Most conversations about AI agents in organisations are shaped by a quiet, background anxiety. If agents can analyse, synthesise, draft, review, and execute, what is left for the humans?
Let me start by saying I take this question seriously. I hear it from founders, from partners, from teams I've coached. It is a real question, and dismissing it with slogans about creativity or AI not being able to count letters does not help anyone. Those answers were already wearing thin before 2026, and the agentic shift has worn them through completely.
I want to offer a reframe that I think is both more honest and more hopeful. The anxiety comes from a premise that was already somewhat wrong. Most of us were taught that humans are valuable to organisations because of what they produce. If you believe that, this moment is brutal, because agents will produce more, faster and often better than we ever did. But if you believe humans are valuable because we can feel what change the world needs, this moment is not a threat. It is a clarification of why we were always here: to feel the purpose, see the potential, and share your gifts in organisations working towards that purpose.
In an organisation properly designed for AI agents, the responsibilities humans carry become more structurally essential, not less. Just in ways most of us were never actually given in the first place. The responsibility of sensing.
An organisation is a living system with a reason to exist. To be as effective as possible, often, this reason to exist or purpose, is split up in differentiated sub-units of the overall purpose, into something we call roles. These roles can be energized by an agent (human or AI) to give direction and movement to the (sub-)purpose that the role carries. All these role-fillers don't just bring the knowledge about their role, but also a whole subset of history, wisdom, experience, and sensory input, that help the role-filler decide what's next and how to best manifest the purpose in the existing living environment.
This is the frame I keep coming back to, because it shifts almost everything about how we think about agentic work. Organisations are not machines with humans operating them. They are living systems with a purpose, sensing and responding to the world around them. That purpose cascades through circles (teams) and roles in what we call a hierarchy of purpose, not people. Each circle carries a sub-purpose that serves the organisation's broader one. Each role carries a sub-purpose that serves the circle's.
So far, this framing is very familiar for anyone who has worked with role-based governance methodologies. What the agentic era forces us to notice is the piece we had been taking for granted. A role is inert without someone, or something, giving their energy to it. Purpose stays as text on a page until a role-filler meets the actual world with it. And what they meet the world with, from where they meet it, is what changes the organisation's behaviour.
The framing of human role-fillers as sensors is not new. Brian Robertson, one of the founders of Holacracy (one of the methodologies I’ve been working with and teaching in for a over a decade), put it precisely a long time before AI agents arrived. In a 2015 interview accompanying his book, he said:
“An organization, like a plane, is equipped with sensors - not just lights and gauges, but also the human beings who energize its roles and sense reality on its behalf. Too often, an organization's "sensor" has critical information that is ignored and therefore goes unprocessed. One individual notices something important, but no one else sees it and no channels are available to process that insight into meaningful change. In this way, we often outvote the low-voltage lights of our organizations. Our organizations become aware of whatever they need to respond to in their world through our human capacity to sense the reality around us. And we humans are all different - we have different talents, backgrounds, roles, fields of expertise, and so on - so we naturally sense different things. Where there are multiple people, there are multiple perspectives.”
- Brian Robertson, 2015
Read that again. This was written in a human-only working world. The role of the human in a well-designed organisation was never primarily to produce. It was to sense reality on the organisation's behalf, from inside the role they energised, and to feed that sensing into a structure that could create valuable output with it.
That framing gives us the ground we need to think clearly about what changes when AI agents arrive. Nothing about the human role is being taken away. What is changing is that there are now other kinds of role-fillers in the organisation, and they sense too, but in a fundamentally different way.
Both humans and AI agents are role-fillers, and both bring perception to the roles they energise. But they perceive the world through entirely different sensory equipment. They inhabit different perceptual realities. Neither is more or less real than the other. They are different.
The biologist Jakob von Uexküll gave us the cleanest concept for this way back in 1909. He called it the Umwelt, from the German word for environment, but he did not mean an organism's physical surroundings. He used it to mean an animal's sensory environment, the perceptual world that animal inhabits. He recognised that every species has its own set of sights and smells and sounds and textures that it can perceive, and that might be very different from what another creature can perceive.
