Back to blog
An AI-augmented organisation is one where people own judgement and the customer relationship, and a set of AI agents and assistants carries the volume work around them, inside boundaries the organisation has written down. The design question is not how many tasks AI can take over. It is who stays in charge of judgement once it does, and AI augmented organisation design is the work of answering that on purpose instead of by drift.
What it looks like in practice
We run our own firm this way, which is why we describe it from the inside. Two senior partners, Anton and Joachim, own every engagement from start to finish. Around them sits an AI workforce, and it extends our reach without replacing our judgement. In outline it has four kinds of role. Analysts read and structure qualitative and quantitative customer material and compare competitors. A drafting role turns what customers said into a job to be done and a value proposition. Content and outreach roles prepare marketing and sales material. And a verifier checks analysis before it goes anywhere. And in the end, nothing reach our customers without a thorough review by us.
Each of them is grounded in our own models and frameworks, including our book Single-Minded: Business Design Built around the Customer, and not in whatever a general-purpose tool happens to assume. Some work as agents that run a defined job and hand back a result. Others work as assistants that help a person in the moment. What they share is a human who owns the outcome.
Four design choices that carry the model
Judgement keeps a named owner. For a product company this means a named person for each process that AI touches, who answers for what goes out and stays accountable when something is wrong. Without that person, AI output belongs to nobody, which means nobody corrects it. The owner's check against the written boundary is described in Using AI Without Diluting Your Brand.
Each role has one job. A role is defined as plainly as a job description: what it receives, what it produces and who checks it. A role called "assistant" with no boundaries is where the confusion starts.
Output is checked where work crosses a boundary. The verifier exists because one AI's output should not go straight into another AI's input. A check at each hand-off is cheaper than finding the error downstream.
The foundation is shared. When each person defaults to their own model, tool and way of working, silos form around the tools. A shared set of assistants, all reading from one written source of truth, closes those silos. How to write that foundation is the subject of Source of Truth: Where to Play, How to Win for AI.
Where the model strains
If a few senior people answer for every output, review becomes the bottleneck, and the volume that AI makes cheap to produce can arrive faster than anyone can check it. A design that does not decide how much output one person can responsibly own trades one backlog for another. This is a risk to design for, not a report from our own experience.
The scarce resource in an AI-augmented organisation is not output. It is attention that can be trusted to check it.
That is also why the model is not a way to remove human capacity. It depends on people doing more of the work that only they can do: the judgement calls, the customer conversations and the decisions about what goes out. The AI workforce takes the volume that used to compete with that work for their time, and the ratio between the two has to be designed, not hoped for.
How to start
Take one flow in your business that people already complain about, such as prospect research, proposal drafts or the handover from sales to delivery. List its tasks and mark each as work AI can carry or work a person must keep. Name one owner for the whole flow, and add one check at the point where work crosses from one stage to the next. Then decide how many outputs that owner can genuinely review in a week. That page is the first version of your design.
Key takeaways
An AI-augmented organisation puts people in charge of judgement and the customer relationship, with AI agents and assistants carrying the volume work inside written boundaries.
Every process AI touches needs a named human owner who answers for what goes out. Without one, AI output belongs to nobody.
Give each AI role one defined job: what it receives, what it produces and who checks it. Place a verification step where work crosses from one stage to the next.
Ground every assistant in your own models and frameworks and in one shared source of truth, not in whatever a general tool assumes.
Ownership concentrates, and review can become the bottleneck. Decide how much output one person can responsibly own before volume arrives.
FAQ
What is an AI-augmented organisation? It is an organisation where people own judgement and the customer relationship, and AI agents and assistants carry the volume work around them, inside boundaries the organisation has written down. The aim is to extend reach, not to replace judgement.
What is the difference between an AI agent and an AI assistant? An agent runs a defined job and returns a result. An assistant works alongside a person in the moment. Both need a human who owns the outcome.
Does an AI-augmented organisation need fewer people? No. The design depends on people spending more time on judgement, customer conversations and decisions, and less on volume work. The AI workforce is meant to free that time.
Where should a company start designing one? With a single flow that people already find painful. List its tasks, name one owner for the flow, add a check where work moves between stages and set how much output the owner can review.
