More AI. More People. Less Output.
The AI story sounds obvious: give people better tools, make them faster, add more AI, get more output. And yet many companies are seeing something stranger — more tools, more AI-generated work, more experiments, more internal activity, but no more shipped product, no faster decisions, no better customer outcomes. Sometimes, even less output than before. How can that be?
The Bottleneck Doesn’t Disappear. It Moves.
AI can make individuals faster while making the organisation slower.
A company is not a collection of isolated tasks; it is a system, and systems are constrained by bottlenecks. This is not a new idea — Eliyahu Goldratt described it in The Goal back in 1984. What is new is that AI creates the illusion that the bottleneck has been resolved, when in fact it has simply shifted.
A developer may write code twice as fast, but the release still waits for code review, QA, staging, and someone to actually press deploy. A marketer may generate ten campaign concepts before lunch, but a human still has to choose one, get sign-off, brief the designer, and coordinate with sales. A product manager may produce specifications at a pace that would have taken weeks before, but engineering capacity, strategic focus, and cross-functional alignment remain stubbornly finite. AI speeds up one part of the work, and the bottleneck simply moves to the next part.
The system’s output does not increase just because one node became faster. Often it gets worse — because that node now floods the downstream stages with more work than they can absorb.
Creation Has Become Cheap. Attention Has Not.
There is a second problem, and it may be more serious: AI makes creation cheap. Drafts, slides, analysis, code, ideas, documents, research summaries, strategy decks, meeting prep, performance reviews — all of it faster, easier, cheaper to produce. That feels productive, and it looks productive. The calendar is full, the folders are full, the Slack messages are full.
But more output is not more value.
Here is the asymmetry that most organisations are not yet accounting for: the cost of creating something has fallen dramatically, but the cost of reviewing, deciding on, and acting on something has not fallen at all. Attention is still scarce, judgment is still scarce, trust is still scarce. When creation was hard, that scarcity acted as a filter — not everything got written down, not every idea became a document, not every opinion became a proposal. The friction was real, but it served a function: it prioritised, so only the things worth the effort got made.
Remove that friction and you remove the filter.
The result is not better decisions; it is more decisions to make, with the same capacity to make them well.
The Old Operating Model Is Still Running
Many companies are making a subtle but consequential mistake: they are adding AI on top of the old operating model. Same meetings, same approval layers, same reporting lines, same unclear ownership, same habit of involving too many people in too many decisions, same instinct to socialise every idea before acting on it. Only now, everyone can generate more material to circulate through that system. This is not transformation; it is acceleration without direction. The pipes are the same, the water pressure is higher, and the leaks are more obvious.
Think about what happened when email replaced memos. Communication got faster. Response expectations got higher. The volume of messages exploded. But the organisational structures that determined who needed to be informed, who needed to approve, and who actually made decisions — those did not change. Most companies ended up with the same hierarchies, only now drowning in inboxes. The speed of the medium outpaced the organisation’s design. AI is doing the same thing, at a much larger scale, with a much more convincing appearance of productivity.
The organisations that figure this out fastest are the ones asking a different question. Not “how do we use AI to do what we already do?” but “what does the way we work assume that AI has now made unnecessary?”
More People Can Make This Worse
Here is where it becomes counterintuitive. Every additional person in an organisation creates more communication overhead — more meetings, more handoffs, more context that needs to be shared, more coordination, more decisions about who decides. This is not a failure of management; it is just how groups work. Fred Brooks captured it in 1975: adding people to a late software project makes it later. Now add AI, and each person can generate more material than before — more proposals, more documents, more feedback, more requests for review, more “can you take a look at this?” messages. Each individual is more capable, and the team becomes more congested.
The organisation becomes busier, but busier is not better, and faster individuals in a slow organisation are not the same as in a fast one. In some cases a smaller team — with AI tools and a much cleaner operating model — will outship a larger team still running the old way. Not because the larger team lacks capability, but because it is spending most of that capability coordinating itself.
