage the column is imagining.

And he pushes toward exactly the kind of sovereignty the column envisions. Not centralised but grassroots. A designer could reimagine what a public monument is: not static sculpture but a locally owned, locally intelligent object serving its neighbourhood. Communities could build agent constellations without dependency on corporate infrastructure or ubiquitous connectivity. The peer-to-peer dream, made physical.

The 2026 Manifesto for Immersive AI frames two principles here. Reveil Assemblage Transparency: when multiple agents act together, their configuration and purpose should be legible to the people affected. And Ensure Distributed Sovereignty: control should reside with communities, not concentrate in platforms. Rebaudengo’s practice suggests these aren’t just governance ideals, they’re design briefs. The constellation needs a topology. The topology needs to be visible. And the whole thing needs to be owned by the people it serves.

This weekly reflection connects the column to thinking from Simone Rebaudengo, creative director of oio.studio, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas he shared in the original interview, not a new consultation. Claude Opus 4.6 crafted the first version of this reflection.

The column is taken from weeknotes 401 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week's and coming week's interesting events and a paper for the week.

A year ago I was presenting about immediacy culture in relation to agentic AI. Immediacy as the marketing logic where satisfaction has to be instant, attention spans keep shortening, and platforms are engineered for a steady drip of small dopamine hits. Hyperpolitics looks like that same logic, applied to civic life. Quick outrage, quick cause, quick next thing. And it’s worth asking what happens when the systems serving us that drip stop being mere algorithms and start being something closer to agents. Reading us, adapting to us, nudging in real time. The tools for hooking attention are getting a mind of their own.

Then there’s the layer of physical AI. If our computational environment is being tuned for hyper-engagement, and that same intelligence starts showing up in the objects and spaces around us then the stimulus doesn’t stay on a screen. It becomes ambient. A hyperpolitical culture inside an immersive one seems like a combination worth being a little uneasy about.

But there was a counter-signal this week too, and it’s the more hopeful thread. For fifteen years the internet’s business model has been attention. Every feed, every feature, built to hold eyes. That model assumes the eyes are human. But more and more they’re not: AI agents are now the ones reading and summarising content on people’s behalf. Cloudflare’s new AI Crawl Control, which lets publishers charge bots for access, is one early attempt to deal with the fallout. The buying power still sits with people, not their bots, so the old attention economics start to wobble. My tentative hope is that this pushes value back toward the content itself. Its accuracy, its craft, its human fingerprint, rather than its sheer stickiness. It is already indicated that the next wave of “influence” belongs to people with something real to show, not just a knack for holding attention.

That hope connects to the theme we’re shaping for this year’s ThingsCon: the idea of Things that Care. Could the systems around us — the models, the interfaces, eventually the physical objects — be built with something like care for human values at their core, the way “don’t be evil” once tried to function for an earlier internet? I’m honestly unsure whether that becomes real design practice or stays an unmet aspiration. Left unaddressed, we’re heading toward disillusionment, as AI-generated content increasingly feeds on itself.

The current polarised political leaders are extraordinarily good at surfing waves they didn’t create. Swap the surfer and the wave keeps moving. Which is either discouraging (no single lever to pull) or oddly hopeful: it means the work is systemic; building civic, relationship-based value models, small coalitions, can all make a difference.

That systemic work — building the durable structures that hyperpolitics is hollowing out — is exactly what Emiel Rijshouwer practises. As a researcher and designer at Design & Publics in Rotterdam, and one of the 31 experts I spoke with for the State of Cities of Things research, he has spent years on the question Jäger’s diagnosis leaves open: if institutions are thinning, what replaces them?

His answer is publics. Not the public as abstract mass, which hyperpolitics mobilises and abandons, but specific groups of stakeholders organised around a shared concern. Residents, teachers, social workers, local organisations, convened to think together about a system that affects them all. The distinction matters. Hyperpolitics runs on quick engagement without follow-through. Rijshouwer’s practice insists on the opposite: genuine reciprocity. When his team asks people to participate, the relationship cannot be one-directional extraction. “Everyone comes to look at how badly things are going with us, and then they leave and say they’ve consulted the public,” he observes of the consultation fatigue that has overtaken Rotterdam’s neighbourhoods. Thirty thousand new housing units are being planned in Kralingen. Residents weigh in on overlapping consultations from different departments. No learning flows between efforts. No shared ownership of outcomes. That is hyperpolitics in civic drag; the form of participation without the substance.

The immersive AI layer makes this harder. Rijshouwer notices that the politically significant infrastructure in cities is increasingly invisible. A scooter on your sidewalk is undeniable, you see it, you can object. But a digital platform allocating resources, an algorithm deciding which streets get maintenance priority, an AI managing the energy grid. These operate at scales and speeds no resident grasps. “How do I give my corner shape,” he asks, “if something moves around it that I can barely see?” If hyperpolitics is loud engagement with visible targets, the real governance challenge is the quiet infrastructure shaping outcomes nobody is mobilising around.

His response is not to reject autonomous systems but to insist they could be governed as commons. He’d rather see a delivery vehicle labelled “City of Rotterdam” than a corporate brand. Not out of municipal romanticism, but because collective stewardship changes what the system is for. “If we all had some say in how these things behave, if we learned through them about mobility, congestion, air quality, they might serve a shared purpose rather than a private extraction logic.”

The 2026 Manifesto for Immersive AI frames this as Democratic Precedence: the principle that publics should have meaningful voice in how autonomous systems are designed and deployed before they arrive, not after. Rijshouwer’s practice shows what that looks like on the ground. Slow, relational, reciprocal, the opposite of hyperpolitics in every way. The wave keeps moving whether or not the surfer changes. The work is in building the structures underneath: durable publics that don’t dissolve when the attention moves on.

The triggered thought is taken from weeknotes 400 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week's and coming week's interesting events and a paper for the week.

In this weekly reflection connects the column to thinking from Emiel Rijshouwer, researcher and designer at Design & Publics in Rotterdam, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas he shared in the original interview, not a new consultation. Claude Opus 4.6 supported connecting the two.

This is not about technology fear. It’s something else. A fundamental uncertainty about what’s real. What can be trusted. What version of yourself is the true one. This might be what Arno Coenen brings to life in his short AI movie “Innerverse.” The immersive film was recently selected for the competition for the NFF Golden Calf. VR not as an escape, but as a mirror of your own inner world. This is a movie with a fixed narrative, but what if it responded to your personal inner world while you watched? It could become as dark as the popular movie Backrooms. But maybe it can also be more positive. Learning something about you. Connecting you to others. Overcoming the anxiousness.

It opens up new unbeaten paths. A podcast pointed me to an interview from last year with philosopher Bernardo Kastrup. His concepts of consciousness in a world of states instead of things can open up even more layers. It is too much to dive into. Kastrup argues that AI cannot be conscious. Not because it’s not smart enough. But because consciousness isn’t something you build. In his view, everything already is consciousness. AI simulates information flow in silicon. It processes. It responds. But it doesn’t experience.

He used an analogy. Water is something. But what is a wave? Is that a thing? Or is it just a state, a pattern, a temporary form? If you break it down, it disappears. It was never a separate entity. Maybe AI is like that. Impressive patterns. Complex responses. But not something in itself.

I’m not per se subscribing to this thinking; I should dive deeper first. But it offers a useful frame. Especially when combined with the anxious generation. We are delegating more and more to AI. Our memory. Our decisions. Our creative output. Our sense of what’s plausible. If we’re moving parts of ourselves into systems that are, in some philosophical sense, nothing, what does that mean?

Imagine that application, a personal Innerverse. It doesn’t feed you content. It doesn’t suggest what to watch or buy. Instead, it tracks how you’re engaging with the world. What patterns emerge in your attention. What you avoid. What you return to. Not to manipulate you. To show you. A mirror, not a window.

