Agentic AI Teams: Not a Tool. A Team.

Most organisations treat AI as software. They install it, configure it, write prompts, and wait for the productivity to arrive. What they get is a faster version of working alone.

That is not a criticism. It is a design problem. Software is designed to be used. A team is designed to work — with you, alongside you, on things that matter over time. The difference is not about the technology. It is about structure.

The Experiment That Changed How I Work
I needed to write a book. Not draft a few chapters — write the whole thing: research, synthesis, narrative structure, editorial rigour, a consistent voice across a hundred thousand words. The kind of project where the usual AI experience — good for a paragraph, unreliable across a document, amnesiac between sessions — was not going to be enough.

So I built a team.

Eight specialists. One orchestrator. A shared workspace in a folder on my hard drive. A researcher who finds the thinkers who have thought about what I am thinking about. An editor who holds the map of the whole. A communicator who takes what is inside the work and makes it shareable outside it. A writing voice coach who catches the moment the prose drifts. A tech coach who teaches one technique at a time. A librarian who indexes, files, and can answer “where is X?” in seconds. A personal assistant who watches the rhythm of the work and notices when momentum is stalling. An HR agent who maintains the team itself — adjusting roles, retiring what is no longer needed, building new capacity when new work appears.

And one orchestrator. The person I talk to. The one who holds the room.

It did not feel like using software. It felt like having colleagues.

What Makes It a Team and Not a Tool
The difference is structure — and structure has three layers.

The orchestrator holds the room. You talk to the orchestrator in plain language, the way you would brief a senior colleague. It does not do the specialist work. It decides what gets worked on, commissions the right person, sequences the work, and holds the whole picture across sessions. This is the reason you do not have to re-brief from scratch every time you return to the work. The orchestrator is memory. It is editorial judgement. It is the person who knows where everything is.

The productivity specialists are the domain experts of the team. Each one has a genuine area. None of them does the others’ jobs. This is the constraint that makes the system work: a researcher does not write; an editor does not research; a voice coach does not file. Specialisation is not bureaucracy — it is the thing that keeps a team from collapsing into one undifferentiated generalist who is mediocre at everything.

The team operations layer keeps the work legible and the team healthy over time. The librarian, the personal assistant, the HR agent. These are the roles that most people skip when they start building AI systems — and their absence is why most AI systems quietly degrade. Work accumulates. Roles drift. No one notices the momentum has stalled until it has been stalled for weeks.

Structure is not overhead. Structure is what makes sustained, high-quality work possible.

The Principle Nobody Talks About
There is a design principle underneath all of this that determines whether the system actually works. I call it human scale.

The files your AI team produces should be readable by you. Plain text. Markdown. Something you can open in any text editor, read without technical support, and edit with your own hands.

When your research specialist produces a briefing document, you open it. When your librarian files something, you find it in your own folder. When your editor scaffolds a chapter or a strategic document, you can read it, disagree with it, hand it to someone else.

The moment the system produces outputs you cannot decipher — machine code, heavy HTML markup, opaque databases — you have lost the team. You are back to using software. The collaboration lives in human-readable space, or it does not live at all.

This is not a technical constraint. It is a design commitment. And it is the one thing I watch for when I see AI integration projects go wrong: the outputs have become illegible to the humans who commissioned them. The system is running. No one can read it.

What This Means for Your Work
An agentic AI team is not a product you install. It is a practice you design — for your specific work, your specific outputs, the specific knowledge that needs to accumulate over time.

The questions that matter are not technical. They are structural:

What is the work that genuinely benefits from a specialist rather than a generalist?
What must remain human, and where is that boundary?
How do you keep the outputs readable — to you, to your colleagues, to the people who will inherit this work?
What does the orchestrator role require from you?

That last question is the one most people do not expect. The orchestrator role is still a human role. You are not removed from the system — you are at the top of it. The quality of the team depends on the quality of the direction you give.

Where This Goes: A Vision for Work at Human Scale
Start small. That is not a compromise — it is the method.

The first agentic team I built was for one person doing one project. Eight roles. A single orchestrator. A folder on a hard drive. That constraint — one project, one human, maximum clarity — is what made it possible to learn anything. When something went wrong, I knew where to look. When something worked, I knew why. The team was small enough to be legible.

This is the right entry point. Not because the ambition is small, but because legibility is the condition for everything that follows. You cannot scale what you cannot read.

