The Value of Friction

For decades, organisations have treated friction as a cost to be eliminated. AI may force us to ask which friction was creating value all along.

There is a challenge with the way vast amounts of organisations are approaching AI, and it is not what you might think.

The focus tends to be on capability: what AI can do, how quickly organisations can integrate it, what it costs, what it replaces, but there is an assumption behind this thinking. The assumption is that friction is the enemy. The measure of a well-run organisation is how smoothly its workflows run. That efficiency and effectiveness are, for practical purposes, the same thing.

AI is about to test that assumption in ways that vast swathes of organisations are not prepared for.

Two ways of thinking about process

Consider a professional who has spent twenty years in project management. Her world is built on a coherent set of principles: good process is repeatable, efficient, standardised, and low-friction. Friction means delay. Delay means cost. The goal of process improvement is to remove the obstacles that prevent work from flowing. She is not wrong. In most of the environments she has worked in, that logic has held.

Now consider an organisation that builds AI into its creative process, not to accelerate output but to introduce deliberate challenge. Where agents are designed to disagree with each other. Where a brief is not accepted until it has survived structured scrutiny. Where the point of the process is not to produce answers faster but to ensure that the answers produced are worth having.

Both of these represent coherent thinking about process. They are not compatible.

The first is viewing friction through an execution lens: friction is inefficiency. The second views friction through a decision-quality lens: friction is a mechanism that drives thinking. Neither definition is wrong. But in a world where AI changes the economics of cognition, only one of those orientations is likely to produce durable competitive advantage.

Intelligence is no longer the bottleneck

For most of the history of organised work, the constraint has been accessing information, expertise, capacity, and synthesis. As a result, processes were designed, in the main, to adjust for that scarcity in order to move the right knowledge to the right people at the right moment. Friction was often the symptom of those structural limits. Remove the bottleneck, reduce the friction, accelerate the flow.

AI changes that logic because options are now abundant. Content is abundant. Research is abundant. The things that have not kept pace are decision-making and judgement. When you flood a seemingly strong process with speed, you do not improve the outcome if a flaw has been managed. You either arrive at the wrong place faster, as the weakest point cannot keep up, or you hit a structural dam.

This is the displacement that most AI conversations miss. Organisations are racing to reduce the friction in their processes at precisely the moment when some of that friction was holding the process together. The question is not how to make the process faster. The question is what the process was for.

The human in the loop who has stopped thinking

The standard reassurance runs something like this: there will always be a human in the loop. AI generates, humans review, and decisions remain with people. The process still has oversight.

That may be technically true. But it contains a hidden assumption: that human judgement remains constant while AI accelerates everything around it.

Human beings are extraordinarily effective at offloading cognitive effort. The brain is, at its core, an efficiency engine - always seeking the path of least resistance, always looking to conserve resources. When a machine produces a credible answer in three seconds, most people do not use the time saved to think harder. They move to the next task. This is not a character flaw. It is how cognition works. It is also how people manage overload: they tick tasks off a list.

The consequence is that organisations may be systematically eliminating the moments where judgement was actually being exercised. The answer arrives faster. The review becomes lighter. The challenge becomes weaker. The process becomes smoother, all whilst the quality of thinking deteriorates.

There is a human in the loop. The human has stopped thinking.

The question that matters is not: is there a human in the loop? It is: is the human still thinking?

Not all friction is waste

Traditionally, people encountered friction throughout a process. A proposal required discussion. A tender response required collaboration across perspectives. A report required interpretation before it became a recommendation. These moments were not always efficient. But they created conditions in which people had to think. Some processes helped to generate judgement, others did not; the computer says no springs to mind.

It is worth being honest about this. Many organisations never had meaningful judgement-generating friction in the first place. Much of what passed for healthy challenge was bureaucracy, approval theatre, and decision avoidance dressed as due process. A sceptical reader might argue that AI is not removing valuable friction; it is simply exposing how little value much of the friction ever created. That challenge is fair. It does not, however, dissolve the problem. It sharpens it. If organisations could not reliably distinguish productive friction from wasteful friction before AI arrived, they are even less likely to do so when speed has been added to the system.

Not all of that friction was valuable. Some of it was administrative overload, repetitive coordination, cognitive clutter dressed as productivity. AI can remove enormous amounts of that kind of friction, and that is good design. The reduction of low-value friction frees human capacity for the thinking that matters.

The danger emerges when organisations treat all friction as the same kind of thing.

Some friction is waste. Some friction is quality control. Some friction is learning. Some friction is in making a choice. Some friction is judgement. They are not the same and eliminating them all in the name of efficiency produces the same result: an organisation that moves faster and understands less.