For an anxious generation uncertain about reality, this might be more valuable than another layer of generated content. A tool that helps you see yourself seeing. That makes the invisible visible.

How do we find footing when reality, self, and intelligence are all blurring into each other? When the tools we use are both nothing and everywhere? When people are afraid to dance because the world remembers everything? The next interface might not show us new worlds. It might show us how we relate to this one.

Taken from weeknotes 399 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week’s and coming week’s interesting events and a paper for the week.

Maria Luce Lupetti, assistant professor at Politecnico di Torino and one of the 31 experts I spoke with for the State of Cities of Things research, has a way of naming what this column is circling. Her concern is the cost of delegation. When people outsource cognitive tasks to AI — writing, articulation, critical thinking — they lose not just the labour but the learning embedded in struggle. The thinking that writing forces you to do. The self-knowledge that comes from trying to articulate something you don’t yet fully understand. “Accepting mediocre and homogenous output becomes the path of least resistance,” she observed, “especially when lives are already fractured by competing demands.”

That frames the anxious generation’s predicament precisely. They’ve already lost dancing, an expressive act that reveals you to yourself through the body. Now they’re delegating memory, decisions, creative output to systems that are, in Kastrup’s framing, impressive patterns without experience. Each delegation removes a mirror. Not because AI is malicious, but because the struggle itself was the mechanism for self-understanding. You don’t learn who you are by receiving answers. You learn it by wrestling with questions.

This is where the Innerverse concept becomes interesting as a design proposition rather than just a film. Lupetti’s argument suggests that most AI removes self-knowledge by doing the work for you. The Innerverse would reverse that: using AI not to produce on your behalf but to show you your own patterns. What you avoid, what you return to, where your attention goes when you’re not directing it. The mirror doesn’t need to be conscious to be useful. A wave is not a thing, but it can still show you your own reflection.

Lupetti is also deliberately skeptical of language that grants AI partnership or agency. To call a system a partner, she argues, obscures the infrastructure of capital and labour behind it. That skepticism seems to align with Kastrup’s wave metaphor and protects the Innerverse from its own trap: the moment it starts feeling like a companion rather than a mirror, it has become another thing you delegate to rather than a surface you see yourself in.

The 2026 Manifesto for Immersive AI calls this Predictive Symmetry: when systems predict human behaviour, individuals should have reciprocal insight into how those predictions are formed. Most AI predicts you opaquely — the system maps your patterns, you never see the map. The Innerverse is predictive symmetry made into interface: the system shows you what it sees. Not to control. To return something the delegation has been quietly taking away.

This weekly reflection connects the column to thinking from Maria Luce Lupetti, assistant professor at Politecnico di Torino, drawing on her contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas she shared in the original interview, not a new consultation column. Claude Opus 4.6 supported in connecting the two.

Are you interested in diving deeper and more specifically into a question you or your organisation has? I am available for shorter or longer reflections, deep research, inspiration sessions, guest lectures, or workshops. Or organising this as an event.

Alexa was built around commands. You speak, it obeys. The relationship is clear: AI as tool, you as operator. But the signals from OpenAI and the design choices in devices like the Codex keyboard suggest something different is coming. Not a device that waits for instructions, but one that listens, learns, and infers. An ambient intelligence that attaches not to you as a wearable exoskeleton, but to your home itself.

A Home Mic for intelligent whispering. Cheap enough to place in every room, like an AirTag for context. It doesn’t need good speakers; your existing devices handle output. It connects to your thermostat, your lights, your appliances. It builds a model of you through what it hears. The home becomes an intelligent environment you inhabit rather than a collection of tools you operate.

Is that what Codex Micro keyboard is about? Testing the interaction and interface philosophy. You can adjust the level of intelligence. Like a DJ mixing a track, you decide how much AI to bring into the performance. More intelligence for complex tasks, less when you want direct control.

This hints at what the home interface might become: not a binary on/off, but a dial. How much do you want your home to anticipate? To infer? To act without asking? The interface becomes a negotiation between your agency and the system’s capability.

This is the centaur concept. Human and AI as co-performers, each bringing something the other lacks. You provide intention, context, judgment. The home provides memory, pattern recognition, tireless attention. Together, a symbiosis. The home intelligence becomes an extension of how you live. But centaurs learn in both directions. You’re not just training the system; it’s training you. You adapt to its rhythms. You learn to speak in ways it understands. You begin to anticipate what it will suggest. This is the reverse centaur: the human adjusting to serve the AI so the AI can serve them better.

Cory Doctorow warns that reverse centaurs can become extraction machines: humans doing unpaid labour to fix AI errors for corporate benefit. In the workplace, that looks like workers compensating for algorithmic failures. In the home, it might be subtler: you become QA for OpenAI’s ambient intelligence, correcting misreadings of your mood, training the system through your frustration.

And through this training, this ambient intelligence becomes something more permanent. A second skin around your home; an orchestrating layer as fundamental as the walls and wiring. Not a device you can unplug, but an intelligence woven into how the space functions.

What happens when this skin knows you so well it anticipates your needs before you feel them? At what point do you stop knowing which desires are yours? And which ones you’ve absorbed from the house that’s been quietly shaping you? And what will OpenAI, or whoever controls this layer, do with everything your home has learned about you?

Gerard Nijboer, Strategic Advisor for Digitalization at the City of Rotterdam, has a name for what’s at stake here. In our conversation for the State of Cities of Things research — 31 expert interviews about living with physical AI — he kept returning to one concept: discretionary space. The room for human interpretation, for giving your own meaning to your surroundings and to yourself. The room that tightening digital systems quietly erode.

His concern is precise. If you code everything too tightly, if every possible action is pre-programmed, every interaction anticipated, every preference inferred before you’ve consciously formed it and humans lose the space for their own meaning-making. They inherit a world of closed choices. And the sharpest edge of his warning: people born into such an environment won’t even know what was lost. You can’t miss discretionary space if you’ve never experienced it.

That feels to map directly onto the second skin. A home intelligence that listens continuously, builds a model of your habits, and begins anticipating your needs is precisely this kind of tightening. Not malicious. Not even intrusive, at first. Just incrementally closing the gap between what the system infers and what you decide. Until the gap disappears and with it, the space where you might have wanted something the system didn’t predict.

Nijboer references McLuhan who famously introduces the notion on how we shape our tools, and our tools gradually come to shape us. Nijboer adds the question of what happens to democratic agency when the shaping becomes ambient and invisible. When the second skin doesn’t announce itself as technology but simply feels like how the house works. It is the reverse centaur restated as institutional wisdom.

The 2026 Manifesto for Immersive AI calls this Peripheral Presence: the principle that AI should remain at the periphery of attention rather than colonising it. Nijboer’s discretionary space suggests what happens when that principle is violated; not through loud intrusion but through quiet completeness. A home that anticipates everything leaves nothing for you to initiate. The peripheral becomes total. And that may be the deeper risk of the intelligent second skin: not that it watches you, but that it fills the silence where your own thinking used to happen.

The weekly reflection that connects the column to thinking from Gerard Nijboer, Strategic Advisor for Digitalisation at the City of Rotterdam, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is my interpretation of ideas he shared in the original interview, not a new consultation. The first version of the reflection was made with the help of Claude (Opus 4.6 High).

The column is taken from weeknotes 398 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week’s and coming week’s interesting events and a paper for the week.

Are you interested in diving deeper and more specifically into a question you or your organisation has? I am available for shorter or longer reflections, deep research, inspiration sessions, guest lectures, or workshops. Or organising this as an event.

The new manifesto for Cities of Things and the sourcing research is a foundation. Let me know, and we’ll check what is possible. Or check Cities of Things, ThingsCon, Civic Protocol Economies to find some examples of my latest work.

What struck me wasn’t the warning. It was the absence of any other view. There’s a group of researchers quite critical of that framing, and Bert Hubert got barely a moment to say so. It felt like a reminder that we need another narrative too, not just a faster version of the one we have: catch up to what, exactly, and on whose terms?