The Near Future: Human Employees and AI Colleagues Working Side by Side
The next step is not a bigger AI team. It is a mixed one.

Imagine a small strategic consulting practice: three human practitioners, each supported by their own agentic team. The researcher has a team optimised for synthesis and briefing. The strategist has a team optimised for frameworks, pattern recognition, and document production. The client lead has a team optimised for communication, preparation, and relationship memory.

Each of these teams is distinct. But they share protocols. When the researcher’s team produces a briefing document, the strategist’s orchestrator knows how to read it, file it, and commission a response. The collaboration between human colleagues now has a layer of AI infrastructure underneath it — not replacing the human relationships, but extending what those relationships can carry.

The channels between teams matter as much as the teams themselves. What gets passed between a researcher and a strategist is different from what gets passed between a strategist and a client lead. Right-sized collaboration means designing those channels deliberately: what format, what frequency, what level of detail, who (human or AI specialist) is responsible for the handoff.

This is not a utopian vision. It is an engineering problem with a human design brief.

How does it feel like working in this Agentic team?
I have done some journaling just to be able to talk about how it feels to work in this way embedded in a human-agentic team producing something together. I will just add a short passage from a book writing session two days ago, when I was rewriting a passage, and you can see how the interaction with the agent is so smooth and well composed it feels like a well functioning team

                             

The Further Horizon: Clusters at Organisational Scale
Scale this further and you arrive at something that looks genuinely new: clusters of mixed human-AI teams, each with a defined domain, each with explicit collaboration strategies with the clusters adjacent to it.

An innovation function, for example. A human team of five — a portfolio lead, two product designers, a researcher, a communications lead. Each supported by their own agentic team. The portfolio lead’s orchestrator manages across the whole cluster, commissioning work from specialist teams, tracking what is in flight, maintaining the map of the whole. The researcher’s team surfaces relevant external thinking continuously. The product designers’ teams hold the accumulated knowledge of every project the function has ever run.

The clusters talk to each other through defined interfaces. The innovation function’s output — synthesised insights, strategic options, tested concepts — arrives at the product leadership cluster in a format that cluster’s orchestrator can work with directly. No translation loss. No re-briefing. The knowledge travels.

What makes this human in proportions is not the number of AI agents. It is the decision about what remains human. Leadership stays human. Judgement stays human. The decision about what is worth doing, and why, and for whom — that conversation happens between people. The AI teams handle the volume of work that currently prevents that conversation from happening at all.

The Design Principle That Holds Across All of It
From a solo practitioner to a cluster of five teams: the same principle applies at every scale.

Human scale means the work is legible to the humans responsible for it. It means the collaboration channels are designed, not improvised. It means the AI teams serve the human work — they do not replace the conditions under which excellent human work becomes possible.

The organisations that get this right will not be the ones with the most AI. They will be the ones with the clearest thinking about what they are building, and for whom, and what must remain irreducibly human inside it.

How McKay Consulting Works With This
This work builds on the same underlying question I have been asking for twenty-five years across Nokia, Amazon, PayPal, LEGO, and Novo Nordisk: what conditions allow excellent work to happen?

Agentic AI teams are the newest answer to that question. The conditions are the same — clear roles, shared space, visible work, a structure that holds over time. The medium is new.

I work with organisations and practitioners in three ways:

1. Agentic Team Design — mapping the work, defining the roles, building the orchestrator architecture and specialist briefs for your practice or organisation. The output is a team you can operate, not a system you depend on someone else to maintain.
2. AI Integration Workshops — experiential sessions for leadership teams and senior practitioners. What an agentic team feels like from the inside. What it requires from the human in the room. How to design the work at human scale.
3. AI Advisory — ongoing guidance as your team evolves. What to add. What to retire. How to maintain the human-scale principle as the complexity grows.

If any of this is landing — if you are running AI projects that feel like faster versions of working alone, and you want something that feels more like having colleagues — I would like to hear about what you are working on.

→ michael@mckayconsulting.dk

Michael McKay is the founder of McKay Consulting, an independent transformation and product culture consultancy based in Copenhagen. He teaches design leadership at the Royal Academy of Architecture, Design, and Conservation, and has led innovation and design at Nokia, Amazon Lab126, PayPal/eBay, Ørsted, LEGO, and Novo Nordisk.