It is also worth noting the opposite failure. Friction does not automatically produce thinking. In large organisations, too much of the wrong kind of friction produces a different chain entirely: friction, then fatigue, then compliance. People stop questioning not because the answer is obvious but because the cost of questioning has become too high. Sometimes thinking happens because enough clutter has been removed to finally focus. The goal is not friction for its own sake. It is friction that is purposeful, placed, and owned.

Consider a production manager who had worked in a print business for decades. When he retired, the operation looked the same on paper - the same equipment, the same process, the same workflow. What it had lost was invisible until it started to wobble. The knowledge that held the process together had never been written down because it had never needed to be. It lived in his read of a job before it ran, his instinct about what would cause a problem three stages downstream, his willingness to slow a job down when something felt wrong. The process was seamless. The judgement was his.

That kind of friction - the pause before proceeding, the experienced eye that checks before signing off - does not appear in a process map. It rarely appears in a handover document. It tends to appear only in its absence.

The same principle holds in organisations that design it deliberately. A manufacturing operation that tests every batch before it leaves the building, that uses output signals not to drive throughput but to ask why something has changed, that rejects product rather than compromise on standard - and whose lines rarely stop for major repair, because the ongoing reading of what the system is telling them means problems are caught before they become failures. Nobody in that organisation is arguing that testing slows them down. The friction is the quality. Remove it, and you do not have a faster process. You have a different product.

Professional services are an instructive case. A tax return can be processed. Tax advice requires judgement. A legal document can be processed. Legal counsel requires judgement. A marketing campaign can be produced at scale. Brand thinking requires judgement. The moment organisations stop distinguishing between the two - and AI makes that distinction dangerously easy to collapse - they begin producing outputs without outcomes. Stuff that fulfils a deemed requirement without knowing if it is the optimal solution.

Momentum is not the same as progress

Most project management systems rest on a single unstated premise: progress is movement. Tasks are created to be completed. Completion creates momentum. Momentum generates the perception of forward motion. The RAG statuses turn green.

But the assumption underneath all of it is that we always know where we are going.

In almost any complex piece of work, that assumption becomes questionable. You can complete every task in a project plan and still arrive at the wrong destination. Execution-centric systems are designed to measure completion. Learning-centric systems are designed to measure understanding. Both matter, but they are not interchangeable, and organisations that run only one tend to become very efficient at implementing yesterday's thinking.

AI accelerates this risk. Historically, the effort required to execute created natural pauses, moments when teams would look up to ask whether the direction still made sense. Reduce the cost of execution dramatically, and those pauses disappear. The organisation becomes extraordinarily good at moving. The question of where it is moving to becomes harder to ask, let alone answer.

Organisations have become very good at measuring movement. They are surprisingly poor at measuring learning and improving movement, even many who practise lean techniques, and in an era of abundant execution, that gap is the strategic risk.

Designing for judgement

The defining capability of modern organisations is no longer access to intelligence. AI has made intelligence abundant. The capability that is becoming scarce (and as a result valuable) is the ability to convert intelligence into judgement, and judgement into value.

That conversion does not happen automatically. It requires deliberate design. Which voices are present in a decision? Where is the challenge introduced? What assumptions have to survive scrutiny before anything moves forward? What friction has been deliberately built into the process - not to slow it down, but to ensure that speed does not outrun understanding and relevance?

There is, though, another risk to consider. Challenge without consequence tends to become intellectual entertainment. If nobody in the process feels the cost of being wrong - if the decision belongs to the system rather than to a person - then friction becomes theatre. Genuine judgement requires ownership. It requires someone who will still be in the room when the outcome arrives.

These are Decision Architecture questions. And most AI training programmes do not ask them. They focus on prompting, tool selection, data governance, adoption frameworks - the mechanics of getting AI into the workflow. Much less attention goes to where humans must remain genuinely in charge, what cognitive safeguards are required, and which friction should be preserved rather than removed.

The organisations already asking these questions are building something different. Not faster processes with AI bolted on, but systems in which challenge is structurally embedded - where the brief does not get accepted until it has been genuinely tested, where agents, both human and AI, are designed to uncover competing perspectives rather than produce consensus, where the goal is not to generate an answer but to earn one.

Bad thinking plus AI is not better thinking. It is faster bad thinking.

The future advantage may not belong to the organisations with the most capable AI. It may belong to those that build the most effective systems of challenge, judgement, and discernment around it.

For decades, organisations optimised for execution. AI will make execution cheaper than it has ever been. Which means the competitive advantage shifts. To the ability to pause, to challenge, to exercise judgement before the next task gets completed. The organisations that thrive will not be those that remove all friction. They will be those that understood which friction was creating value all along.

If you want to understand how decisions are being made in your organisation — and where the judgement has quietly been designed out — the Decision Architecture Review is a good place to start.