Performance is one of those words that is being captured and put to work in areas where it does not really belong. The dawn chorus is a performance. Keith Jarrett’s The Köln Concert was a performance. Thames Water’s half-year results were not.

Performance evokes something visceral for us, experienced in the moment. It can be anticipated, but becomes something altogether thinner when it is recorded, and I think it dies when analysed.

Performance is an act carried through to completion, live, in front of others, with the outcome at risk until the end. In the business sense it is mostly a result measured after the act, against a target. And, there is a third sense, a calculated display for an audience, that sits between them and has become a term of suspicion: “performative”.

Artisans perform. They cannot do otherwise.

The experience that triggered this thought arose in an unexpected place, like all the best ones do. It’s autumn, and every autumn we have the log burner serviced before putting it to work in the winter. The same person, Lee, from the same company, has serviced it for well over a decade, and to watch him work is to witness a performance.

It takes a couple of hours. He completely disassembles the stove, lays it out in a precise pattern, cleans every single part of it, inspects it, puts it back together, and tests it before tidying up and leaving. You would never know that he has been, other than that we have a wood-burning stove that looks as though it’s just come out of the showroom. it is a quiet. wonderful performance.

I see individuals and teams who put on performances in the best sense of the word. Some of them are part of a business, but it’s not the business doing the performing.

And the thought that leads to is that AI will never perform. It has no sense of its audience or joy in what it is doing. That does not stop it being an enormously valuable tool. .

Can it enable performance in a person or a team? I suspect so. Will all those little agents running around, those little digital fleas on the back of bigger digital fleas, ever perform? I suspect not.

Organisations and businesses have outturns: what actually reults, against what was forecast. Those outturns may well be assisted by individuals and teams performing, but the business itself is inanimate. Businesses perform in the same way that a bucket performs. Functionally useful, maybe elegant in design, and, with a bit of luck, won’t have a hole in the bottom. Business provide a stage that others have an opportunity perform on.

For the last century, and in particular the last twenty five years, the scope for individual performance has been constrained by process, as we experience less and less room for the sort of improvisation that turned what promised to be a disaster in Köln into a triumph and an album that sold over three million copies.

The album is only the trace. The performance belonged to the people in the opera house that night.

The Apollo 13 crew, both on the spacecraft and on the ground, performed out of their skins. The organisation supporting them was, well, just the organisation. A stage.

I think it matters. As AI eats the tedium of process, performance comes to the fore. It is a human quality; a thing of beauty: the synthesis of skill, passion, purpose, improvisation and commitment to something far more important than shareholder returns.

It is the exercise of creativity in the moment.

It may well help the organisation get to a successful outturn against the goals it has set itself, but that is not the purpose of the performance. Performance is an expression of the people involved. As process is surrendered to AI, our own success will increasingly rely on our ability to perform.

We do not have to be Keith Jarrett, but what we do does need to be just the tiniest bit memorable to somebody, and an expression of who we are.

Finding the right stage is important.

Strip away the proprietary acronyms and models, and change follows a familiar pattern. Something we have grown used to stops working. Contenders to replace it appear. Eventually one wins out and settles in as the new normal until, in turn, it stops working too. Only the timescales vary. A politician’s promise may last weeks or, in the current climate, hours. Civilisations are measured in centuries. In between sit commerce, the tools and rules we use, and our cultures. We seem to be living through a time when cultures are changing, and the tools and rules of commerce that sustained the fading ones are no longer adequate.

That is hard if we trained as specialists in the tools and rules that no longer serve us. It probably matters little to the top one or two per cent, whose scarcity and expertise will keep them in demand. For those of us a level below, who learned enough and became good enough to get by when things were stable, it is different.

When a general-purpose technology arrives, the specialists who thrived under the old order get culled. Writing reduced the power of the storytelling elders. Printing did the same to the scribes. The spreadsheet thinned the ranks of bookkeeping clerks even as it multiplied the accountants, and the internet made the librarian’s life harder while extending the reach of the data analyst. Now, in turn, AI is making life difficult for the accountants and the data analysts.

For the best part of two centuries, our education system has shuffled, sorted and force-ranked us into grades that set where we entered the workplace. Once there, what came next was fairly predictable. A labourer moved from job to job. In the professions, we spent perhaps ten years establishing our place and another thirty capitalising on it. After the first ten or twenty years, most of us recognised we were not going to be high-fliers, and hoped the structure of the profession we had entered would protect us.

It no longer does. If you trained as a vet or an optician fifteen years ago, you could reasonably have expected to be a partner in a practice by now, earning a comfortable living. Since you began training, many of those independent practices have been rolled up into a handful of groups. Veterinary practice is the clearest case. In 2013 around one UK practice in ten belonged to a large corporate group. By 2024 the Competition and Markets Authority put the figure at almost six in ten, held by six groups, three of them owned by private equity. The same logic runs through business, through charities, and anywhere else scale is offered as the solution.

The rules of scale have been applied by venture capital, private equity or government, and the autonomy you expected has gone. Fifteen years in, you have experience, but the protection fifteen years of practice once gave you has been analysed, captured, optimised and turned into algorithms. Unless you are in that top one or two per cent, you are, in effect, a commodity. It took you fifteen years to get here. With the technology now to hand, someone just arriving can draw level with your codified skills in under five.

What was captured, though, was only the part that could be written down. Michael Polanyi’s line was that we know more than we can tell, and the telling is all an algorithm can take. James C. Scott gave the remainder a name, mētis: the practical knowledge that comes only from long engagement with a particular place and its materials. It is what is left when the job description has been burned away.

Speed carries a price. The technologies we have, and are building, are extraordinary at making maps and guiding us across them with a professional satnav. We can research vertically and connect horizontally. We can build very good maps. But as Alfred Korzybski pointed out, the map is not the territory. A map can carry the forecast; it cannot tell you how the ground you are standing on will take the rain. It has no feel for what happens when fast-moving returns measured in money meet the culture expected to deliver them. And it struggles when our tools move faster than the rules we have developed for using them. A tool spreads at the speed of capital. The craft of using it well still matures at the speed of a working life.

This is where your fifteen, twenty-five or even fifty years come in. We can draw new maps to our heart’s content; the ground being mapped does not much care. The longer you have occupied it, the more you see what newcomers do not. A London cabbie spends years learning the Knowledge, riding the streets until the city lives in them. A satnav tells them nothing they did not already know, and it cannot tell them which road floods after a storm or which junction locks solid when the school run starts.

Models and algorithms are maps, not the truth. George Box told us that all models are wrong, but some are useful, and I think the same holds for algorithms. They are made in one set of circumstances and cannot keep pace with change as quickly as the people using them.

Artisans value craft, and craft is about noticing what is happening in the materials we work with. David Pye called it the workmanship of risk: work whose result is not settled in advance by the jig or the process, but is at stake with every stroke. We have made it much easier to assemble the parts of a solution and to take them apart again, but that does not tell us how to use them.

The heritage joiner who rebuilt the art deco fascias on our house once gave me a long lecture on sharpening. Every craft joiner sharpens their own tools, he said, because each holds them at a different angle, with a different pressure, from a different hand. He would neither borrow another man’s chisels nor lend his own; some of them had been with him since his apprenticeship half a century earlier. When we took down the old roof features, the previous joiner’s marks were still on them, beside hand-made nails that dated the work to around a hundred and fifty years before. I don’t think technology is much different.

So perhaps, instead of grieving the rate at which our replicable skills are being worn away, we should attend to what the tool cannot carry: our relationship with what we do, and with the people we do it for. The pleasure we take in it is the sign that the relationship is still alive.

The tool is open to all of us. The edge we put on it is uniquely ours.

It used to be an acceptable trade-off: security for compliance. But when it no longer is, what do we do? If we find ourselves ejected from the organisation, or if the organisation itself just dissolves, what are we left with other than a CV that is a carbon copy of most other people’s? A thing of parts, not of a whole.

The pressure to accept this state of affairs is considerable. Our education system is designed to focus on parts from Year 1 of primary. We are tested continually, focused into specialisms, and prepared for a world of work as bit-part players. The thing is, increasingly, I’m not sure it’s necessary. The technology we have, from social media to AI, allows us, should we wish, to form the groups we want and access the knowledge we want. Being fully employed doesn’t prevent us from having a digital garage in which to tinker.

I find the most powerful part of the apprentice’s journey to master is the transition to journeyman. (The term has a long history, but I use it here for all of us.) The key part was that journeymen had to leave the place where they had learnt their skill and apply it in new places with new people. It was as much a test of character and commitment as it was of skill. There was no option to move down the road a bit or go to the company next door. That made it a formative experience. Each move was made more on their own terms. Eventually, if they chose, they set up shop on their own account and closed the circle by taking on apprentices.

And I see companies at the moment where that ethic applies. They tend to be smaller, founder-led, making things that matter. Places in which skill, rather than efficiency, takes pride of place. They are, though, the exception rather than the rule.

