AI Won’t Fix Your Organisation. It Will Show You What’s Broken.
And that’s the part nobody is advertising. The gap is almost always an AI organisational design problem — and it was visible long before the AI arrived.
In 2012, my team built a robot.
I was the vision. They built the thing. Not the kind that walks — the kind that stands in a hotel lobby, watches you approach through an Xbox Kinect sensor, recognises your presence before you’ve said a word, and asks how it can help. We called it HAL — Hotel Assistant Liaison — a tongue-in-cheek nod to Kubrick’s homicidal spaceship computer that felt entirely appropriate. The product we demoed was branded AL. We’d dropped the H. Whether that was branding pragmatism or a subconscious act of mercy, I’ll leave open.
Watch the original demo from 2012 →
HAL was a patchwork of what was available at the time: Kinect for depth sensing and skeletal tracking, Google’s early cloud services for speech recognition and synthesis, and a shallow neural network called Skynet — built by a developer in California I’d found on a forum whilst hunting for something that could serve as HAL’s brain — for intent mapping. Before transformers, before large language models, before AI had a press office, you built intelligence by daisy-chaining services together and hoping the seams didn’t show.
The technology worked. HAL could hold a conversation. It could answer questions about breakfast hours and WiFi passwords and local attractions. It could free up the human concierge for the things that actually required a human: the difficult conversation, the upset guest, the moment where presence mattered more than information retrieval.
The startup ran out of money before we got to market. And if I’m honest, my head was somewhere else — my family and community were facing an existential threat that came from being a minority in a society that had decided we were a problem to be managed. I was trying to hold something together on the outside whilst building something on the inside that was supposed to help people see more clearly.
HAL never saw a hotel lobby.
But the question it was built around has stayed with me for thirteen years, and it’s the question I think most organisations are about to be forced to answer — whether they want to or not.
What HAL Was Actually Doing
The framing we used at the time was augmentation. HAL would handle the repetitive 80% so the human concierge could focus on the complex 20%. That was the pitch, and it was a good pitch.
But looking back from 2026, I think we undersold what was actually happening.
HAL wasn’t just answering questions. HAL was measuring the gap between what guests needed and what the organisation had designed itself to deliver. Every time a guest walked up to the kiosk and asked where the gym was, that was a signal. Every time someone asked what time checkout was, that was a signal. The questions HAL fielded weren’t random — they were a precise map of where the hotel’s information architecture had failed.
The concierge already knew this. They answered the same questions forty times a day. But knowing something and having it measured, timestamped, and visible are different things. The first can be dismissed. The second cannot.
This is what I mean when I say AI is the most powerful diagnostic tool, It is the most powerful tool we have for AI organisational design — not because it solves problems, but because it makes existing ones impossible to ignore.
Not because it solves problems. Because it makes the existing ones impossible to ignore. Redundancy is the clearest case, and I wrote about what leaders do once AI makes it visible.
The Pattern I Keep Seeing
In the years since HAL, I have worked across contexts that look completely different on the surface: pharmaceutical teams in Europe, FMCG sales teams navigating post-conflict Lebanon, agronomy teams in Malawi, consumer goods operations across eleven markets in the Middle East, East Africa and Turkey. Oncology teams in the US. Drug development in Germany. Telecoms in the Netherlands. Healthcare in the UK. The sectors change. The dysfunction doesn’t.
In pharma: A team I worked with built a machine learning model to predict physician behaviour four months in advance. Clean data, solid architecture, a team that cared deeply. The model worked. And then nothing happened fast enough. Predictions landed and sat in a queue between data science, marketing, and sales — each team with a slightly different definition of “actionable,” each with a slightly different sense of who owned the next step. The AI was ready. The organisation wasn’t.
We spent three months tuning the algorithm. We should have spent the first month on AI organisational design — specifically, redesigning the handoff.
In Lebanon: A sales team lost 226 outlets during the conflict. They recovered — and then grew, expanding coverage by 45% — not because they suddenly had better data or more sophisticated tools. They recovered because the crisis stripped away the governance layers that normally slowed their response cycle to a crawl. They stopped waiting for approval. They adapted daily. The PDCA loop ran on instinct because there was no time for anything else.
When I came in afterwards and looked at what had changed, it was a masterclass in AI organisational design by necessity — clarity of ownership, speed of decision, visibility of accountability. The most striking thing was how visible the decision ownership had become. Everyone knew what they were responsible for. Everyone knew how fast.
In Malawi: The agronomy team ran a pulse survey after a lean meeting reset. The data showed something precise and uncomfortable: people knew what the priorities were — that metric scored 3.67 out of 5. But they were less clear on who was accountable for acting on them — decision clarity sat at 3.58. The gap between knowing what matters and knowing who does what by when is smaller than it sounds. In practice, it is where most organisational friction lives.