AI Exposes What Was Always True
This is perhaps the most important thing AI is doing right now, and it is not getting enough attention: AI is not creating new organisational problems, it is making pre-existing ones visible and urgent. The bottleneck was often never individual productivity. It was decision-making speed. It was unclear ownership — the situation where everyone is involved and no one is accountable. It was too many approvals required before anything could move, weak product judgment — the tendency to rely on consensus rather than taste — and a lack of focus, the organisational inability to say no to things that sound reasonable.
These problems existed before AI. They were tolerable because the pace of work was slower, and AI removes that tolerance. When everyone can produce more, the structural inefficiencies that slowed everything down are no longer background noise — they become the loudest thing in the room. This is actually good news, if you are willing to look at it directly.
The Real Redesign
The companies that will benefit most from AI are not the ones that have deployed the most tools. They are the ones who have used AI as an occasion to redesign how they actually work.
What does that redesign look like in practice?
Fewer handoffs.
Every handoff is a place where context gets lost, speed decreases, and accountability diffuses. Most organisations do not know how many handoffs their core processes contain, because no one has ever counted them.
Start there. Take one high-frequency process — a product feature going from idea to launch, a sales proposal going from qualification to signature, a support issue going from report to resolution — and map every step. Count the handoffs. Then ask: which of these exist because the work genuinely requires a different person, and which exist because of how the organisation is structured?
Clearer ownership.
“We need alignment” is one of the most expensive sentences in business. It sounds responsible. It usually means ownership is unclear and no single person feels empowered to make decisions.
Pick a domain — pricing decisions, content approvals, hiring calls below a certain level, product scope changes within a sprint. Ask: who is actually accountable for the outcome here? Not who gets consulted, not who gets informed — who is accountable? If the answer is more than one person, the answer is effectively no one.
The redesign is not about removing collaboration. It is about being clear that one person carries the decision, consults whom they need to consult, and then moves. Everyone else trusts the outcome or escalates explicitly — they do not slow the process by default.
Smaller decision-making groups.
The research on group decision-making is consistent: past a certain size, groups do not make better decisions. They make more cautious, slower, more political ones. Each additional voice adds surface area for objection and delay, not quality.
In practice, this means auditing your standing meetings and recurring review processes. If a meeting has more than six people and a decision is supposed to come out of it, something is wrong. Either the decision does not need that many people — in which case, reduce the group — or it is not really a decision meeting, it is an alignment meeting, in which case, ask whether the alignment could happen asynchronously instead.
One useful test: if you removed three people from this meeting, would the quality of the decision actually suffer? In most cases, the honest answer is no.
Faster feedback loops.
Most organisations treat speed as a risk. Move faster, make more mistakes. The redesign inverts this: moving slowly is itself a risk, because it delays the information you need to make the next decision well.
The goal is not to be reckless. It is to get something real in front of real users or real conditions as fast as possible, then use what you learn. A marketing team that launches a rough campaign, sees the data in a week, and adjusts will outlearn a team that spends a month perfecting a campaign before launch. A product team that ships a limited release to ten customers in week two will know things that a team doing internal reviews for six weeks simply cannot know.
AI can help here directly — not by replacing the judgment, but by compressing the time between “we have an idea” and “we have something testable.” Use it for that.
More discipline about what not to do.
This is the hardest one, and the most important.
AI makes it easier to start things. A new initiative that would have required weeks of groundwork can now be prototyped in days. A report that would have taken a team a week can be drafted in hours. The activation energy for new work has fallen dramatically.
But finishing things, and finishing the right things, still requires the same focused attention as before. The discipline to stop a project that is no longer the priority, to say no to a request that sounds reasonable but is not, to remove a meeting that everyone attends but no one would miss — that discipline is now worth more than ever, because there are more things competing for the finite capacity of your organisation.