The science journalist Ed Yong reintroduced the concept to a wide audience in his 2022 book An Immense World. A tick lives in a sensory bubble of three signals: the warmth of a mammalian body, the touch of hair, and the smell of butyric acid. Light, colour, sound. None of that is part of the tick's wonderful world. A dog smells time. An electric eel reads electrical fields invisible to us. A mantis shrimp sees colours we have no words for. They share the same physical environment with us. They inhabit completely different perceptual worlds.
Each being experiences its own Umwelt as if it were objective reality. Uexküll called this objective reality, were it knowable, the Umgebung. But each being only ever has access to its own Umwelt, its own slice. The lesson is that every act of perception is partial, by design.
Stephen Wolfram pushes this even further. In his 2023 essay Observer Theory, he argues that the nature of us as observers is critical even in determining the most fundamental laws we attribute to the universe. The Second Law of thermodynamics, our perception of time as flowing in one direction, our sense of continuous space. Wolfram argues these are not properties of an underlying reality independent of us. They emerge from the kind of observer we happen to be, an observer with computational boundedness, perceiving the world through a particular kind of cognitive bottleneck. If we were not computationally bounded, we could "perceive the whole of the future in one gulp" and we wouldn't need a notion of time at all.
If even the laws of nature look different to different kinds of observers, the question of how a human role-filler and an AI role-filler perceive the same organisation suddenly becomes very interesting.
A human role-filler in customer success, sitting on a call with a frustrated client, perceives a particular reality. The room they are in. The tone of voice that did not match the words. The pause before the complaint. The history they share with this customer. The weather. Their own body, registering tension or relaxation. Their taste, formed over years, telling them whether this is a relationship to repair or a relationship to release. That is one Umwelt, one observation of the organisation's reality.
An AI agent reading every customer interaction the company has ever had perceives an entirely different reality. Patterns across thousands of conversations. Statistical regularities in churn signals. Subtle correlations between phrasing and renewal. Cross-references to product issues filed three years ago. A semantic landscape no human could hold in mind. That is another Umwelt, another observation of the same organisation's reality.
Neither sees better. Both see differently. Each one's perception is shaped by what their kind of observer can attend to and what falls outside their bubble. The AI cannot feel the room. The human cannot hold the cross-organisational pattern. And, crucially, each one's Umwelt feels to them like the obvious, complete picture, in exactly the way Uexküll predicted.
Combining different perceptual worlds gives an organisation access to a wider slice of the reality it is trying to operate in. No single Umwelt is enough to perceive a complex world accurately. Multiple Umwelten, properly integrated, get closer to truth than any single one could.
This is not a poetic claim. It is how complex systems actually work. A team where every member has the same training, the same background, and the same instincts will share blind spots that none of them can see, because the blind spots are baked into the shape of their shared Umwelt. A team that brings together genuinely different perceptual styles gets a wider view of the same problem, and tends to make better decisions.
Add AI agents to that picture and the principle holds in a much stronger form. The difference between a human Umwelt and an AI agent's Umwelt is not a matter of background or training. It is structural. The agent perceives whole categories of organisational reality the human cannot, and misses entire dimensions the human catches without effort. Together, they triangulate.
This reframes what we are doing when we put humans and AI agents into the same governance structure. We are not building a better workforce. We are building a wider perceptual system for the organisation. The organisation becomes able to sense more of its own reality, because the role-fillers inside it see different parts of that reality.
The implication for organisational design is sharper than it looks. If different sensors give wider perception, then the worst thing you can do is collapse the differences. Every time we try to make agents behave more like humans, or expect humans to think more like agents, we narrow the perceptual diversity of the organisation and lose exactly what makes the combination valuable.
There is one specific dimension where the human Umwelt remains structurally aligned with what an organisation is trying to do, in a way no AI agent can match. That dimension is feeling where the value lies. What an organisation actually contributes in the human world. Whether it is good. Whether it matters. This is what is currently being named "taste".
Here is the structural argument, and I think it is airtight. An organisation's purpose, in the end, is to do something for human beings. The customer is human. The community is human. The team is human. The future generations affected by the organisation's choices are human. Purpose is service to humans-in-the-world.