In Nigeria: After fifteen months of building agile habits across sales, marketing, and activation, the pattern that emerged wasn’t about speed — it was about signal flow. Insights started moving across functions instead of sitting inside them. The new SKU launch succeeded not because the team moved faster in isolation, but because the walls between insight, decision, and action got thinner. The friction became visible. Then it became workable.
What Actually Gets Diagnosed
When you introduce an AI agent into an organisation, three AI organisational design gaps become measurable that were previously invisible.
Decision latency. The agent logs a recommendation at 9am. The action is taken the following Thursday. That gap existed before the AI arrived. But now it has a timestamp. You can no longer claim the process is moving at a reasonable pace, because the process is now on the record. Latency, it turns out, is rarely about speed. It is almost always about unclear ownership at the point where a decision needs to be made. Nobody is slow on purpose. They’re slow because nobody is sure it’s their call. This is an AI organisational design question, not a performance one.
Accountability fragmentation. In every organisation I have worked in, there is a version of the same conversation: the one where three people in a room each believe someone else is responsible for a particular decision. The AI surfaces this structurally. When an agent produces an output and nobody picks it up, you have not found a lazy team. You have found an accountability gap. The agent did not create it. It just made it countable.
Process ownership gaps. These are subtler. They are the places where work moves between functions — from strategy to execution, from insight to action, from recommendation to implementation — and nobody has formally designed the handoff. The process works when people who happen to know each other manage it informally. It fails when those people leave, or the organisation scales, or the pace increases beyond what informal coordination can absorb. AI operates at a pace that informal coordination cannot absorb. So it finds these gaps automatically, because it stops at them.
Why Organisations Won’t Look
OpenAI’s announcement of its Frontier Alliances with McKinsey, BCG, Accenture, and Capgemini contains one line that deserves far more attention than it received. The limiting factor for enterprise AI value, OpenAI said, is not model intelligence. It is how agents are built and run inside organisations.
This is a significant thing for a technology company to admit. It is also, I suspect, more commercially convenient than it might appear.
Because naming the problem as how agents are built and run keeps the solution in the domain of the technology firms and consulting firms. It frames the challenge as one of implementation and integration. AI organisational design is not an implementation question. It is a structural one. It does not name the harder thing, which is that what AI will surface — in most large organisations — is a set of structural dysfunctions that have been deliberately, if unconsciously, maintained.
Accountability gaps don’t persist by accident. They persist because clarity of ownership is also clarity of consequences. When nobody owns a decision, nobody fails. When a handoff is informal, it is also unauditable. When the process is murky, the murk protects people. Not maliciously. Just humanly.
This is the political economy of self-knowledge in large organisations: the very structures that AI will expose are often the structures that make institutional life manageable for the people inside them. Asking an organisation to look clearly at itself is not a technical request. It is a political one. It asks people to become visible in ways that carry risk.
Most organisations will respond to AI’s diagnostic function by optimising around what it reveals rather than redesigning in response to it. They will create new approval layers to manage the agent’s outputs. They will build governance frameworks for AI recommendations that mirror, exactly, the governance frameworks that slowed down human recommendations. They will solve the agent problem without touching the organisation problem.
This is not cynicism. It is pattern recognition. I have watched it happen with agile transformations, with digital transformation programmes, with data initiatives. The technology arrives. The dysfunction absorbs it. The organisation continues at the speed of its decision systems. AI organisational design is what gets skipped. It is also what determines whether any of the rest of it works.
What the Honest Version of This Looks Like
The enterprises that capture real value from AI will not necessarily be the ones with the most sophisticated models. They will be the ones willing to do something harder: to treat what the AI surfaces not as an implementation problem but as a design brief.
This is what AI organisational design actually asks of you: not better tooling, but clearer architecture. It is a question about where authority actually lives in your organisation, and whether the people who hold it know they hold it.
Accountability fragmentation is not a communication issue. It is a question about whether your organisational design has kept pace with your operational complexity.
Process ownership gaps are not handoff issues. They are questions about whether you have designed your organisation for the world you operate in, or the world you used to operate in.
These are not comfortable questions. They are not the questions that get funded in an AI transformation programme. But they are the core of what AI organisational design requires you to confront. They are not the questions that get answered in a slide deck about agent deployment.
But they are the questions that HAL was pointing at in 2012, in a hotel lobby it never reached.
And they are the questions that every AI agent being deployed inside every enterprise right now is, whether anyone has planned for it or not, already beginning to ask.
AI may move at machine speed.
Most organisations will still move at the speed of their decision systems.
The question is whether they have the courage to look at what the gap reveals — and the will to redesign, rather than accommodate.
Most will find reasons not to look.
The ones that do will not just be better at AI. They will have done the harder work of AI organisational design – and they will be better organisations for it.