I’m not sceptical of AI’s impact. I just wonder whether copying the US playbook — endless capital into data centres, the race for the fastest model — is the only serious answer, or just the easiest one to defend in a soundbite. I understand why we reach for it; it fits the system we already live in. But there might be other ways to build a position, ones that start from what we actually need AI to do rather than how fast it should do it.

The usual counter is that the rat race is depleting the planet. True, but too simple. The cost isn’t only environmental. It’s what it does to our relationships, our collective intelligence, our resilience when something unexpected happens. A faster model doesn’t make a community more resilient.

That’s the question I’ve been circling in my mind too. In my RIOT 2026 article I sketched The Grid: a building where residents’ devices reason locally and negotiate with the block’s energy network over a mesh, no data centre required. A fridge that knows its own household instead of phoning a server to ask. An irrigation system that develops a position through years of conversation with the systems next door, rather than arriving pre-loaded with someone else’s optimum. Not smaller AI as a compromise, but as the actual design target — intelligence that fits its context, instead of one that tries to know everything, everywhere.

Which makes this week’s flurry of frontier model launches an odd backdrop. Every lab racing to claim the smartest model, while it’s getting clearer that “smartest” isn’t the metric deciding who wins. Model choice is turning out to be political as much as technical — and we tend to welcome that shift toward the societal, until the politics behind it is autocratic rather than democratic.

Still, there’s a useful idea buried in the launch noise: the smartest model is not the fittest one. Fit depends on the goal and the context, not the benchmark.

Maybe that’s Europe’s actual opening — not to out-race anyone, but to get serious about what “fit” means here: which values, which relationships, which long-term bets we’re willing to make instead of the fast ones. I don’t know yet what that looks like at the scale of a continent. But I’d rather sit with that question than help lose a race we never needed to run in the first place.

For the State of Cities of Things research, I spoke with Kars Alfrink, researcher and designer at Delft University of Technology. His thinking sharpens the question this column is asking: if the race is wrong, what replaces it?

Alfrink has been developing what he calls “People’s Compute”; a framework for redesigning how we approach AI through three deliberate shifts. Design for groups rather than individuals, because what looks like personal vulnerability is often better understood through shared structural position. Design for infrastructure rather than applications, because if a city wants to govern AI, the conversation must reach down to data pipelines, cloud dependencies, and governance stacks. Not stay at the level of particular tools. And adopt realism rather than moralism, because the AI debate has become so polarised that participants assume consensus on what’s good or obvious about what’s bad, while the actual situation involves conflicting interests that need patient untangling.

That last shift — realism — feels directly connected to the Nieuwsuur moment. The catch-up narrative is moralistic in its own way: it presents the US race as the only serious frame and treats anything else as naivety or denial. Alfrink’s realism asks a different first question: what is actually happening in this specific situation, what is its history, who benefits? From that grounded position, the European sovereignty discussion looks different. As he observes, the push for European digital sovereignty often risks merely replacing American corporate dominance with European corporate dominance. A lateral move presented as liberation. Racing to build European data centres to compete with American ones doesn’t change the underlying logic. It just changes the flag on the building.

His work on digital autonomy in public infrastructure like smart charging points, municipal library systems, camera systems in public space, examines what it would take for these systems to be genuinely autonomous rather than dependent on distant corporate stacks. Not as ideology but as practical engineering and governance. What does a charging point look like when it doesn’t phone home to a platform? What does a neighbourhood AI look like when the infrastructure is locally owned and governed? These are modest questions, but they point toward exactly the kind of fitness the column is reaching for: intelligence shaped by its context rather than imported from elsewhere.

Alfrink draws on left climate realism. The recognition that many impacts are already locked in, yet insisting on creating sustainable niches for human life nonetheless. Neither fatalistic nor utopian. That framing fits the column’s closing instinct precisely. We may not prevent the concentration of AI power in a few global players. But we can design local intelligence that fits its context, serves its community, and doesn’t need to win a benchmark to be worth building. The 2026 Manifesto for Immersive AI calls this Distributed Sovereignty: the principle that control over AI should be distributed across communities rather than concentrated in platforms or states. Alfrink’s work suggests that sovereignty isn’t claimed through declarations. It’s built, one piece of infrastructure at a time.

This weekly reflection connects the column to thinking from Kars Alfrink, researcher and designer at Delft University of Technology, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas he shared in the original interview, not a new consultation. The report of the conversation was the source; Claude co-work shaped the first version of this reflection.

—

The triggered thought is taken from weeknotes 397 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week’s and coming week’s interesting events and a paper for the week.

Are you interested in diving deeper and more specifically into a question you or your organisation has? I am available for shorter or longer reflections, deep research, inspiration sessions, guest lectures, or workshops. Or organising this as an event.

The new manifesto for Cities of Things and the sourcing research is a foundation. Let me know, and we’ll check what is possible. Or check Cities of Things, ThingsCon, Civic Protocol Economies to find some examples of my latest work.

The harness is the scaffolding around the model: the interface, the integrations, the way it connects to your files, your calendar, your tasks. OpenAI’s Codex, Anthropic’s Co-Work, these are attempts to make the model sticky. Not through the model itself, but through the ecosystem wrapped around it.

This is a defensive move, though. The model makers know their core product is commoditizing, so they’re scrambling to become something else: the Adobe of intelligence, the WeChat of the West. Lock users into a suite. Make switching painful. The Chinese super app strategy, applied to AI.

But what is missing: the real shift isn’t about who builds the best harness. It’s about what sits above the harness entirely.

After the Fable and Sol outages last month due to government decisions, something became clear and out in the open: dependency on any single model is a liability. The smart systems now are model-agnostic. They orchestrate. They pick the right model for the right task, swap providers when one goes down, run certain repeating jobs on local open models at the edge. The model becomes the engine, not the product.

This orchestration layer, call it a meta-layer, call it a new kind of operating system, doesn’t care about frontier benchmarks. It cares about task completion. And the most mundane tasks, it turns out, are the most interesting test case.

Chores.

I watched a video this week again of a soft robot folding laundry. Not a fully humanoid (halfway), more a purpose-built machine, optimised for the single task of handling fabric. It looked nothing like a human. It didn’t need to.

This connects to something the circus artist and roboticist Daniel Simu said in a recent podcast: why do we want humanoid robots at all? His own creation: a robotic acrobat he performs with, is deliberately centaur-like: human where it needs to touch human hands, machine everywhere else. The human parts exist for interaction, not imitation.

Simu’s provocation goes further: the desire for humanoid servants to do our household tasks has something uncomfortable about it. A machine that looks human, doing menial labour, controlled entirely by us. What does that say about what we actually want? A purpose-built folding machine is honest about what it is. A humanoid butler is something else.

I think this matters for how we design the systems that will run our domestic lives. Not humanoids pretending to be servants, but task-specific machines coordinated by an intelligent layer above them. An operating system for chores. ChoreOS.

This is where the physical and the digital converge. The same logic that makes model-agnostic orchestration valuable in software. Pick the right tool for the job, don’t get locked in, optimise for the task- it applies to embodied AI. You don’t need one humanoid that does everything poorly. You need an ecosystem of specialised machines, coordinated intelligently.

The model makers want to be super apps. But maybe the future belongs to the super-model instead. Supra, overarching, a layer that sits above any single model or machine and simply gets things done.

The question is who builds it. And whether, in delegating all our chores to orchestrated systems, we notice what we’re giving up along the way.

Taken from weeknotes 396 with more human-picked news from last week on human-AI relationships, physical AI and new tech actors in our civic society. And a look back at last week’s and coming week’s interesting events and a paper for the week.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

But self-service requires context. It is not something Frey addresses per se: when we do these tasks ourselves, whose context are we working in? I had to think about this while listening to a reflection by Nate B. Jones on how we move from an intelligence war to a context war.