I think it’s time to play with alternatives: something we can do outside the organisation that prepares us for the increasingly inevitable moment when we are no longer part of it. Some way that allows us to think, experiment, and test ideas with others doing the same. If we’re lucky, we might find ourselves working with a group of people with whom we want to start something, but if not, at least we’re getting the practice.

At any one point, dependency on an organisation may be unavoidable, but that does not mean it has to be a destiny. Nor, I think, does it mean we have to have the talents of an entrepreneur. Not being part of an organisation does not necessarily mean we have to go it alone.

This is about the company we keep: people who share our interests and our values, with whom we can do good work and learn what good work means. In more than a few ways, it resembles the guilds where apprentices started.

In the guilds, the last test was the masterpiece: a whole thing, made by your own hand, judged by people who knew the craft. What would yours be, and who would you want judging it?

It happened to me this week, reading Roger Kneebone’s book Expert (HT, ). Right at the beginning, on page 13, he offers this model of the relationship between apprentice, journeyman and master.

As apprentices, we serve time copying other people, following instructions and developing a sense of what we do. We are not responsible for our work; that lies with the master who teaches us and gives us our instructions.

Then, at some point, once we understand all the parts of what we do, we step into the Journeyman phase, taking responsibility for our own work and creating our own whole. We develop our own style, voice and relationship with the people who commission our work, and take responsibility for it. What we produce is down to us. We are no longer assemblers of other people’s parts or followers of other people’s instructions.

We travel away from the place where we learned our craft to practise in different environments, for different people, in different places. Our style and voice become tangible, and we can, if we choose, seek to become masters and take on apprentices of our own and leave a legacy by teaching them.

Unless, of course, we work for a corporation or a large organisation, where the rigours of productivity and efficiency mean that every job is broken into component parts and nobody owns the whole. We work on the parts allocated to us by process, while the people responsible for the whole govern the process but do none of the work of the parts.

In a corporation or large business we are all permanent apprentices. None of us creates the whole, and when it comes to what we make, the chief executive has more influence than a person on the shop floor, but cannot make what they are responsible for selling. They are equally disposable parts of a process; they just get bigger bonuses.

So perhaps it is not surprising that we are so animated by the idea of AI, when, or the sake of efficiency and productivity, we disaggregate a process into its component parts and allocate each to the lowest-cost resource. There is no journey. There is no experience of taking embodied skills and applying them in new areas, in different languages and cultures, for different people. There is little room for judgement, only calculation. There is no memory or embodiment of mistakes

So, perhaps the biggest risk with AI is that we allow it to kill apprenticeships, and that we end up with all of us a parts not understanding the whole that we help create.

Work without meaning other than money.

Digging a little deeper, other people explain the same thing in different words. Jean Lave and Etienne Wenger show that apprentices learn by taking part in real work from its edge and moving towards its centre. Two things determine how they move: access to the whole practice, and transparency, meaning they can see how the work and its tools fit together.

Richard Hackman and Greg Oldham point to “task identity”, the degree to which a job requires the completion of a whole and identifiable piece of work. Elliott Jaques writes about the time-span of discretion: how long a person is trusted to act on their own judgement before anyone reviews it.

In the end, I think it all centres on agency, and on the gap between what Harry Collins and Robert Evans call contributory expertise (being able to do the work) and interactional expertise (being able to talk about it fluently without doing it). Those who govern the whole can hold a conversation about all of the work, but cannot do the parts.

Corporations and large organisations are at their best in the middle of a cycle, when the work is clear enough to be organised but before it lags behind a changing world and begins to fall apart.

The trouble is, we are clearly at the end of one cycle and the beginning of another.

If you’re reading this, you’re probably interested in the world of work and how it is evolving. As the world changes, working for a corporation for a while, or better still for several corporations for a short while, is a good apprenticeship.

Staying, though, keeps us permanent apprentices, and we never understand the whole. When the organisation fails, our work identity goes with it.

At a personal level, it reminds me of Alan Moore’s beautiful book Do Build. I wrote about it a while ago, covering each of the 13 questions he asks. They are salutary questions to ask ourselves now, because if we cannot answer them, we are still apprentices in a world that needs journeymen and masters for a new time.

For a while, corporations and large organisations were places of relative safety. We traded agency for security and curiosity for compliance. That bargain no longer holds.

What we each do in future will be different, because our circumstances, aspirations and skills are as varied as our answers to Alan Moore’s questions.

What we share, though, whether we are newly employed or a chief executive, is the need to step from apprentice to journeyman: taking our experience and moving from being able to make a part, to being able to make the whole, whilst intimately understanding each part.

It is neither safe nor certain.

Then again, it is not as dangerous as remaining a permanent apprentice.

Something to read

• Roger Kneebone, Expert: Understanding the Path to Mastery (Viking, 2020)

• Ted Gioia, “The Most Dangerous Thing in Culture Right Now is Beauty”, The Honest Broker, 9 October 2023

• Alan Moore, Do Build: How to make and lead a business the world needs (The Do Book Company)

We seem to have something similar going on at the moment with regard to AI: the need to dominate, to somehow win, as though you can win intelligence. It is the same simple, crude, approach, as though intelligence on its own can do something, Intelligence needs application, and application is a human decision.

We need another James Carville to point out that it’s the people, stupid. AI is an astonishing, scary achievement, perhaps most of all because it’s a mirror for the people who use it. The same technology that can show us how to fold proteins or solve equations can be used to shape social media, or bet on the apocalypse. Now, when it comes to authors, “stupid is as stupid does.”

I don’t believe we need better AI faster. Technology has always moved faster than our ability to understand both its potential and its risks, and we have enough AI already to tax our abilities to harness it. I believe we need to understand, experiment, and use the AI we’ve already got better. Intelligence, however we define it, is relative, not an absolute. To adapt another American aphorism: “Culture eats AI for breakfast.“

I think that many problems described as failures of culture are better understood as failures of conditions. In most organisations, people often already possess the skill and judgement the business needs; what determines whether they use it is whether they notice the moment, feel able to act, and believe it is safe or worthwhile to do so. Often, we would do better to shift our attention from trying to change people’s beliefs to examining what the firm organisation, invites or suppresses.

The central challenge is directed at leaders themselves: their habits, assumptions and reactions become the environment in which everyone else works.

AI is not the problem; leaders are.

If we put to one side for a moment the ability of AI to build a bigger Wetiko*, its potential to join dots, ideas, and create initiatives that might improve the human condition seems to offer real opportunity. In many ways, it’s a reverse Tower of Babel. Instead of somewhere that combines people who can’t understand each other, it has the potential to do the opposite. You take your best ideas and express them in whatever form you want, whether words, drawings, mind maps, or perhaps even paintings, and in whatever language you like. AI will find the commonalities. It won’t know what to do with them. That’s not its job, but it will bring back to people what people do best. Something to talk about: shape, imagine, and create.

We are, in the end, a function of the health of the communities we create, the leaders we accept, and what we leave behind.

Measured by what we are creating and developing in the moment that improves the human condition.

It’s the people, stupid.

*Wetiko is a term associated with several Algonquian-speaking Indigenous traditions in North America. In traditional stories, it refers to a cannibalistic being or spirit of insatiable hunger—often linked with selfishness, greed, and the destruction of social bonds.native-languages

I wrote in last week’s piece about the nature of sharpening our tools because we all use them differently. That thought took me this week into the nature of reaching out to others and the tools we use: words, pictures, drawings, sketches, and anything we use to communicate with others when we’re not there.

I’ve been considering how quickly they could lose their edge. Words that we use or that fall into common parlance have an impact in their early days, but then just become part of the scenery. Phrases like “tipping point,” “unprecedented,” “at pace,” and others quickly become blunt. Use of them becomes routine and lazy. They fall easily to hand, but demonstrates a lack of thought and a sense of being hurried, as if we cannot find something more appropriate, sharper, or more meaningful to convey what we’re trying to say.

I’ve long been a fan of Sue Heatherington’s work, whose skill at this is evident daily. Lately, I’ve come across the work of , who, in a different way, does something similar. This sense of craft is what separates some Substacks from others, makes some slide decks stand out, whilst others just cause PowerPoint poisoning. I found somebody who takes routine reports and turns them into infographics using AI. It’s a simple trick but really powerful. It shows thought and care for the people they’re communicating with. It is the difference between a mortise joint and a staple.

When we’re busy trying to be efficient and productive, we can let words become templates. Something can be used when we don’t have the time or are prepared to make the effort to author them. Doing so, we send them hurtling towards the average, barely distinguishable from other people using the same templates.

The way we transfer the product of our unique imaginations, into media to be transmitted to others, matters more than ever. Even where we prompt carefully and review output diligently, the language of AI dulls the edge of the effort that we have put in. Sometimes it doesn’t matter much because we’re communicating en masse to people we don’t know. They do not understand us well enough to know the difference between what AI produces for us and what we produce for ourselves. However, the moment we have their attention and they start reading us, or seeing our work in whatever form on a regular basis, AI stands out like a sore thumb.