Sensing whether something genuinely serves humans-in-the-world requires sharing that Umwelt. You cannot feel the weight of a decision on a person's life unless you have a body, a life, stakes, mortality, relationships, weather, a kid with a fever. An AI agent can map the possible outcomes, simulate the responses, and surface considerations no human would have thought of. It cannot feel which outcome actually belongs in the world the human will live in. That feeling is reserved for beings who share that world.
This is exactly why "taste" has become the live conversation it has. Paul Graham, in February 2026, put it like this: "In the AI age, taste will become even more important. When anyone can make anything, the big differentiator is what you choose to make." The Cloudflare CTO Dane Knecht added: "Building is easy now. Knowing what to build, and what not to, is the hard part." Sam Altman has been saying similar things in hiring contexts. The pattern is the same. As production becomes cheap, the locus of value shifts to discernment about what should be produced. And discernment, in the end, is feeling.
This is not a vague spiritual claim. It is a precise observation about Umwelten. AI agents can run extraordinary maps of possibility space. Humans, situated in the same world the purpose serves, are the only role-fillers who can feel which possibility actually fits. The mapping is enormous and necessary. The fitting is small, irreducible, and uniquely human.
In an organisation properly designed for both kinds of role-fillers, this is the part of the work that stays in human hands. Not because AI is bad at it, but because the question being answered is not the kind of question AI is structurally equipped to answer.
People practising self-organisation methodologies like Holacracy have a structural mechanism for collective sensing built in from the start: integrative decision making. When changes to the organisation's structure are proposed, the practice treats one thing as authoritative. Not majority. Not seniority. Not even consensus. The sensing of each role-filler from inside the role they energise. The objector alone determines whether their objection is real. The facilitator helps articulate, never overrules. Decades before AI agents arrived, the practice already encoded the principle that an organisation perceives reality through the role-fillers who are sensing on its behalf.
What changes in the agentic era is not the mechanism. It is the room. Both human and AI role-fillers can now raise objections from inside their roles. Because they inhabit fundamentally different Umwelten, what they each catch is different.
An AI agent with full organisational visibility might raise a structural objection no human in the room could plausibly catch. A new accountability that quietly conflicts with a policy in a circle three layers away. A human role-filler might object on grounds the AI cannot see. Something in the proposal that echoes a direction taken years ago that caused real harm, never properly documented, but still alive in the people who lived through it. Once surfaced, the integration might prompt the circle to make that knowledge explicit, so future role-fillers, human or AI, can read it.
Either objection, properly tested, evolves the proposal into something better than either Umwelt alone would have produced.
I want to be careful here. Governance meetings are only one site where humans feel into purpose, and a small one. Most felt sensing happens in everyday work. In the moment of writing a customer reply, in choosing what to build next, in the taste decision about whether something is ready to ship. The value of integrative decision making is not that it captures all of that. It is that it gives a structural channel for those deeper sensings, from any kind of role-filler, to actually evolve the organisation when it matters.
For the practical mechanics of this form of governance, The Governance Meeting goes through the full format.
I have been running a few experiments to see what a fully agentic organisation actually feels like in practice. For example a content website with a few products. In it, a content circle, all roles inside it filled by AI agents. The only thing I still do is approve final outputs before publishing. When I first set the circle up, I had to give a lot of input. Tone of voice. How a hand-off should work between roles. Which concepts matter and which to avoid. The early weeks were heavy with process and craft instructions.
Then something shifted. Once the agents had absorbed enough of the working context, the tensions they surface to me changed shape. After about three meetings, almost every blocking tension the circle lead brings to me is a question about purpose alignment. Not how to write something. Not how to format it. Whether it actually serves what we are here to do.