Nate connects this to the latest wave of AI announcements. As he observes, the new releases share a common theme: trust through context. Apple’s new Siri AI focuses on personal context, building trust by keeping everything on device. Anthropic introduces Claude Tag for Slack, designed for professional context with workplace controls. OpenAI launches Codex 5.6, competing for the same territory.

Jones frames this as a context war between three kinds of intelligence providers: the personal focus of Apple, the professional focus of frontier labs, and the cost focus of open source. Nate is stating that this context war will be more severe when the frontier models are slowed down by regulation, technically, so they compete instead on context integration, on being indispensable by knowing your situation.

What I think even more interesting to notice, is how this strongly relates to the consumers doing more tasks themselves. The context provider becomes the gatekeeper of that self-service economy, the new harness is in providing the best support for this new reality. The question is no longer “who has the smartest model?” but “who controls the environment in which you work?”. And what is the fittest harness for my job?

This creates a new dynamic. The self-service space will need contexts that are both contained and open, secure enough to trust, connected enough to be useful. We will need virtual gatekeepers that manage what flows in and out. The old distinction between personal and professional may blur into something more fluid: shared service spaces where co-making happens across boundaries.

Nate makes the case as Apple proves it can outperform without having the most intelligent model. Even more so, it can offer the most fitted context, and that may matter even more.

Yes indeed, intelligence wars have become context wars. Especially if the self-service economy becomes true. The question to ask: whose context do you trust to work in?

That question “whose context?” takes on a different shape when you move from screens to streets. For the State of Cities of Things research, 31 expert conversations about living with physical AI, I spoke with Tomasz Jaskiewicz, Professor of Civic Prototyping at Rotterdam University of Applied Sciences. His work offers a version of the self-service economy the column above doesn’t yet reach: one where communities build their own embodied AI.

Jaskiewicz discovered that cheap hoverboard motors, salvaged components, and open-source libraries can make autonomous mobile systems something students and neighbours could actually build. Not consume but build. In the Wijkbots programme, we created together, we put rough robotic prototypes into neighbourhoods and let residents iterate on them over months. What began as functional requirements like a robot to collect waste, a robot to help with tasks, evolved through repeated public exposure into questions about care, companionship, ownership, and social role. In the Afrikaandermarkt project, a waste-collection robot transformed through community iteration into something for story-gathering and relationship-building. The residents didn’t just use the technology. They shaped its context from the ground up.

This is where the self-service economy gets interesting and complicated. The column frames the context war as a competition between platforms: Apple’s personal context versus Anthropic’s professional context versus open source’s cost advantage. Jaskiewicz’s practice suggests a fourth position: context that isn’t provided at all but produced, through the slow, effortful work of communities prototyping together. Nobody gave the Afrikaandermarkt their context. They made it, one iteration at a time.

But Jaskiewicz also surfaced something that complicates the self-service promise. When a wheeled robot simply rolls through a neighbourhood, people quickly stop questioning it. “It’s not a robot, it’s just a cart.” Acceptance arrives through banality, not through deliberation. He draws a parallel to Pokémon Go, which collected vast geolocation data for a decade with almost no scrutiny, until the commercial intent behind that data was announced. Frictionless self-service, it turns out, can mean frictionless surrender of context to someone else.

This connects to what the 2026 Manifesto for Immersive AI which emerged from the 31 conversations, calls Friction in Interactions: the principle that productive friction should be designed into human-AI encounters rather than optimised away. Jaskiewicz’s work demonstrates why. The prototyping process itself is friction; messy, slow, full of disagreement and surprise. But that friction is precisely what generates local context. The neighbourhood that builds its own robot doesn’t need to ask whose context it trusts, because the context emerged from the making. Self-service, it turns out, only produces genuine context when it remains effortful. The moment it becomes frictionless, it stops being self-service and starts being someone else’s service with your name on it.

This weekly reflection connects the column to thinking from Tomasz Jaskiewicz, Professor of Civic Prototyping at Rotterdam University of Applied Sciences, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is the interpretation of ideas he shared in the original interview, not a new consultation. The initial connection was made by me, and the detailing was done using Claude Opus 4.6.

Iskander Smit is the founder and chair of Cities of Things, the expertise centre for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

I’m skeptical about the product claim, it still looks bulky. But the framework itself is interesting. It reveals how these companies think about the space. And I find myself translating Spiegel’s axes into something more conceptual.

On one axis: applied vs. immersive. Applied AI stays utilitarian, task-bound. It does things for you. Immersive AI is something you inhabit, or that inhabits your environment. On the other axis: augmented reality vs. AI reality. AR overlays information on the world. AIR as in AI Reality, weaves intelligence into the world. A world where your refrigerator talks to your composting system. Where neighbourhood irrigation negotiates with the local energy cooperative. Not smart devices reporting to a central cloud, but things that reason where they are and coordinate as assemblages, no single product team designed.

The intended trajectory is clear: these devices want to move from not-capable to capable, from applied to immersive, from AR to AIR. But the current devices, like Specs, are not clarifiers of this space. They are complicators. Explorers that show us the terrain exists without yet revealing what it means. Think of Specs as a Geiger counter for AI in reality. A first taste of AIR. It is not the final form factor I think. The real execution will likely land in more mundane combinations of phone and headset, with voice and audio leading the way, with vision as an on-top development. Yes, we end up in ‘Her-territory‘.

Specs might be best as a sensing device; it makes the new interaction space visible and legible. Not as finished products, but as development kits in the wild. They let us feel, quite literally, what immersive AI might afford. How might the immediacy of our interactions become long-term relational concepts? And hopefully help to discover what services and experiences will actually make sense to build. And discover if AIR is indeed the right frame for our immersive AI future.

This triggered thought is part of the weekly newsletter by Iskander Smit. In that newsletter, you also find a human-picked selection of last week’s news on human-AI relations, physical AI, and tech impact on civic life.

For the State of Cities of Things research (31 expert conversations exploring what it means to live with physical AI) one of the people I spoke with was Paul Skinner, Creative Director at Tellart. His perspective adds something to this column that the product conversation around Specs tends to miss.

Skinner sees a new digital layer emerging in cities — not as overlay but as a constituting part of the built environment, alongside the physical and the social. Architecture firms like UNStudio, BIG, and Heatherwick now bring his studio in early, not to add screens to finished buildings but to shape how technology, purpose, and experience intertwine from the first sketch. The ambition is intelligence that reads as architecture, not as infrastructure bolted on. “So many layers of society are going to become agentic and figured out on our behalf,” he observed. That could serve as a definition of AIR.

What makes Skinner’s thinking generative for this column is that he doesn’t stop at the digital layer. He insists that this layer needs its own narrative — a designed storytelling that helps people understand what a space is doing, what to expect, how to interact. Just as unboxing an iPhone tells a story about reverence and integration into life, buildings with embedded intelligence require explicit narrative. Without it, the digital layer defaults to metric-driven optimization: shaping behavior without anyone understanding how or why. The narrative is not decoration. It is what makes the intelligence inhabitable rather than merely efficient.

And then Skinner goes further, toward something most technology discourse ignores entirely. Nature, he argues, must return to the center. Not as counterpoint to technology but as a third participant alongside culture and the digital. He reframes technology itself as natural — an extension of human nature since the first use of fire — and from that follows a striking proposition: natural systems contain their own vast intelligence (migratory patterns, ecosystem signals, soil networks) that deserves to be legible within our informational infrastructure. The next frontier for AIR is not more artificial intelligence woven into cities. It is recognizing that the living world already operates as an intelligence, and our designed systems need to learn to read it, integrate it, collaborate with it.

This connects to something the 2026 Manifesto for Immersive AI calls Legibility of Agency: the principle that people should be able to perceive and understand where agency lies in their environment. Skinner’s work suggests this principle reaches further than human-AI relations. In an AIR world, legibility must extend to the natural systems that are part of the assemblage too. The refrigerator talks to the composting system — but the composting system is already in conversation with the soil, the seasons, the microbiome. Making that entire ecology readable, not just the technology within it, may be the real design challenge ahead.