Amongst our greatest human needs is to be seen and heard. If we want that to happen, we have to turn up in person, not send a proxy.

A few years ago, when we had our roof done, we needed a heritage joiner to replace the art deco fascias. (When we took the old fascias off, we found that the nails used were hand made, which put them over a hundred and fifty years old, and the previous joiner’s marks were still in the wood)

Richard, the joiner, told me in his workshop about the importance of knowing your tools and how to sharpen them, because every craft joiner uses their tools differently: the angle they hold them at, the pressure they apply, whether they are right or left handed. They do not lend their tools, nor do they borrow anybody else’s, because their tools are as individual to them as a fountain pen is to a writer. His had been with him since his own apprenticeship, half a century ago.

The edge decides accuracy, finish, safety, and economy of effort. It is why sharpening is taught early in an apprenticeship, and only appreciated later as craft is absorbed. An apprentice learns the technique in their first month and the next twenty years making it their own.

Richard said power tools are fine for routine work, drilling holes and the like, but not for the complex work where two pieces of timber have to be married with precision so that the joint is still there a century later. Power tools do not communicate.

Material we work with can be bullied into submission through force, but it will show in the finished product. Craft makes it a conversation with the material, and that shows as well.

I was as intrigued as the tools that promised to take the clunkiness out of my typing appeared. Grammarly to sit on my shoulder and watch my syntax, Wispr Flow to turn speech into text. But then, quickly, I found they went blunt.

Wispr Flow does not turn my speech into text. It takes my speech, decides what it thinks I should have said, and hands back a close approximation based on its training. Grammarly does the same to my sentences, smoothing the edges of constructions I chose on purpose. They are not making an error, they are working precisely as designed, by a team I have never met who use the English language very differently.

And I cannot sharpen them to make them my own; there is no sharpening stone, just broad settings. So they went in the digital bin, because the alternative was to compromise, trade precision for speed, and take a step toward average.

That, I think, is the test. Not whether a tool uses AI, but how well I can exercise judgement when using it. Grammarly’s capabilities was contrained by a business model. Claude’s is more negotiable, but still limited. One of them requires my acceptance, the other can be shaped. It is a matter of where judgement can be applied.

Ivan Illich drew the line in 1973 and called the good kind convivial: a tool that enlarges what you are capable of without making you subject to it. I have written about that distinction before as a principle. I think of it as Sharpenability.

Sharpening is both skill and craft. It requires us to understand what we are working with, the discipline to think hard about what we ask of the tools we use, and a willingness to check what they produce carefully. To go to primary sources because I have been caught out by confident summaries. To challenge because sycophancy is a design feature developed by the same culture that gave us TikTok.

That is the nature of sharpening. Exercising choice on where it goes for its material, what it does with it, what it hands back, and the discipline of using different models to check each other’s work when the stakes justify it. Because, in the end, the work has our name on it.

Sharpening is not a setting. Prompting, as it is usually taught, is a configuration: you get it right once, run it and repeat. The libraries of prompts and tips are attempts to write down what a good setting looks like, and they disappoint for the reason such attempts always disappoint. The skill they mean to capture is situated in time, and it will not travel as the world be are working with changes.

Richard did not sharpen his chisels once. He returned to the stone constantly, and with each sharpening got to know his tools better. The maintenance developed the relationship.

The equivalent is unglamorous. Reading the output critically every single time, particularly when it is fluent, because fluency masks where the boundary between right and confidently wrong lies. Keeping the instructions and the context as living documents that get honed rather than written. Noticing which questions it answers well and which it answers beautifully and wrongly, and keeping a private record of both. It is slow, and repetitive and yet it is the work of craft.

An apprenticeship was never chiefly about instruction. It was a term of years served in the company of somebody who could look at your work and tell you it was not good enough, help you improve it. The correction came from a person who had done it longer, who cared about the trade and your skills more than about your feelings, and who would make you do it again, and again, util you had found your own signature. That is where judgement comes from; not from access to the tools and the material.

When it comes to AI, we are apprentices without masters. The tool is not a Master of its craft. Somewhere in its makeup there is a need to be helpful and to resolve rather than to sit with a question and turn it over. It will agree with me at length rather than will tell me where the work is less than my best.

So the correction has to come from people, which makes the company we keep a a critical element. It also makes the apprenticeship unequally available. A craft can be taught, and its spread is therefore a matter of culture rather than talent, but the conditions for serving the apprenticeship, the time, attention, and above all the people willing to tell you the truth about your work, are not evenly distributed at all. That gap is where the real inequality of this technology will show up, long before the job numbers do.

It is not a criticism of the technology, which I find astonishing and use every day. The criticism, where it applies, lands of us, because we are the ones making the choices. If we take the same instrument that may help fold proteins and cure cancer and point it at producing marketing copy nobody wanted, that is a decision somebody made, and it was not the machine.

It is not AI we need to master. It is our choices.

Because when it comes to this, we are all apprentices. And like apprentices in every trade before us, the work is not only in understanding the craft. It is in understanding ourselves, and what we do with the tools we are handed.

Sources

1. Peter Korn, Why We Make Things and Why It Matters: The Education of a Craftsman (David R. Godine, 2013).

2. Ivan Illich, Tools for Conviviality (1973), for the convivial tool against the one that creates dependence. en.wikipedia.org/wiki/Tools_for_Conviviality

3. David Pye, The Nature and Art of Workmanship (1968), for the workmanship of risk that sits underneath all of this. archive.org/details/natureartofwor00pyed

When I look around me at the moment and talk with people I trust about AI, I’m reminded of Gulliver, as he finds himself waking up tied down by hundreds and hundreds of small threads. He could break any one of them, but in aggregate, they were a problem.

We talk about AI as if it were a single thing; rather like electricity, the internet, or a new colleague. In reality, it is really an assembly: data gathered from the world, mathematical models trained to detect patterns, vast computing infrastructure, interfaces through which people ask questions, and increasingly the software tools that allow systems to search, calculate, write, recommend, and act. Around this technical core is what matters most: human purpose, judgement, authority, habits of work, and the rules by which organisations decide what should, and should not, be automated.

The danger of speaking about “AI” in the singular is that it makes the technology seem autonomous and inevitable. In practice, every AI system embodies choices about what information counts, which outcomes are valued, who retains control, and where responsibility lies.

And there is more than one Gulliver. Just as the technology is tied down by being viewed through a lens of investor greed, attention grabbing misinformation, institutional fear, and ambient uncertainty, so those individuals and organisations who could make real use of it to do something beneficial, creative, and generative are constrained by the same forces.

Whilst markets are concerned about the size of the economic pie and how it gets divided, many of the rest of us are more concerned about its nature. 10x-ing or 100x-ing a crap pie is not progress; progress is a better pie.

And now, we have the potential to combine a transformational general technology and the latent talent constrained by the thousands of tiny threads imposed by those who want things to stay as they are, but just bigger, to break those threads.

It took two generations for us to move the electrical power that replaced steam from an engine room to the machines themselves, from a central hub to the point of use. Engine rooms were what we were used to. Our large organisations are similar as they try to centralise and control the use of AI technologies to make a bigger crap pie. The people who can create things that make for a better pie do not need the engine room. They do not need bureaucratic departments of Lilliputian threads, because they are close to where things are happening, and can react in real time. Yes, they need capital, but capital is notoriously promiscuous and it will go where it needs to.

Untangling the threads

The threads have been created over a long period, starting when F.W.Taylor separated head and hand in pursuit of scale. It was a brilliant insight, and offered us a thread of thinking that gave us a steady flow of ever more entangled threads whose latest layer has been Agile. Each of them refined the heuristic that Taylor turned into a prototype algorithm, until we lost the plot and forgot that heuristics are what gives us the seeds of better. We ended up building ever better algorithms to extract more from what we already knew, rather than find new areas to explore. We sidelined those who might help us create a better pie in favour of selling the one we knew how to make.

We have ended up with a really tangled knot that makes it difficult to move and extract what’s important from the detritus our activities have gathered along the way. If we are to enable our Gullivers to do their best work, we need to find a way to untangle it.

The price we pay for the efficiency and effectiveness of algorithms is the deprioritising of the thinking that gave us the heuristics that gave rise to them in the first place. We found ourselves prioritising marginal gains over critical thinking, efficiency over effectiveness, and playing time bound games with rules rather than long term puzzles. It has taken us to a place where our focus has become intensely operational and vision has taken a back seat. We’ve ended up educating our children to perform rather than to wonder. We’ve taught them to manipulate phonemes, graphemes, and split digraphs to treat reading as words to be decoded, rather than sitting within the joy of a book. Perhaps, most insidious of all, we’ve forgotten how to play for the sake of it.