A practical starting point: every quarter, before asking what to add, ask what to stop. Not what to pause. What to actually stop, close, and remove from the list. If that question is harder than the question of what to add, your organisation is accumulating work faster than it is completing it. AI will make that problem worse, not better.
The Tools That Should Not Survive the Redesign
Here is the part that most AI conversations avoid. The standard narrative is additive: buy more tools, deploy more capabilities, build a bigger stack. But if the redesign is real — if you genuinely reduce handoffs, clarify ownership, shrink decision groups, and cut what is not working — then certain categories of tools become redundant. Not because they are bad tools. Because they were built to manage complexity that no longer exists.
Status and coordination layers.
Project management platforms are full of boards, cards, status fields, and update threads. Most of that infrastructure exists for one reason: because ownership is unclear, no one trusts that things are moving, and visibility has to be manufactured through process. When one person owns a thing and is accountable for it, the status is either done or not done. You do not need a system to track what a person already knows about their own work. If your project management tool is primarily used to answer the question “where does this stand?”, ask why that question keeps needing to be answered at all.
Approval workflow tools.
There is an entire category of software built around the assumption that content, decisions, and spend must pass through multiple sign-offs before anything moves. Contract approval chains. Content review platforms. Multi-stage request management systems. These tools are well-designed solutions to a real problem — but the problem they solve is diffuse ownership, not the nature of the work itself. Redesign the ownership model, and most of the workflow the tool was managing disappears with it.
Meeting infrastructure.
AI meeting summarisers, automated note-takers, transcript tools, scheduling assistants — all useful in a world where meetings are a fixed constraint. But if the redesign reduces the number of meetings, particularly the alignment meetings that exist because no one is empowered to decide without them, then the tools built around those meetings lose their purpose. The right response to a bad meeting is not a better summary of it.
Reporting and visibility dashboards.
Management dashboards often exist because leaders lack direct visibility into the work, and teams are too large and layered to provide it. The dashboard is a proxy for trust and proximity. Smaller teams, clearer ownership, and faster feedback loops reduce the need for dashboards that tell you what is happening — because you are close enough to already know. This does not mean metrics are wrong. It means that many reporting layers are a symptom of organisational distance rather than a management best practice.
AI tools built on top of broken processes.
This is the most uncomfortable one. A significant portion of the AI tools being sold right now are automating steps in processes that a well-designed organisation would not have in the first place: AI that summarises the output of a meeting that should not have happened, AI that drafts the status update that exists because ownership is unclear, AI that speeds up an approval workflow that could be handled by one person with authority.
Automating a broken process does not fix it. It makes it faster, cheaper, and more permanent.
The question to ask of every tool in your stack is not “does this work well?” It is “does the problem this tool solves still exist after we have redesigned how we work?” If the answer is no, the tool should go — regardless of how much was spent on it, how long it has been in use, or how many people have built habits around it.
Fewer tools, used with intent, in a system designed to move, will always outperform more tools layered on top of an organisation that was not designed to move at all.
The Question Worth Asking
The key question is not “how can we add AI to this workflow?” It is: if we were designing this workflow today, with AI available from the start, would we build it this way at all? The answer, in most cases, is no. Most workflows were designed around the constraints of a world without AI — the cost of writing things down, searching for information, drafting, and first-pass analysis. Many of those constraints are gone, but the workflows remain, because workflows are sticky, because people are comfortable, and because redesigning how work gets done is genuinely difficult and disruptive. The companies willing to do that work — to start from a blank sheet and ask what the process should be, not what it has been — are the ones that will actually capture the value that AI makes possible.
The Uncomfortable Conclusion
The competitive advantage will not belong to the companies with the most AI tools, the largest AI budgets, or the most AI mentions in board presentations. It will belong to the companies that can turn additional capability into actual progress. That requires something AI cannot provide: organisational clarity, a shared understanding of what matters, the discipline to focus, and the courage to remove things — layers, approvals, roles, meetings, processes — that feel safe but slow everything down.
Otherwise, the outcome is predictable. More people. More AI. More activity. Less output.