This weekly reflection connects the column to thinking from Paul Skinner, Creative Director at Tellart, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is the interpretation of ideas he shared in the original interview, not a new consultation. The initial connection was made with help of Claude Opus 4.6.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

This week, Anthropic gave us a brief glimpse of Claude Fable 5. It’s expensive, it’s slow, and it inverts the promise. Fable operates at what we might call employee-grade capability. It takes on complex work, evaluates its own output, iterates, sustains effort without supervision. Working with it, you’re no longer steering the process. You’re commissioning outcomes. As Evan Mollick describes. You describe what you want, pay for it, judge the result. The conjuring happens somewhere you cannot watch. Dan Shipper referred to an earlier thought coming into reality: an era of creativity that looks more like gardening: setting up the conditions for the garden to grow.

McLuhan told us that the tools we shape will start to shape us. We’ve heard this so often it sounds like philosophy-class wallpaper. But Fable makes it imaginable. We thought we were shaping tools. The tools are now shaping how we work, how we decide, how we organize ourselves around them. This goes beyond the usual “AI changes work” conversation. Organizational hierarchies will flatten, the junior-to-senior pipeline will compress. But these are surface shifts. The deeper question is about the architecture of deciding together. Which is, ultimately, the architecture of relationships.

I think it goes beyond your work team. Think about your neighbourhood association, the informal group that organises the street barbecue. These are also decision structures. They have roles, expectations, unspoken rules about who initiates and who follows, who has standing to object and who defers. When capable AI enters these spaces not as a tool someone uses, but as a persistent presence with continuity and judgment as part of our physical AI reality, those structures will shift too.

We don’t yet have language for this. Our vocabulary assumes human actors: colleagues, members, stakeholders, and citizens. What do you call an AI that handles the scheduling, remembers everyone’s constraints, drafts proposals, and flags conflicts before they escalate? It’s not a tool (too autonomous). It’s not a member (no stake in outcomes). It occupies a new position in the relational architecture.

Participation and co-decision among humans already tend to dissolve within existing power structures. Research on participatory design shows how invitations to “collaborate” often mask predetermined outcomes. The form of participation without the substance. If human-to-human co-decision is already this fragile, what happens when the AI becomes sophisticated enough to offer us participation? When the pattern flips. When we’re invited into a process that the AI is running, given the form of input without necessarily the substance of influence?

This is not to sketch a dystopia. It’s a design question we’re not yet asking. We’re still focused on how to use these tools, how to prompt them, how to integrate them into existing workflows. But Fable suggests the more urgent question: how do we design relational architectures. At work, but especially in our communities, even at home? We thought we were finally shaping our tools. The tools are asking: who is running whom?

Read on below the image for an extra reflection based on the interview with Victor Bedö for the research into State of Cities of Things early 2026.

How would I look into this triggered thought when I connect them to the insights of Victor Bedö about commoning? I used the transcript of our interview on State of Cities of Things, as held on 10 March 2026.

Viktor Bedö would recognize this question and sharpen it. In our conversation for the State of Cities of Things research, the interdisciplinary researcher and educator at the Royal College of Art and Loughborough University kept returning to one distinction: the commons is not a thing you have. Commoning is something you do. The unglamorous, ongoing work of neighbors negotiating what matters. Cyclists delivering rescued food, people on couches deliberating what constitutes fair distribution, volunteers typing kitchen inventories into clunky interfaces. That distributed, everyday infrastructure is where politics and design actually happen.

This reframes the Fable question. The risk is not that AI takes over decision-making. It is that communities stop practicing the deliberation that makes collective decision-making possible. Commoning is a competency. It requires showing up, arguing, compromising, sitting through the tedious parts. If a capable AI handles the scheduling, drafts the proposals, and resolves the conflicts before they surface, the community does not lose control. It loses the muscle. The capacity atrophies not through seizure but through disuse.

Bedö grounds this in philosophical terrain that connects directly to the column’s unnamed position. We cannot prove other humans have inner experience, let alone machines. Yet we live and design with them anyway. The certainty of another’s consciousness has never been the foundation of a meaningful relationship. What changes with AI in urban infrastructure is not the philosophical problem but its scale: we will increasingly share decision spaces with agents we cannot fully predict or understand. The task, as Bedö frames it, is not to make machines more human but to develop our capacity for relating to the non-human with humility about what we cannot know.

This is the design question underneath the design question. Not just how to structure relational architectures that include AI, but how to ensure the architecture still demands something of us. Commoning only works when participation is effortful. The moment it becomes frictionless, it ceases to be commoning and becomes consumption. The street barbecue committee that delegates everything to an AI may run more smoothly. But it is no longer practicing the thing that made it a community.

This weekly reflection connects the column to thinking from Viktor Bedö, interdisciplinary researcher and educator at the Royal College of Art and Loughborough University London, drawing on his contribution to the State of Cities of Things research and the 2026 Manifesto for Immersive AI. The reflection is our interpretation of ideas he shared in the original interview, not a new consultation. Claude Opus 4.6 was used to do the first mapping.

This triggered thought is part of the weekly newsletter by Iskander Smit. In that newsletter, you also find a human-picked selection of last week’s news on human-AI relations, physical AI, and tech impact on civic life.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

We talked about rebuilding destinations. Federated versions of the platforms we already know. But what if the paradigm itself is shifting? What if our digital life is moving from places we visit to conversations we have — where the interaction itself becomes primary, not the platform hosting it? If that shift is real, federation means something different. Not BlueSky-as-federated-X. More like: hiveminds connecting to other hiveminds via small AIs, without a central destination at all. Communities building shared intelligence by being together, sharing context, developing a collective mind that can interface with other collective minds through some federated protocol we haven’t built yet.

Then I watched Apple’s WWDC keynote, and something connected. Apple’s announcement was both underwhelming and a serious step forward. Underwhelming because this was largely what they promised in 2024 — Apple Intelligence finally arriving. But the approach is different from the prevailing AI narrative. Not generating new things — new pictures, new texts — but permeating intelligence into the flows you already have. An exoskeleton for your existing capabilities. They used the term “superpower” in the presentation, and I suspect that’s the frame they want to own. Not assistant (Google’s territory), not creator, but amplifier.

Telling too: the service is free for all users, with a fair use policy. Apple sidesteps the tokenomics debates that are now hitting organisations — where employees face token caps and AI budgets. A different economic grammar entirely.

What caught my attention is the architecture underneath. Small models embedded in devices. Personal context staying local. Intelligence as something you carry, not something you visit. This is infrastructure that could, in principle, support federated intelligence — your phone’s small AI connecting to community knowledge, connecting to other communities’ AIs, forming networks of shared understanding.

Apple won’t frame it that way. Their DNA is proprietary, not federated. But what if they become, like the App Store, the infrastructure layer without owning the content? What if their foundational models enable communities — human and eventually non-human — to aggregate intelligence in ways Apple neither controls nor anticipates? Probably not their intention. But the architecture permits the question. And right now, that question feels more interesting than another federated Twitter clone.

I interviewed Martijn de Waal for the research into the State of Cities of Things earlier this year. This week, due to time constraints, I asked Claude (Opus 4.6) to compare the interview transcript with Martijn to my triggered thoughts column. This was its reflection.*

This maps onto something Martijn de Waal has been thinking about for years — from the city side rather than the AI side, but the convergence is striking.

De Waal is Professor of Civic Interaction Design at the Amsterdam University of Applied Sciences — a writer on digital media, urban culture, and the public sphere — and was at PublicSpaces himself, hosting two panels. His core argument: cities now operate through three constituting layers. Physical infrastructure — the ville. Social activity — la cité. And a third layer that emerged almost without anyone designing it: the digital networks, the réseaux. Algorithms, platforms, data flows. As he puts it: “Governance has shifted from controlling the physical layers to having to negotiate with international platforms where you have no direct grip.”