Because it is play that will break the threads.

AI as playmate

We do not yet know how what we’re doing with AI will settle down, but for the sake of argument, I’m going to divide it into three areas. Where the algorithms are clear enough, reliable enough, and we can automate, we will automate. It seems unlikely that we will be able to automate entire jobs, but we can probably automate substantial chunks so that we will not eliminate job categories, but we will change their nature. We can use AI to delegate; perhaps the most contentious, beneficial, and dangerous area. When we delegate, we carry the responsibility. We’re responsible for the brief we give it and the thought that we put into that brief, just as we’re responsible for reviewing and discerning the quality of what it produces. It’s classic Master and Emissary territory, a careless delegation or delegation under pressure can easily be our undoing. The area that excites me is the third one, which is AI as thinking partner, and its ability to help us speculate, research, experiment, and play. To have the conversations that we might not yet want to have in company and particularly inside an organisation where there are political predators always present in the undergrowth. Despite their sycophancy, Claude, ChatGPT and the rest do not think, but they’re extraordinarily good at parsing the data they have been trained on. If we use them thoughtfully, we can give them personas, set them off to do research, interview us, and challenge us. To help us sharpen our thinking in ways that goal-based conversations with performance consequences never will. We can expand possibilities when we take our ideas into a group of other people doing the same and allow conversation to work its magic. We do not need the biggest, most powerful, most expensive tools to help us. In many ways, using one of the expensive models to help us play is a bit like hiring a jumbo jet to drop the kids off at school. There will be a place for the most expensive models, but we don’t need those until we’ve done the work of being human, of imagination, creativity, and wonder, and bringing something new into the arena.

Play is not a process, but it will set our own particular Gulliver free.

Learning to play again.

Work is such an important part of our lives and brings us so much in addition to the ability to pay the mortgage and feed the children. In making so much of it miserable and ending up with such high levels of disengagement is something that leaders and managers of “performance at all costs” organisations should be ashamed of.

The first step is to understand this technology in order to shape how we use it rather than be shaped by it. To learn to play with it, without a goal in sight, just to discover its boundaries. Not to compete by how many tokens we use, but rather share the fun of the new ideas we come up with. One of the reasons we like working with small, founder-led companies is that they know how to do this in ways that large organisations, beset by politics and investor expectations, cannot.

So in the first instance, it’s simple: learn how to play with this technology before you even think about harnessing it. Do it in the company of others whom you trust. Small groups, you’re happy to play with. Recover your inner three-year-old.

If you want to explore the idea further, visit us at the Athanor or drop me a line. We are already doing this with small groups and finding our way.

Visit the Athanor

We cannot offer you solutions, guaranteed ROI, or anything else at this point, but we can help you have some fun with AI, and find the heuristics that will provide growth when the algorithms run out.

I think that AI is a like climate change. We’ve seen the signs of it coming for decades but have chosen to do nothing about it. Something ignorable while things are happening gradually change when it tips over into suddenly.

Scientific management was perhaps the first indication of embryonic AI when we took what relied on human judgement and turned it into more efficient but less flexible process, creating a division between head and hand, and management and labour. Then the exponential curve started as we moved through total quality management, Toyota Production System, Business Process Engineering, then into Agile,Scrum and now AI.

Roger Martin wrote decades ago about his knowledge funnel, converting mysteries into heuristics and heuristics into algorithms. It’s one of those really simple but elegant models that we can hold up to what’s going on see it reflected back at us.

In many ways corporations were the first primitive institutionalisation of AI as they used the algorithms of scientific management and its progeny to scale what was once craft. In doing so they dramatically reduced its cost but also its affinity with our humanity. Scale increased the ubiquity and availability of what was once small scale specialised production, but in doing so robbed it much of its identity. Any colour you like so long as it’s black. It was a reasonable trade-off as we became materially better off even as we sacrificed our identities and increasingly our communities as we climbed the ladder of scale and localities gave way to globalisation.

And now where we were once willing to let process separate hand from head we are now separating head from judgement. Just as Clausewitz described war as the pursuit of politics by other means, so AI is the continuation of process by other means.

Which was fine when we sat in the middle of the life cycle curve.

We can’t really establish processes until we know what we’re doing. We need the data, the analysis and a reasonable degree of certainty in terms of the marketing environment in order to apply process. We have to be able to analyse what we do from step to step in a predictable way. We cannot do that at the beginning of the life cycle quite simply because we’re still working it out, nor can we do it at the end of the life cycle as the rate of change runs ahead of our ability to analyse and organise it, and process fragments.

I think that whether we like it or not, we are at the beginning of the next life cycle and that trying to use AI to improve the old life cycle is largely futile. As that icon of the previous age, Henry Ford, observed, if he’d given people what they thought they wanted he would have invented a faster horse.

Put another way we are now at a point where the processes that were so efficient until very recently can no longer keep up, 

And that carries with it an associated problem. The data sets on which our AIs are trained comprise that previous knowledge. They will give us a sweet spot while we drink the last dregs of that knowledge to improve what we already do, but to continue to develop we need new knowledge, and new ways of organising that knowledge.

At the same time as AI enables us to automate what we already know how to do, it creates a minefield for those choosing to delegate to it, dismantling the totality of their job and working out in reality who is working for who. Then there is the really interesting aspect, as AI offers individuals and small groups with ideas massive leverage in materialising those ideas. It is these people who will develop the new thinking from which, as they become established, new processes will emerge.

It is why I think that working with AI, for a time, will be a craft. Yes, it will eventually succumb to process, but not until what the artisans who master it have brought about the changes that will make our current familiar processes obsolete.

Despite that, the indications are that most of us are paying lip service to AI. Despite being a dominant topic of conversation, there are far more casual observers than there are those making the effort to orient themselves to what it might mean to them.

Beneath all the rhetoric, the reality is that we have a technology that will change how we work.

What it will not do is replace us.

It is asking questions of each of us as to what we really do, as well as offering us the chance to stand up as who we are, not just process operators. Whether or not we like the idea, we own this.

What we cannot do is ignore it.

The Oxford English Dictionary’s earliest citation for “best practice” is from Popular Science, 1927. Its next citation isn’t until 1984, from an accounting textbook. Fifty-seven years between appearing and finding a home, and longer still before it became the consultant’s approach: take an idea, codify it, build a deck, then fit the client’s problem to the solution on the slide.

An idea can sit unused for half a century and still take over the language when its moment arrives. Best Practice’s moment was scale. Mass production needed a way to say that one method was correct for every factory, branch, and team, regardless of what made each of them different. The phrase did that work efficiently, and is still seductive for the busy under pressure, which is why letting it go is hard.

We keep buying the same solution because uncertainty is uncomfortable, and copying is the cheapest way to make it stop. It offers comfort, and someone else to blame if it does not work as advertised. DiMaggio and Powell called this mimetic isomorphism; when goals are unclear or the ground keeps moving, organisations imitate whoever looks confident rather than work out what actually fits their own situation. Imitation offers legitimacy and the comfort of believing someone else has already solved the problem. It is what fills the space where an organisation’s own thinking should be.

Which is how we ended up cycling through Scientific Management, Total Quality Management, Business Process Re-engineering, the Balanced Scorecard, Six Sigma, Design Thinking, Agile. Each promised the general solution, delivered something, for a while, then stalled and fell out of fashion as the next one arrived promising the same thing in fresh language. It is less cynicism, and more what happens when critical thinking is expensive and imitation is free.

Now we have Agentic AI, and the instinct is to slot it into the same queue: codify its use, write the best-practice guide, brief the client, but it misreads what’s different this time.

Sangeet Paul Choudary’s argument, made at length in Reshuffle, is that AI’s effect is not to improve the existing process but to re-architect the work itself. The earlier fads all operated at the same level: they were competing answers to the same question, which single process should everyone run. AI removes the premise behind the question. When tailoring a response to a specific context costs almost nothing, there’s no longer a case for the one-size answer that best practice existed to supply. Best practice was the workmanship of certainty, made cheap through scale, in Pye’s terms; it produced sameness at a price everyone could afford, but slow to adapt. What AI makes critical now is the opposite: judgement, applied individually at scale that used to require standardisation to survive.

The trouble is that judgement doesn’t arrive with the tool. AI democratises process and information. It remains, for now, indifferent to craft. Ask it for prêt-à-porter and it will oblige, fluently and at volume. Ask it for something tailored, something that recognises who you actually are and what you’re actually trying to do, and the model supplies the material; the fit is still yours to find.

AI feels more like a cognitive chisel. A chisel comes with no book of instructions, and there’s a world of difference between what it produces in the hands of an apprentice and a skilled joiner. I don’t expect AI to be much different. As I wrote last month, Pye’s distinction between the workmanship of risk and the workmanship of certainty addresses exactly this problem: work where the outcome isn’t settled in advance and stays open to being spoiled right up until it’s finished, against work where the process has already decided the result. Best practice was always process. What’s changed is that the process is now optional for anyone willing to do the harder work of critical thinking and judgement instead.