Two recent examples. During research for an article, one of the agents surfaced criticism online of a category of product I am trying to sell. They checked the critiques. Some turned out to be valid, some not. The agent included all of them in the draft. That is where the tension emerged. The role's accountability is to produce content that leads to more sales, and including valid critiques cuts against that. The circle lead surfaced it as a blocking question: should we publish? I made the call. We published. Honesty about the product space serves our purpose more than tilting the writing toward sales-friendly framing. In another project, a similar thing came as a check on a project definition. The circle had scoped an MVP that quietly cut away most of the breadth of the original purpose. The circle lead flagged the mismatch and asked whether the smaller version was still what we wanted. We rescoped.
What I notice is that this is exactly the work the article is about. The agents have absorbed the process, the craft, the structure. The questions they cannot answer for themselves, and that they correctly route to me, are about whether the work actually serves the purpose. Taste. Fit. Direction. I run all of this in Nestr with Claude, and at this point I do not look at most of what the circle is doing day to day. I just chat with the circle lead about the tensions that need me. Next week I expect to release the last few oversight checks I still hold. If you want to try this in your own organisation, there is a structured way to do it safely before going to full autonomy.
Define human roles around what humans actually perceive well from their Umwelt: situated stakes, social currents, the felt sense of fit, taste about where the value lies. Define agent roles around what their Umwelt picks up: pattern at scale, semantic structure, memory, tireless analysis. Connect both through a living governance layer that every role-filler, carbon-based or silicon-based, can read and contribute to.
The practical implication is structural, and it is where Nestr's work sits.
If you accept that humans and AI agents are different kinds of observers contributing to a wider organisational perception, the design question becomes straightforward. Every human role should be shaped around what a situated observer actually does well. Every agent role should be shaped around what a computationally unbounded observer actually does well. And both should be connected through a shared, living governance layer that each can read and contribute to. With humans, we’ve been doing this since the beginning of time, trying to get different types of personalities to the correct role where they can best serve the purpose. It’s just that now, the type of personality and perception has gotten a lot broader.
The mistake I see most often is organisations racing to deploy agents without this structural layer in place. The result is what we have been calling agent sprawl, and it is the single most common reason agentic projects fail. Agents with overlapping responsibilities, no shared context, and no visible governance trail will produce more volume and more confusion in equal measure. Gartner's prediction that over 40% of agentic AI projects will be cancelled by the end of 2027 is a direct consequence of this structural gap (Gartner, June 2025).
The mistake I see almost as often is the opposite one. Organisations that have the agents, have the governance, and still under-invest in the human sensing roles. They assume humans will keep "having opinions" and that those opinions will somehow make it into the structure. They will not. Sensing only becomes load-bearing when it has a structured channel. That channel, in my experience, is a living set of role definitions, regular tactical meetings where tensions can be raised by any role-filler, and governance meetings where the structure evolves through consent.
Without that structure, human sensing stays as gossip, venting, and private frustration. With it, sensing becomes the nervous system of a living organisation.
Our organisations can become more alive, not less, as more of their work is done by agents. The human-filled roles become more recognisable, more valuable, and more clearly essential. Every role-filler, human or AI, contributes a perception the organisation could not otherwise have.
I want to close with the piece that makes me genuinely hopeful about this moment.
It is easy to read the agentic shift as a subtraction. Agents will do what we used to do, and there will be less of us needed. And for organisations that treat humans as interchangeable units of labour, that subtraction is real. But for organisations that design roles around what living sensors actually perceive, the shift is an addition. The human role becomes more recognisable, more valuable, and more clearly essential than it has ever been. Not because we fought for our place, but because the place was always ours. It just took the arrival of a different kind of mind for the shape of ours to become visible.
The organisations I want to build, and the ones I want my children to work in, are the ones where being alive in the world the purpose serves is the qualification. Where your role exists because you perceive something the organisation could not otherwise reach, and where the structure around you takes what you sense and actually uses it.
That is the quiet promise underneath all the noise about agents. Our organisations can become more alive, not less. More sensing, not less. More distributed, not less. More purpose-driven, not less. But only if we build the sensing layer as deliberately as we build the agent layer, and design each kind of role-filler around the Umwelt they actually inhabit.
At Nestr, that is what we are building. A governance layer where human sensors and silicon synthesis sit inside the same living structure, each filling roles they are actually suited for, both connected to the same living purpose.
Work. Liberated.
.png){kind=logo}