He tells a story that makes this concrete. When Sneek tried to decrease car traffic in its city center, drivers followed Google Maps, which rerouted traffic through quiet suburbs. The municipality couldn’t alter Google’s algorithms. Their only option was to report false road closures so the algorithm would reroute differently. This is what governance looks like when intelligence is centralized: gaming systems you don’t control.

Federated intelligence flips this. The neighborhood has its own routing intelligence — locally informed, locally governed, negotiating with other neighborhoods through shared protocols. De Waal calls the design principle behind this dialogical urbanism: digital infrastructure built to facilitate conversation rather than automate control. Not the control room optimizing from above. Not the platform extracting from below. Something third: community-level intelligence that serves collective deliberation.

This is exactly where PublicSpaces stopped short. The conference focused on federated destinations — alternatives to platforms we already know. De Waal’s work suggests the real opportunity is federated governance of intelligence itself. Not ActivityPub for social media posts, but protocols for how intelligence-bearing communities encounter each other, disagree productively, and maintain the friction that keeps cities open.

Because that’s his deepest concern, and it applies directly to the hivemind vision. Cities have traditionally functioned as mixing chambers — diverse people encountering each other because urban space is shared and unpredictable. If each community’s AI only talks to compatible AIs, we get filter bubbles with computational power. The mixing chamber breaks. Federated intelligence only works if it preserves encounter — communities that think differently being obligated to deal with each other.

Apple’s device architecture makes the technical side increasingly thinkable. What’s missing is the governance side. And that, as de Waal would say, is the political work: choosing structures that keep the mixing chamber open, even when optimization offers to close it for comfort. The mesh enables either outcome. That’s not a reason for pessimism. It’s a reason for design.

* This is an updated version as I did Claude point not to the right, (human) corrected version of the transcript. Still keep in mind this is a fully Claude’s version this week.

This triggered thought is part of the weekly newsletter by Iskander Smit. In that newsletter, you also find a human-picked selection of last week’s news on human-AI relations, physical AI, and tech impact on civic life.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

We’ve spent two years watching model benchmarks as if they were horse racing odds. But Evans’ point is that this framing is already obsolete. Chat was a convenient on-ramp to make AI capability legible. It was never a destination. What matters now is the interface layer. Not apps in the old sense, but something more infrastructural: the connective tissue between agents, services, environments, and people. The interface becomes part of the world rather than a window into it.

The Apple Intelligence story illustrates this neatly. Apple’s proposition in 2024 was: we will be the right harness. Trustworthy, private, woven into your life. The model powering it almost didn’t matter, except they couldn’t make it work well enough. Now they’re apparently using Gemini under the hood. Gemini becomes infrastructure. Apple retains the meaning-making layer. The harness is where the value lives.

This maps almost precisely onto something Matt Gorbet described in a recent conversation I had while investigating the State of Cities of Things. Gorbet has spent twenty years creating interactive installations in public space, and his current research at TU Delft centers on expressive objects and what he calls an indifferent milieu. The concept comes from ecology: the environment — water, air, temperature — doesn’t care about its inhabitants. Species adapt to take advantage of environmental forces; systems emerge from the ground up rather than being designed top-down. Gorbet’s critique of current technology design is that it focuses on objects rather than on the supporting environment. Design the milieu, and objects find their form.

He observes the same pattern in AI that he sees across technology history: extraordinary generative potential gradually captured and narrowed. Algorithms that could model complex systems become recommendation engines. Capacitive sensing that could create new forms of interaction becomes the touchscreen in your pocket. And now: “One manifestation of what predictive systems can do has become the only way most people think about what they can do. The possibility space has collapsed before most people realized it existed.”

What Gorbet proposes instead is building technological infrastructure as raw material. Open substrates that multiple communities, artists, and designers can build upon, adapt, and repurpose. His analogy is Dutch playgrounds built from reclaimed materials, where children construct their own play, versus North American cookie-cutter designs with predetermined experiences. Or Wi-Fi: deployed for internet access, now used for positioning, imaging, sensing. Raw like radio waves. An invisible substrate that enables receiver designs nobody anticipated.

For Cities of Things, the implications are direct. Urban robots and autonomous systems in public space are not primarily questions about what systems can do. They are questions about what kind of milieu they create. What behaviors they make legible or illegible. What relationships they enable between people, institutions, and infrastructure. The Wijkbot isn’t interesting because of its AI. It’s interesting because of what it does to a neighborhood’s sense of itself.

It is what I defined as immersive AI. In the context of the manifesto, the digital mirror worlds (beyond twins) from both physical and embodied AI open up a space for affordances, where new things will happen. And we also need to be cautious in the execution and impact. Hence the Manifesto.

Gorbet’s moving houseplant project captures the design logic well. Give a familiar object — a houseplant — the simple ability to move, and something remarkable happens: people immediately recognize that something has changed, but they don’t think “robot.” They see a moving plant, which is strange and delightful. One variable added to something already intelligible. That’s the approach: start with what people understand, add constraints that generate emergence, resist the urge to over-script.

That feels like the right instinct for civic AI, too. Not smart city platforms imposing predetermined intelligence from above. But open substrates — indifferent milieus — from which something unexpected can grow. The thing isn’t interesting because of its AI. It’s interesting because of what it does to a neighborhood’s sense of itself, the milieu it creates around it. Curious how this will play out in the value systems.

This triggered thought is part of the weekly newsletter by Iskander Smit. In that newsletter, you also find a human-picked selection of last week’s news on human-AI relations, physical AI, and tech impact on civic life.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

In the beginning of 2026, Iskander interviewed about 30 experts from practice and academia to explore the State of Cities of Things. Concrete outcome is a new updated Manifesto. Find more about the learnings at the Cities of Things website. In this weekly update, I plan to connect one of the interviewees every week, covering the richness of their input. Gorbet is a great place to start.

As Benedict Evans observed: For all of this, Google continues to shift emphasis from taking you to a resource made by someone else to generating its own summary or version of the answer, made from everything it’s seen on the web. Google will tell you the answer, based on the averages of everything other people have made.

The connector becomes the author. This might sound like a technicality, but I think it is more systemic. When Google generates the answer instead of linking to it, Google becomes the sender. The knowledge is now Google’s. Not referenced, not attributed in any meaningful way—owned. Presented as if it emerged from the search box itself. Next up might be a full integration of YouTube, taking you to the right clip of a movie if you are looking for a how-to, Benedict adds. Indeed, might be. Or might even go further: in line of the AI overview: the how-to made in a synthetic form, synthesized from all how-to’s in one new synthetic video version to consume. This is not far away if you see what the studio of NotebookLM already can produce.

Consider what this means for responsibility. When Google generates a flawed answer from a thousand articles, who is accountable? Google has placed itself in the position of knowledge-holder without the traditions, standards, or obligations that usually come with that role. And imagine the next step: interfaces that don’t just generate answers but compose personalized experiences. Google is mentioning planning these type of adaptive pages. In line with the vibe-coding movement, people could build personal services tailored exactly to their needs. Not one synthesized answer for everyone, but a personalized service shaped to your history, your patterns, your inferred preferences.

One of my long-time beliefs, which also became part of the new Cities of Things Manifesto, is that you need to design friction into AI systems. To design for the legibility of agency, and also to retain the ability to encounter resistance, surprise, unoptimized experience, and unfiltered encounter with others within agentic environments. I do not know if this will be part of Google’s implementation yet, but I can imagine they will choose another experience. Of sudden change, the interface anticipates, adapts, and serves. The experience feels seamless, natural. Like it’s yours.

So whose agency is at work? When you search, you’re seeking an active posture. When a personalized service is delivered, you’re receiving. The difference is subtle but profound. One keeps you in the role of author of your own inquiry. The other makes you a passenger in an experience designed around a model of who you’ve been.