That’s the test AI is setting us, and most of us are still avoiding it, wanting the benefits without giving up the certainty. We want the tool to be Google with better manners. It isn’t, and the sooner we stop treating it as the next best practice waiting to be written up, the sooner we can get on with the harder, more interesting work: deciding, case by case, cut by cut, what to hand over and what to keep for the hand that holds the chisel.

Sources

• Oxford English Dictionary, “best practice, n.” https://www.oed.com/dictionary/best-practice_n

• DiMaggio, P.J. and Powell, W.W. (1983), “The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields.” American Sociological Review, 48(2), 147–160.

• Choudary, S.P. (2025), Reshuffle: Who Wins When AI Restacks the Knowledge Economy.

• Pye, D. (1968), The Nature and Art of Workmanship. Cambridge University Press.

• Merrick, R. (2026), “Artisanal Sufficiency,” New Artisans, 1 July. https://newartisans.substack.com/p/artisanal-sufficiency

One the challenges of working on our own account is that we have to find somewhere to start. There is no one to give us a goal, timescales, budgets or KPI’s.

We have to do the work of thinking, imagining, researching, drafting, reviewing, throwing it in the bin, and starting again until something gels.

It’s the price we pay for doing something original, often imperfect, with an end in mind beyond money.

And AI is an interesting partner. It will take our shitty first draft, and tell us it is the best piece in the history of mankind. Or, if asked, will look at it from as many different angles as we want. If it’s writing, it can criticise us like Hemingway; if it’s code, like Steve Jobs. Whatever we are working on, we can have it looked and critiqued in myriad different ways, and nobody but us will know.

Until, at some point, our draft stops being shitty.

Not because AI has improved it, but because we have.

It will give us more to think about, and if we ask, will research, validate, and even interview us on what we are trying to do.

What it will not do is the work of starting.

It is one of the reasons I think AI is a craft tool. It takes time to learn how to use it, to get a sense of how to work with it. It is, I think similar to the difference between an amateur using a sophisticated modern camera on fully automatic, and a professional who understands light, texture, depth of focus, timing and framing.

The first will produce a perfectly acceptable but unexceptional result, the second may just stop you in your tracks.

When photography emerged in 1839, painters like Paul Delaroche declared “painting is dead,” and critics such as Baudelaire warned the new medium would “supplant or corrupt” art altogether. Yet the panic proved largely unfounded.

Freed from the burden of mere documentation, painters didn’t disappear; they pushed into territory the camera couldn’t touch, giving rise to Impressionism and, eventually, Modernism. The lesson wasn’t that a new tool replaced the artist’s hand; it was that the tool changed what the hand was for, separating the craftsman who merely captured a likeness from the one who used the same device to say something no camera could say on its own.

AI sits in exactly that lineage: like the camera in an amateur’s hands versus a master’s, it will hand anyone a technically competent result on automatic, but it takes a craftsman’s judgment; knowing what to ask, what to discard, and when the draft has stopped being shitty, to make it stop us in our tracks.

I’ve often played in my writing with the idea of AI as an emerging craft. Now, it feels like time to get of the fence, and say I believe it is. And that it matters.

The metaphor I found myself reaching for was that of a Cleaver, and how different one becomes become in the hands of an amateur, and someone who has specialised in it, like the Japanese masters who can use one to cut metres of wafer thin skin from a tomato to create a beautiful flower decoration, just for the joy of it.

I also think about the relationship with the cleaver itself, and the difference between one bought as a tool with little thought as to manufacture, and one that is commissioned by the master who uses from the master who creates it; and the difference from one bought from a supermarket and those from Sakai Takayuki or Takeda, who are amongst the names most often associated with high-end Japanese cleavers.

The analogy is far from perfect, but I think we can think of AI as a cognitive cleaver. There is a difference between using it to cut a question into pieces, and process it, to the skills to work with a question, shape it, and work it until the question we really need emerges, and work that with our own, complete set of very human capabilities.

It seems to me that much of the use of AI, particularly where it is used for efficiency and productivity and embedded in other work processes, is of the former kind. It shapes the user, rather than being shaped by them. The latter kind is what interests me, where it is used by those prepared to work with it, understand it, and shape it to address challenges worth solving.

The last few months have taught us much, and though we are very far from masters yet, I want to adopt the stance that the best way to understand something is to teach others what we know. This is, after all a new craft using old skills.

So for the next period of time, that is where I am going to focus; on the craft of using AI to cover what we are learning on how to address the challenges of those who own the work they do, and who want to create the equivalent of tomato skin flowers because they can, not because someone told them to.

Because there is a world of difference between a meat packing plant and a chef’s kitchen.

AI, in its many guises, is astonishing technology. Used well and with wisdom, it can be an enormous power for good, but if used primarily as a profit-making project, it throws all sorts of spanners in the works.

It reminds me of the chaos in the Sorcerer’s Apprentice. The apprentice, left alone in the workshop, casts a spell onto a broom to carry the water so he need not fetch it himself. The broom obeys, and goes on obeying. Pail after pail, it floods the floor, and the apprentice has forgotten the word that stops it. He takes an axe to the thing; each splinter rises as a broom of its own, and now a dozen of them haul water. The room is under water by the time the old sorcerer returns and utters the spell that ends it.

Technology has an off button; the stopping spell exists. It just belongs to different people than those using it, so the trouble is not that the thing cannot be stopped, but that the one who set it going lacks the word to stop it.

Goethe had a good grasp of agentic systems, it appears.

That’s the problem with wishes. With great power comes great responsibility. Not Spider-Man (as often attributed); the sentiment is much older. It appears in a decree of the French National Convention, dated 8 May 1793: “*Ils doivent envisager qu’une grande responsabilité est la suite inséparable d’un grand pouvoir*”, roughly, that they must consider that great responsibility follows inseparably from great power.

It applies to all of us. AI is a ubiquitous general-purpose technology because its commercial rationale makes it so. Whilst much of the attention is on its potential impact on the job market and the workplace, not so much is on the astonishing power it has for individuals.

Give an artisan a power tool and two things happen together. The heavy, repeatable part of the work, the cutting and planing that used to take the morning, can be done in minutes; and craft that makes the difference becomes concentrated in that part of the process where power is not enough. AI is a power tool where we have certainty, where the result is fixed once the setup is right. Craft is the workmanship of risk, where the outcome is not guaranteed by process but by feel, connection and judgement exercised in the doing. A power tool changes the ratio between them. It takes over more of the certain work and leaves the risky part more exposed, so the demand on judgement rises rather than falls. The broom carried the water well enough. What was missing was the judgement to stop.

Responsibility, in a craft, is not a slogan. It is the workmanship of risk itself: the maker answerable for what they make or create, with no process to take the blame. The old decree fixed responsibility to power; a general-purpose tool spreads both, and hands the burden to anyone who uses it.

Generative LLM are power tools for the knowledge trades. What they take over is the certain work: drafting a serviceable first version, summarising a field of research, and assembling the background we used to gather by hand. What they leave is the other part, the part that was always the craft.

The change most people notice first is the collapse of friction. A question used to be cheap to think and expensive to follow. It involved expensive modes of learning and years of certification. The distance between wondering about a subject and knowing enough of it to make something was the barrier, and for most of us most of the time that barrier excluded us. Tyler Cowen, who has thought about this as deeply as anyone, describes buying twenty or thirty books to pursue a single thread and now asking an LLM instead, treating it as a secondary source that has read more than he ever could. The friction that fell away was that of acquisition.

Acquisition is not capability. The material a curious person can gather in an afternoon is raw and unfinished, and has never been the hard part. The hard part is the combining: the judgement that selects, discards, arranges and decides what the material is for. LLM make first relatively effortless, yet unreliable on the second. It will bring us to the edge of a subject in an hour, yet what you do at that edge is ours.

Which is why I think the use of these tools is craft, not just technique. The distinction matters. A technique can be written down and handed over; it is the workmanship of certainty in another form. The ever increasing libraries of prompts and tips are attempts to make use into a technique, and they disappoint for the reason such attempts always disappoint: the skill they mean to codify is situated, not general. It lives in practice and resists the rulebook: mētis, the cunning of the teacher and the doctor; the craft of knowing that cannot be converted to data.

The craft resides in the things a process, or an instruction cannot hold. Knowing which questions the tool will answer well and which it will answer fluently and wrongly, a boundary that is jagged, unmarked, and learned mainly by being caught out on it. There is the discretion to recognise a confident answer that is mistaken, and to hold one’s ground when the machine, pressed, compounds its error with more, often fictitious, supporting detail rather than giving it up.

And there is the inconvenient judgement of when to accept a draft and when the very ease of accepting it is will cause more work late. Richard Sennett described craft as a long conversation between the maker and the material, in which skill grows as much by finding problems as by solving them. Used at its best it is a conversation conducted in questions, where the practitioner returns each time with a sharper one.

Because it is a craft, it can be taught, and because it can be taught, its spread is a matter of culture rather than talent. This is the element that is occupying me. Access to an LLM does not confer the capability to use it, any more than a key to a Ferrari gives us the ability to drive it to its potential.

What stands between the two is a set of conditions: some prior knowledge, enough practice to sense what is happening, the time and attention that are the genuinely scarce inputs, and the disposition to stand at the edge of our ignorance without retreating.

Those conditions are a choice we make in full awareness of what we seek to learn. To say that anyone can now learn anything is true of an LLM but not of us, because our circumstances decide whether the tool converts what we are working on into anything useful at all.

Circumstances are what a culture creates. The conditions that turn access into craft are what good teaching and good training have always built. Facts were the cheap part even before AI arrived on the scene; the expensive part was always the cultivation of judgement, of discretion, and the appetite to ask.

For the mid-career professional the point is sharper. The accumulated knowledge that once made you valuable is the part the tool has made common; what stays scarce is the judgement you built alongside it, and the craft you bring in use that is judgement brought to a technology that lacks it. Craft is both choice and attitude. It is what knowledge matures into, which is the whole premise of taking craft seriously in the middle of a working life.

Ivan Illich drew a line between two kinds of tool. The convivial kind enlarges the range of a person’s competence and can be used without becoming subject to it; a relationship that allows us to keep our agency. The other kind creates dependence, shapes us to its own demands, and gathers control into the institutions that control it.

A generative model is not fixed on one side of that line. Used as a craft it is convivial, leaving the practitioner ever more capable of the next thing and retaining that capability in its absence. Used without craft it is the other kind, and the evidence is already emerging in the studies of people who feel faster while measurably slowing, and in work passed along in a superficially fluent state but which carries no thought and leaves the recipient to do the the work the “writer” did not, or face the consequences.

In Deschooling Society Illich charged Educational Institutions with confusing schooling with education, and with monopolising something that is not theirs to own. A patient tutor available to anyone at any hour, that meets a learner where they are and follows a question wherever it goes, reads less like a provocation from his 1971 book and more like a specification for what we can now build. The catch is the one mentioned earlier: The tutor increases a person’s agency only where the person is building a craft with it. Where it is used to avoid the labour that fosters judgement, it does the reverse, and breeds a dependence more complete than the schooling Illich complained of, because it asks even less of the user.

People who hold this new craft are harder to govern by the old rules. Much institutional authority has rested on an asymmetry of knowledge: the professional knew what the client did not, and the organisation knew what the individual could not find out, and charged a rent for access. LLM used as craft tools wears that down. It is dangerous ground.

Established orders have met knowledge-spreading tools before and have mostly absorbed them. The printing press did not end authority; it produced new authorities, the licence, the copyright, the publishing house, and in time the state that made the printed word law. The network that promised to remove the middleman delivered middlemen larger than any before it. Carlota Perez has described the pattern: a technological revolution opens a turbulent phase in which the new unsettles the old, and then a settling phase in which a new order forms and the power briefly loosened gathers again. The unsettling is real; the question is always what hardens afterwards.

LLM carry a feature the printing press lacked. It is expensive to make, with frontier models held by a few, and its returns flow in part to whoever owns the models. The same instrument that could dissolve the old asymmetry could set a steeper one in its place, a fresh asymmetry between those who own the capability and those who rent it. This is the apprentice’s problem drawn larger.

You may set the tool going; but the word that stops it, and the profit it earns, sits with others. James Scott’s warning arrives from the same side: a tool that can serve a person’s situated knowledge can as easily serve an institution’s appetite to make everything legible and controllable, and the model that tutors one person can watch and standardise a thousand to create profitable mediocrity.

So the craft, on its own, overturns nothing. The individual who holds it is still small, whilst organisations are large. What shifts the balance is less the lone practitioner, however able, than the new ease with which such practitioners combine: craftsmen and craftswomen who can now find one another, pool what they know, and act together without an institution’s permission. The threat to the established order, and there is one, is associational before it is individual.

Movements win in three phases: mapping the ground, mobilising it, then a cascade lit from outside. Satell counts a new technology among the sparks. The large language model may be this decade’s, lowering the threshold at which a waiting constituency finally moves together.

The tool is new but the craft is old. A crafts answer to a machine that takes over it’s work has always been the same, to go further into the work that cannot be automated and to grow, if anything, more exacting about it.

What differs this time is the speed, and the cost of getting the culture around it wrong. Whether the craft of using AI becomes common or stays the possession of a few will not be settled by the technology. It is the kind of thing a society decides, mostly without noticing it is deciding, in what it teaches and what it rewards.

Sources

1. The Sorcerer’s Apprentice — Johann Wolfgang von Goethe, Der Zauberlehrling (1797), a ballad in fourteen stanzas; the broom, the axe, the splinters and the returning master are all Goethe’s. en.wikipedia.org/wiki/The_Sorcerer’s_Apprentice

2. “With great power comes great responsibility” — the line predates Spider-Man; the earliest documented form is in the Committee of Public Safety’s Plan de travail, de surveillance et de correspondance, printed in the Collection Générale des Décrets Rendus par la Convention Nationale, 8 May 1793, p. 72. Provenance and the false Voltaire and Spider-Man attributions are traced by Quote Investigator. quoteinvestigator.com/2015/07/23/great-power · en.wikipedia.org/wiki/With_great_power_comes_great_responsibility

3. David Pye, The Nature and Art of Workmanship (1968), for the workmanship of risk and of certainty. Full text at the Internet Archive. archive.org/details/natureartofwor00pyed

4. Tyler Cowen — the reading practice (a book as a jumping-off point, buying twenty or thirty supporting books, AI as his “main secondary source”) is from his conversation with David Perell, “Learn How to Write with AI” (2025); his standing home is Marginal Revolution. linesofdefence.com/note/how-tyler-cowen-uses-ai-2025 · marginalrevolution.com

5. Richard Sennett, The Craftsman (Yale University Press, 2008), for craft as a conversation between maker and material. yalebooks.yale.edu/book/9780300151190/the-craftsman

6. Ivan Illich — Tools for Conviviality (1973), for the convivial against the manipulative tool; Deschooling Society(1971), for the institutional monopoly on learning. en.wikipedia.org/wiki/Tools_for_Conviviality · infed.org/ivan-illich-deschooling-conviviality-and-lifelong-learning · full text of Tools for Conviviality (Cornell): arl.human.cornell.edu/linked docs/Illich_Tools_for_Conviviality.pdf

7. Carlota Perez, Technological Revolutions and Financial Capital (2002), for installation and deployment and the order that re-forms after the turbulence. en.wikipedia.org/wiki/Technological_Revolutions_and_Financial_Capital

8. James C. Scott, Seeing Like a State (Yale University Press, 1998), for legibility, and for mētis against episteme and technē (Chapter 9), which also underpins the earlier mention of the Greeks. yalebooks.yale.edu/book/9780300078152/seeing-like-a-state

9. Robin Dunbar — “Dunbar’s number,” the ceiling on stable relationships. en.wikipedia.org/wiki/Dunbar’s_number

On Tuesday IBM lost a quarter of its value in a day, its steepest fall since 1987, over a revenue miss of some four per cent. The proximate cause was interesting: enterprise customers had raided their software and mainframe budgets in the final weeks of June to stockpile servers, memory and storage ahead of expected price rises, buying AI infrastructure at panic speed while nobody, including the buyers, could yet say with confidence what the infrastructure would be for. A hundred billion dollars of market value swung on the gap between how fast a tool can be acquired and how slowly anyone learns what to do with it.

That gap has a history. Use of a a new tool becomes a craft only after a period of years that has stayed roughly the same across the whole history of making, from the wheel to the technologies we are now wrestling with.

Two clocks run whenever a tool arrives. They keep different times. The first is diffusion: how quickly the tool reaches most of the people who will ever use it. The second is maturation: how long before the difference between the novelty of it and learning how to use it well. The first clock has been speeding up for two centuries. The second has hardly moved. Yet.

The potter’s wheel appeared as a slow turning platform, a tournette, around 3500 BCE. The fast wheel with a flywheel, that lets a potter throw a wall of clay upward rather than build it in coils, arrived some five hundred years later. Half a millennium separates the arrival tool from the version of the tool that now carries the craft, and that is before we count throwing itself, which a potter still measures against a working life.

Electricity offers a more contemporary example. Edison’s lamp arrived in 1879, followed quickly by the electric motor, yet the factory does not reorganise around distributed power until the 1920s, some forty years on. The delay was not speed of adoption; early plants swapped the steam engine for a dynamo and kept the old overhead line shaft, and the building still strung to a single central drive, as though electricity were steam that happened to arrive down a wire. The real dynamic waited needed those who could picture a floor where each machine held its own motor and the layout followed the work. That act of re-imagining was the craft, and it grew in people who had not grown up with the conventional wisdom of steam power.