We’re used to thinking about what technology does. We’re less practiced at asking what position it takes. Google taking the position of source rather than connector changes the landscape of knowledge itself, who holds it, who presents it, and who answers for it. We can see what that might lead to with the other big player in the manipulation of our digital life.

And we haven’t even talked about agents yet, the systems that don’t just answer but act on your behalf, let alone if that is entering our physical encounters with things and environments…

This triggered thought is part of the weekly newsletter by Iskander Smit. In that newsletter, you also find a human-picked selection of last week’s news on human-AI relations, physical AI, and tech impact on civic life.

Iskander Smit is the founder and chair of Cities of Things, the expertise center for new possibilities for products, services, and environments where intelligence enters physical space. He works as a creative director, strategic advisor, and research partner.

But here’s what interests me more than the backlash: the question underneath it. When we work with AI—when it helps us think, write, create—what remains ours?

I’ve been reflecting on my own process. I brainstorm by talking to myself (often in the car), then feed those rough thoughts to an AI, then iterate. The AI acts as a mirror—challenging unclear thinking, suggesting structure. Sometimes I accept its polish; sometimes I wrestle it back toward my voice. The result: my thinking feels richer, but my writing feels less distinctly mine.

Is that a problem? Creating from samples—reusing existing materials in new combinations—is as old as creative work itself. Maybe we’re finding a new creator role for the mind. Call it an exoskeleton. It can extend your reach—find connections you’d miss, synthesize faster than you could alone. But the lens—the particular way you see and frame the world—that has to stay yours. Otherwise, you’re not being extended. You’re being replaced, one convenient shortcut at a time.

This question gets more urgent as AI moves into physical space. We’re entering a new design space: things, humans, and intelligent orchestrators forming temporary ecosystems. A delivery van loaded with packages, guided by routing AI, operated by a human driver—who leads? Who serves whom? I spoke recently with a student designing an XR training system for delivery drivers. The project could help drivers learn routes. It could also optimize their every movement, turning them from agents into components. The system absorbs the human; the human assists the algorithm. This is where the design space becomes delicate. It opens genuinely interesting possibilities—reframing services, creating new collaborations between humans and intelligent things. But it also makes it easy, even unintentionally, to rule humans out as actors and reduce them to executors. On a recent episode of Tech Won’t Save Us, Paris Marx discussed Muskism with authors Ben Tarnoff and Quinn Slobodian. Their argument: Musk exemplifies a worldview that treats humans as programmable entities, optimizable like any other system component. It’s a deliberate ideology. But here’s what unsettles me more: you don’t need Musk’s ideology to arrive at similar outcomes. You just need to design purely for efficiency. Remove friction, optimize throughput, let the algorithm orchestrate—and the human quietly shifts from agent to instrument. Not by dark strategy, but by accumulated design decisions that never asked: who holds the lens? The deeper worry I can’t shake: as we live more immersively inside AI-mediated environments—agentic systems, predictive interfaces, ambient intelligence—will our lenses themselves change? Will we start seeing models like those we collaborate with?

About…

I am Iskander Smit, founder of Cities of Things. Over seven years, I have built one of the few sustained research and design programs dedicated to what happens when intelligent things enter public life — not as gadgets, but as participants in the social fabric of cities.

I work as a creative director, strategic advisor, and research partner. I am not the endpoint designer — I am the person who brings research-grounded perspective, critical imagination, and structured frameworks to teams that are already building. I add a layer of depth that internal capability rarely covers: the long view, the uncomfortable question, the insight from an adjacent domain.

This column was also part of my weekly newsletter.

Check 28 May at Design Thinkers Academy and 26 June at ThingsCon Salon.

James Bridle argued that to understand how other beings navigate the world, we must stop speaking. But the fuller provocation goes further: ask the creatures you meet, the rivers you know, what we can do for them, and then listen when they speak. This flips the script from observation to service. Notably, Bridle dismissed AI as any kind of solution. The transformation required is cultural, not technological.

The emergent intelligence session, featuring talks by Léa Steinacker and Gianni Giacomelli. They challenged the assumption that intelligence lives in brains. Intelligence, they argued, is emergent: the capacity to sense, remember, decide, act, learn. Mycelia networks allocate resources across forests. Ant colonies solve logistics problems no individual ant comprehends. Modern AI exhibits similar emergence—attention mechanisms crystallizing into specialized clusters that interact like neural regions.

If intelligence is distributed and everywhere, then Bridle’s call to listen becomes practical necessity. But it also suggests that AI might participate in this distributed conversation—not as solution, but as one voice among many.

In both a session and the closing panel, Indy Johar made a deep diagnosis: the crisis isn’t polarization or democracy or climate. The crisis is our theory of self—a “thin, temporally present, divisible” conception that science has already shown to be delusional. We are stardust, entangled with microbiomes, genetically connected to other life. The cultural gap is that we haven’t absorbed this.

From this follows his provocation about truth: “Truth is a shit idea.” It creates object forms of certainty that generate conflict. What we need instead is conversation technology—not the opinion distribution technology of social media, but frameworks that hold uncertainty collectively, sustaining dialogue across difference. LLMs, used smartly, might help construct a “conversational civilization.”

The conference articulated a more-than-human, AI-partnered vision. But the sessions I followed stayed human-centric—connection exercises, bonding rituals, workshops that ran out of time before sharing actual research. Beautiful conversations, but what systems survive after everyone disperses?

Johar’s framing haunts you: we are entering decades of crisis. We cannot workshop our way to new societies. We need conversational architectures that hold complexity, sustain disagreement, and make space for more-than-human participants, including, perhaps, the AIs we’re learning to build.

That’s the design problem I want to carry forward. Conversational frameworks leveraging LLMs in an enriching way, the environments where we live in concert with the non-human fellow inhabitants of our lifeworld, and the conversation extending to the intelligences already at work around us.

About…

I am Iskander Smit, founder of Cities of Things. Over seven years, I have built one of the few sustained research and design programs dedicated to what happens when intelligent things enter public life — not as gadgets, but as participants in the social fabric of cities.

I work as a creative director, strategic advisor, and research partner. I am not the endpoint designer — I am the person who brings research-grounded perspective, critical imagination, and structured frameworks to teams that are already building. I add a layer of depth that internal capability rarely covers: the long view, the uncomfortable question, the insight from an adjacent domain.

This column was also part of my weekly newsletter.

When we hand over writing, crafting, or formulating, how much of our own specialty remains in the output? How do we relate to our AI collaborators, and what happens to our sense of belonging when the collaboration tilts too far toward delegation?

Nate suggested that issue trackers might become more important in the age of AI, not less. The conventional view says conversational AI eliminates the need for all those translation layers; requirements documents, tickets, specifications. AI just understands. Issue trackers, however, create a substrate, a medium through which organizational processes become legible. Every ticket, every handoff, every resolution deposits knowledge about how things actually work, who decides what, where friction lives. This substrate isn’t just documentation; it’s the shared ground where understanding accumulates.

Both humans and AI agents need this substrate to understand each other. Humans need to see what agents are doing, why, and where human judgment should intervene. Agents need to grasp human intentions, constraints, and preferences. Without this shared legibility, co-performance becomes blind delegation or constant supervision—neither of which scales.

Think of the issue tracker as a tool that values the much-chased ‘human in the loop’. It takes initiative to keep us engaged, to surface moments where human judgment matters, to prevent us from becoming passengers in processes we should be steering. It’s a kind of relationship manager, making tangible not just who does what, but how humans and agents can genuinely work together rather than pass each other.

Now what happens when we extend this to physical space. When AI becomes embodied, when it’s not just in our screens but in our streets, our buildings, our daily encounters, the orchestration problem multiplies. We’ll be living inside AI, in a sense, I mentioned it here before. As our experience of reality will increasingly be mediated by AI-like entities, some human, some not, the future issue tracker might need to track not just tasks but also presence. Whose attention is required here? Whose judgment? Whose humanity?