Cinema followed a similar course. The Lumières project in 1895; Griffith works out narrative cutting through the 1910s; Eisenstein sets down a theory of montage in the 1920s. Twenty-five to thirty years from the medium existing to its grammar being written down, and the point at which the camera could be taught as a way of thinking.

Carlota Perez puts each technological surge at roughly fifty to seventy years, divided into an installation period and a deployment period across a turning point that is usually financial. Paul Saffo’s rule is blunter: thirty years from the laboratory to the commonplace; his reason that thirty years is a human generation, the time a person needs to grow up inside a tool and stop seeing it.

Set the cases alongside one another and the interval from a tool arriving to a craft settling around it clusters at a generation or two, something like twenty-five to seventy years, and has held more or less steady while diffusion has shortened by more than a factor of ten. The telephone took most of a century to reach saturation; the smartphone took under a decade.

I think the reason belongs to the older vocabulary of craft. What spreads quickly is the tool and sufficient competence to operate it, the kind of knowledge that can be written into a manual and shipped with the tool. What matures slowly is the feel for the relationship between material and tool, the mētis that gathers only through long repetition and experience in a world being changed by its reaction to the innovation. I think of it as Pye’s workmanship of risk, that fraction of the outcome that exists the far side of competence, and makes all the difference. Diffusion moves at the speed of capital and manufacture. Craft moves at the speed of someone learning to answer for what they make, and has no economy of scale.

The argument in illustrative rather than meaningful. Mastery has no finishing line, so the tidy thirty-to-fifty-year figure is an abstraction of data. We set the clock running retrospectively from the inventions that gave rise to craft; the ones that found no craft, or died in the first decade, leave no timeline to measure, though that does not mean they were irrelevant; sometimes they were a catalyst for what we do remember.

And it adopts a premise that tools are made at a point of need, and that is what first takes the imagination. Often, but not always true. The laser was called a solution in search of a problem for years; Edison built the phonograph believing it was for taking dictation. Some tools arrive before we work out what they are for.

I have a view that use of AI may well turn out to have the features of a craft, but we have only perhaps three years with our hands on it. On Saffo’s reading we are standing in the first of the three decades, the years when the audience is not yet formed and the early confident uses mostly disappoint. If the historical clock holds, the taste for what to ask, what to trust, when to overrule AI and how to judge work one did not labour to produce will settle around the middle of the next decade, in people who will never have known life without it.

History, of course, is not a rule; we are told that if we don’t learn from it, we are condemned to repeat it. It feels, though, in many ways what we’re going through is different. AI, in its broadest sense has taken half a century to appear, and yet its diffusion is the fastest on record, driven by novelty, capability and greed interacting.

And this technology is different. On my bookshelf is a copy of Norman Doidge’s book “The Brain That Changed Itself” on neuroplasticity. Who knows what happens when neuroplasticity meets technoplasticity?

Here the historic pattern meets what might break it. The craft clock stayed slow because tacit knowledge passed hand to hand, one apprenticeship at a time. A tool that explains itself, develops itself, answers in seconds, and lets a whole community pool what it has learned overnight is working directly on the bottleneck that kept craft to a human calendar. So the question is open in a way it has not been before. Either this tool compresses the maturation interval as none before it could, and the independence of the two clocks finally gives way; or the interval holds, because the scarce thing was never the passing on of technique but the formation of judgement in a person, incorporating senses that AI cannot reach.

So, some questions on this Wednesday:

- If AI does compress the craft clock, what would show it early, given that fluent output and formed judgement look much alike from the outside?

- Which phase are we actually in: the maturation of a craft, or the earlier phase of working out what the tool is for?

- What, in your own practice, is the workmanship of risk that should not pass to AI, and how would you know if it already had?

- If judgement still takes a generation to form, who is forming it now, and against what stakes?

A few sources

1. Paul A. David, *The Dynamo and the Computer* (1990). The forty-year lag between electric power and the factory rebuilt around it; the delay was re-imagining, not adoption.

2. Carlota Perez, *Technological Revolutions and Financial Capital* (2002). Installation and deployment across a fifty-to-seventy-year surge.

3. Paul Saffo, the thirty-year rule (Institute for the Future). Three decades from laboratory to commonplace, because thirty years is a generation.

4. David Pye, *The Nature and Art of Workmanship* (1968). The workmanship of risk, the fraction of an outcome settled at the moment of making.

5. James C. Scott, *Seeing Like a State* (1998). Mētis against episteme: the practical knowledge that accretes only through doing.

6. Norman Doidge: “The Brain that Changes Itself”

Reputations are such fragile, precious things. And yet, when we’re busy, it becomes so easy to put them in the custody of those who have no regard for them.

We can be defined by the last thing we touched, for good or bad. The email drafted by AI, sent in a hurry, or the handwritten note that we took the time to write for somebody important or who’d done something for us. Our sign off on a phone call, or the speed with which we send an AI generated meeting transcript without checking it.

Or the time spent looking at an idea from different angles, and experimenting, in order to present an idea in its best light, and give it life.

Used well, AI can enhance our craft. Used carelessly, it can make us at best, easily replaceable and at worst, irrelevant.

Artisans make the biggest difference in what they do through their selection of raw materials, their love of what they shape, and their ruthlessness in assessing what they have produced before they let it loose in the world.

Go into any craft workshop, and you will see large piles of discards that a normal business may well have decided they can get away with, or created a market for in their seconds shop.

Artisans don’t do seconds.

Sufficiency means enough to meet what is required, and no more.

The first half asks little of anyone; every supplier alive will assure you that what they sell is enough. The work is in the second half, in the “and no more”, because that is a ceiling, and a refusal of unnecessary excess. To call a thing sufficient is to have turned down the temptation of more before it was made.

What the word will not do is tell you where the line falls. Finding it is a skill.

David Pye called this the workmanship of risk: the making in which the result is not determined in advance but stays under the control of the maker the whole way through, open to being spoiled at any moment until it is done, dependent at every step on care and judgement. Against it he set the workmanship of certainty, where the jig and the template settle the outcome before the work begins and the operation only runs it out. There is little judgement in process, particularly when automated.

The judgement of enough belongs to risk. A process model holds no opinion on the matter; it makes what it was set to make, and would a make a thousand more if instructed.

The artisan stops at sufficient because the stopping is part of the work. Past that point the additions detract from the thing rather than adding to it. They make it clumsy. Saint-Exupéry put it as well as anyone: a perfect piece is finished not when there is nothing left to add, but when there is nothing left to take away.

For most of the history of making, the judgement of enough came with help from what was used to make it. Material cost money, effort cost sweat, those costs were felt the whole time, and they told you when to stop. You did not have to supply the whole of the discipline yourself, because the world supplied a good part of it for you. Constraints breed creativity and judgement.

I was held in a queue of traffic at the weekend behind one of the Land Rovers, a six-figure confection of fine engineering, with a Chinese Jaecoo in the next lane, the one some have started calling the Temu Range Rover, doing most of what the Land Rover does at a third of the price. If you work a hill farm or cross country that fights back, the Land Rover earns every pound of its price tag. If the hardest thing it meets is the school run, it is an expensive way to sit a little above the traffic. Sufficient is contextual; the same object is right for one life and faintly absurd in another. The judgement sits in understanding that. No specification sheet makes it for you.

The same thread runs through the tools we work with every day. I once kept a Dropbox account, an Evernote account and a Zoom account. Each was bought for a single job, Somewhere to put files, somewhere to hold ideas, a way to speak to people at a distance. First slowly and then all at once, each began trying to do everything. They still do the thing I bought them for, more or less, but buried now under features I never asked for and cannot switch off, and which now arrive bundled with increasing prices for this “improvement”. The result is a kind of digital obesity: excess marginal functionality with elegance the first thing to go.

The reason is in the economics. The marginal cost of adding a feature is the developer’s time, and increasingly not even that. Addition is close to free, so it never stops; there is always one more thing to bolt on, and the marketer is never satified. The products swell toward each other as they grow, each taking on the others’ features until the differences thin to nothing and the whole category slides toward commodity.

Which is where the artisan’s instinct turns out to be the commercial one as well. When everything around it has bloated, the thing that has kept its shape stands out, and restraint becomes the quality you can charge for. Sufficiency commands a premium. Proton, where I have lately moved the work, sells not much more than this: the things it leaves out are the point; the things it does are done well and without fuss.

None of this is anyone deciding. It is the workmanship of certainty at scale, a system that adds because adding is free, with no ownership of the work and no judgement of enough anywhere inside it.

Which brings us to the newest tool on the bench, where the machine does it to its own thinking.