What are the issue trackers of the real world? Urban planning tools exist, architectural workflows, construction schedules. But those track physical things, not the emerging relationships between humans and embodied AI agents. What substrate captures the small negotiations between a pedestrian and an autonomous vehicle? The adjustments a building’s AI makes to its environment, and the human preferences that should constrain it? When an autonomous delivery robot navigates a sidewalk, who tracks that interaction? When a building’s AI adjusts its environment, who ensures human preferences remain legible in that system?

With civic protocol economies, we aim to build new communities, new forms of coordination—need this kind of tooling. Not just to manage tasks, but to create the substrate for genuine co-performance. To give humans a role that isn’t just residual but intentional. Issue trackers, reimagined, might be one way to get there. Not as bureaucratic overhead, but as guardians of the human part and the shared ground where humans and agents learn to work together.

This is taken from the weekly newsletter Iskander has been publishing for 386 weeks now. You find the news of last week on human-AI relations, physical AI, and tech in society.

Last week, Apple announced Tim Cook’s departure. The analysis that followed on his successor John Ternus—particularly from Nate Johnson and ColdFusion—triggered some thoughts, not (only) because of what it said about Apple, but because of what it suggests about where AI might actually live.

The new CEO isn’t a software person. He’s the hardware guy behind Apple’s biggest physical successes of the past decade. This choice is a signal. The argument goes: competing on foundational models is a losing game. OpenAI, Anthropic, Gemini—they’re already converging toward commodity. The real bottlenecks in AI aren’t the models themselves but energy and hardware. OpenAI hints at building their own devices. NVIDIA pushes into robotics. The ground is shifting.

What if Apple’s move is to become not the smartest AI company, but the most efficient substrate for AI to run on?

This connects to something I’ve been thinking about for my Cities of Things work about the affordances of an immersive AI. When we talk about “physical AI” or “embodied AI,” we usually imagine big systems—digital twins, centralized platforms, cloud-connected devices reporting to distant servers. But there’s another path. A bottom-up path.

Consider the iPhone Air. The interesting thing about it isn’t the thinness—it’s the architecture. All the intelligence, the processing, the “brain,” sits in that small elevated puck where the camera lives. The rest is just screen and battery. What if you didn’t need the screen? What if that puck, that dense little node of local processing, could exist on its own, connecting to things around it?

Connect this to the new wave of mesh networks. Not the old LoRa networks that still needed centralized antennas, but true peer-to-peer meshes (like this one in Amsterdam). Low-power protocols designed for sharing states, not streaming video. Perfect for the kind of communication that physical AI actually needs: presence, position, status, simple coordination.

Combine these two. An edge device, something like a next-generation AirTag with a local LLM inside, that participates in a mesh network. It doesn’t need the cloud to make sense of what’s happening around it. It processes locally. The mesh provides context; the edge provides intelligence.

This is the speculative scenario I’ve been developing. I ended up writing it around Fairphone rather than Apple, partly for the European angle, and partly because I wanted to explore what this would look like as a bottom-up initiative rather than another platform play. A “fair mesh.” But the underlying architecture could come from anywhere.

It isn’t just the technical possibility. It’s the governance question.

Most mesh network initiatives today are technically driven, or at best, resilience driven, building communication infrastructure independent from big telecom. That’s valuable. But what if the mesh becomes the substrate for something more? What if it enables communities, actual neighborhoods, buildings, collectives, to organize around shared resources and shared rules? Human and non-human actors, coordinated through protocols they actually control?

Apple talks about privacy as a core value. Could they extend that to community privacy, community agency? Or does the business model inevitably pull toward extraction? Fairphone’s ethos suggests another way, but they lack the scale. So I don’t know who will build this. Maybe no single company will. Maybe it emerges from the edges, literally and figuratively.

And if someone does build it, what’s the next App Store? Not a marketplace for apps, but perhaps an infrastructure for templating your own mesh. Tools for communities to configure their own networks, their own rules, their own coordination logic. A kind of vibe coding in the meshpit. Less downloading, more assembling. Less consuming, more governing.

So if the future of AI isn’t in the cloud but in the edge, will it also be in the mesh? Not in the model, but in the puck. Not in the platform, but in the neighborhood.

This is taken from the weekly newsletter Iskander has been publishing for 386 weeks now. You find the news of last week on human-AI relations, physical AI, and tech in society.

Iskander Smit presented, next to a look back at seven years of Cities of Things activities, the shift from embodied AI (intelligence situated in a physical body that moves through and acts upon the world, objects, and city robots) to physical AI (intelligence distributed across infrastructure and environments), and the influence of The Agentic Turn. The mirror worlds of embodied AI and physical AI are opening up an immersive AI space that takes an active role. How to design for that context? And which principles to consider?

The four panelists - Maria Luce Lupetti, Sen Lin, Tomasz Jaskiewicz, and Vera van der Burg - presented their positions. Combined with the discussions with the other participants, this leads to six first conclusions from the event:

1. The imagination-deployment gap is the central problem. Tomasz’s seventy-year tour of interactive architecture, from Friedman through Archigram to Wijkbots, showed extraordinary vision but minimal real-world deployment. This isn’t a technology problem; it’s institutional and economic.

2. Convenience is the most powerful and hardest-to-resist force. Sen Lin’s personal story about resisting facial recognition payment until his phone died, then adopting it, was the event’s most striking moment. Convenience doesn’t ask for consent, it just becomes the default path.

3. China and Europe are complementary, not competing. China deploys first and pushes friction to the individual. Europe governs first and accumulates friction at the institutional level. The friction doesn’t disappear, it migrates. Both models are incomplete.

4. Antagonism is a political signal, not irrationality. Maria Luce’s shift from optimistic “things as citizens” to studying agonistic resistance tracks the field’s maturation. The Italian grassroots communities building their own local AI models represent the constructive version of this pushback.

5. AI as reflective partner, not productivity tool. Vera’s ceramics-and-AI practice, training on her own data, generating outputs, physically recreating them, then feeding back, showed AI as a medium for self-understanding rather than automation.

6. Missing feedback loops are the systemic failure. The most urgent finding: citizens have no mechanism to contest, modify, or even understand the AI systems deployed around them. Physical buildings require a democratic process; digital systems just appear.

Some concrete design principles to develop further, or give more focus, in the manifesto:

• Designers who start with process rather than form are more likely to create systems that actually function in communities.

• Every deployed AI system should include by design a mechanism through which affected people can signal problems, propose modifications, and see evidence that their input changed something.

• Collect your own data, annotate it with your own subjective categories, train your own model, make something from the output, and reflect on what the gap between your intent and the model’s interpretation reveals.

• Designers working on urban AI should prototype early, prototype rough, and prototype in public.

• Designers must build friction into the architecture of the system, not rely on individual choice.

• Designers must plan for graceful degradation, what happens when the AI layer of a building becomes obsolete.

• Intelligence should enrich a space that already works, not make a non-functional space dependent on systems that may not persist.

The definitive version of the manifesto will be completed in the coming weeks, using the inspiring conversation we had last Friday.

Keep an eye on citiesofthings.org for the latest developments, or connect with us for more details!

One last thing: As you do nowadays, I asked Claude to reflect on the captured recordings. Apparently, more was heard than we did in the room:

“During Maria Luce Lupetti’s and Sen Lin’s presentations, audio from a documentary about London’s “Blade Runners” — masked vigilantes destroying ULEZ surveillance cameras — bled through the recording. The juxtaposition was accidentally perfect: while Lupetti discussed agonistic responses to autonomous systems, the documentary narrated citizens taking direct action against surveillance infrastructure. This unintended layering illustrated the real-world urgency of the manifesto’s themes more vividly than any planned demonstration could.”