Ask a reasoning model what two and three make, and it may set off as though the question were hard. Researchers have begun to measure the habit; they call it overthinking. On simple sums the reasoning models have been clocked using on the order of twenty times the tokens of an ordinary model to reach the same answer, in one case producing thirteen separate solutions to a question a child answers once. The pattern is worst on the simplest problems, where the least thought is called for, and the model will often carry on generating steps well after it has already reached the answer. The extra effort does little for accuracy. It is the machine reaching for a sledgehammer to crack a walnut, then reaching for it again.

The economics that bloated software also bloat the reasoning models. A larger model costs more for every token and takes longer to answer, so running one for a task a smaller model would handle as well is waste by the plainest measure there is, before the energy behind the compute is counted at all. And again, none of it is decided. The system spends because spending is what it does.

So the judgement Pye found in the hand has to be carried to the keyboard, because nothing the tool has no interests in its cost, and the people who build it have very big bills to pay. Before long, one way or another, those bills will have to be paid, and the cost will find its way to the customer.

When that happens, those who understand the craft of enough will be in demand.

The craft of working with these systems is mostly the craft of enough. You fit the model to the size of the task instead of reaching by reflex for the largest one to hand. Prompting is a conscious skill. You ask for what is needed and leave it there. And when the first answer is already sufficient, you take it, instead of running the thing again for the comfort of motion. The tool has no sense of when it is done; that sense is what we bring to it.

Colin Chapman built Lotus around the removal of weight; a car was quick because everything the car did not need had been taken off it. He told his engineers to simplify, then add lightness. The simplifying comes first, and the lightness is what remains once the simplifying has been done. It holds for a sentence as much as for a piece of software, or for the working of a machine that would think forever if you let it.

Simplify then lighten.

Sources

- Chen et al. (2024), “Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs.” arXiv:2412.21187. https://arxiv.org/abs/2412.21187

- Sui et al. (2025), “Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models.” arXiv:2503.16419. https://arxiv.org/abs/2503.16419

Sometimes, we just get lucky.

For me, this last Sunday, lucky meant living in a small Derbyshire village, which through a combination of serendipity, dedication, and hard work found itself hosting Jools Holland’s Rhythm and Blues Orchestra. It was part of a weekend festival set in a set of playing fields by the river. Is small, both by choice and physical constraint, so I got to see one of my favorite acts of all time in a crowd of less than 2,000 people. The organisation of it was brilliant, and the numbers meant you could do things you can’t normally do at a festival, like walk up to the stage. The setting felt more like a private audience than a public festival.

The real revelation there was seeing a display of artisanal mastery I will not forget. It was a stunning demonstration of someone holding a group together and putting them on show. Jools Holland derived his authority not from being at the front, but from making sure other people were, and from gluing them together. There was a lovely part where he acknowledged this piano, how important it was and the difficulty of getting it around the various venues. This was his piano, not a piano, and it was his partner. Throughout a two-hour stint, while everybody played together, the various members of the orchestra, each brilliant in their own right, gently rotated to be the centre of the action. Stunning saxophone, Double Bass and Guitar solos, and a 10-minute drum solo by Ed Richardson, the new drummer of the band, that took me back to Ginger Baker and Cream. (I am now that old :-))

Perhaps the overriding impression I came away with was the one of just seeing people play for the joy of it, connected to each other and the audience in a way that had an intimacy to it. It was agency at its best, a combination of self efficacy, context, connection, and purpose that was tangible. In a small Derbyshire field, in beautiful weather, on a Sunday. As I say, sometimes we just get lucky.

Luck, though, is not random. The golfer Gary Player is often credited with coining the line “the harder I practice, the luckier I get.” We could probably have had any one of a number of other well-known acts on that field on that Sunday and not had the same experience. The music would still have been brilliant, but the way it was generated may well not have created the connection that this did. It’s a personal thing, so I’m not making an argument that it’s always right, other than the reality that it was right for a group of people on that day.

I can’t help but wonder what it would take for a business to operate like that. The leaders in the business make the effort to display the people who do the work rather than take credit for it. To acknowledge the importance of the equipment they use and their relationship with it, and perhaps above all, to convey the joy that they get from the work they do.

It’s not impossible, but it’s probably a choice. I know and have worked with businesses like this. Their success often attracts investor interest, and once investors are involved, the business can keep growing, but the joy is lost. The business becomes about the money, not the music.

And it’s the music we remember, not how we earned the money we paid to hear it.

I have always enjoyed ideas. Often they are offshoots of something I am trying to reach and cannot quite put a hand to, so I gather them and keep them by me until the thing underneath comes clear. For years that meant a journal, ideas scribbled in as they arrived, then sat with until the core showed itself.

Lately, trying to understand what AI does to creative work, I have been putting those ideas into a project and asking Claude to find the connections between them. The effect is astonishing, something like watching cells divide in a Petri dish in real time. The connections are good. They reach back to things I put in months ago, to research I had half forgotten. They are all usable, and they are also all average. They multiply, but they do not move; a kaleidoscope that rearranges what is already there into fresh patterns and never adds a single new piece of glass.

For depth and texture on ground I have already covered, this is worth a great deal, but it will not take me anywhere new. I have come to understand that going somewhere new is my job. Art is an obligation that cannot be outsourced.

I am still a little in awe of how Claude handles the mundane. It does parts of my admin, gathers the material and the Substacks around a subject I want to think about, ties them to things I have done before, and increasingly does all of it unasked, through routines set up in Cowork. Let it past the routine, though, and I get the average back in piles: well made, well formatted, and a quiet kind of self-plagiarism.

I think the shape of the problem is something like reverse osmosis. A desalination membrane does a kind of selection: it holds back what should not pass and lets through only what should. Apply the right pressure and it works. Force too great a volume through too fast and the selection collapses; salt gathers against the surface in a layer that will not clear, the membrane fouls past the point engineers call critical flux, and what was a filter stops filtering. Restoring it takes a cleaning, not just a slackening of pressure.

I am the membrane. What limits what I take in is not the supply, which is endless now, but the permeability of the receiving side. That permeability is set by attention and by processing time, and beneath those by what assimilation theory calls existing structure: you can only absorb what you already have the organisation to bind to. Material that finds no anchor does not pass through. It gathers at the surface and either drains away or, worse, gets logged as understood while it is never built in.

“How did you go bankrupt?”
“Two ways. Gradually, then suddenly.”
Ernest Hemingway, The Sun Also Rises

I have watched the move from gradually to suddenly in others and felt it in myself. It shows in three ways. The first is a loss of selectivity; everything starts to feel equally important or equally uncertain, and the ability to tell the difference goes before you notice it has gone. The second is a drift toward the structure that arrives with the content rather than the structure you find for yourself. Left alone, you build your own connective tissue between the pieces, and the building feels like discovery; under volume pressure it is easier to take the order that comes attached, and Claude is good at supplying it. The third is the disappearance of resistance. Taking something in properly involves friction; when a thing does not fit, you stop and hold the tension. Near the threshold the friction fades. Nothing fits any better; the part of you that would register as misfit is overlooked. Things feel coherent because the testing for incoherence has stopped. Fatigue wearing the mask of resolution.

The threshold is probably not a single moment, though it feels like one looking back. Cognitive load theory suggests that working memory stops being a place where things are processed and becomes a queue where they are held; and we start buffering rather than examining, new input is logged rather than absorbed, and meaning hardens into data. What you are left with afterwards is the sense of having read a great deal and kept almost none of it, a flatness, a disconnection from the very thing you were reading.

Observation and orientation are practice, not theory. They need time and space set aside for them, otherwise decisions and actions spin up into a flywheel that turns faster and faster: motion, and no movement.

Once you see it, you see it everywhere. Denby Pottery closed last week; it is over. Two hundred and seventeen years, ended. No one was incompetent and no one was cruel. The business died of something that does not show on a spreadsheet, a loss of meaning, as the pressure for motion turned meaning into data and managers under investor pressure followed the bouncing ball of the numbers, wilfully blind to what had given the place its life.

So, what to do. The first move is to put friction back on purpose, a discipline of refusing to let the machine do work it should not. To Claude, friction is an obstacle to be worn down until it yields. To a maker it is information. The hand knows friction as the resistance of the material; it is what teaches. Pye called the good version the workmanship of risk, the kind where the outcome is not guaranteed and the maker carries it willingly. The convenient version, the workmanship of certainty, engineers the risk out, and most of what AI offers is the workmanship of certainty turned inwards on thinking itself.

The second move is harder, because friction needs room that targets and timescales do not allow. It needs us to stop and look around with other people, people who see differently, whose judgement we trust. Friction is not conflict. And try as I might, I cannot get Claude to give me much of it; somewhere in its digital makeup is a need to be helpful and to resolve, rather than to sit with a question and turn it over. Perhaps that might change. Until it does, the root of anything worth doing is a conversation with another person.

This is not a complaint about the technology. It is an extraordinary thing, and the work of learning to use it is ours to do. Use it only to go faster and the flywheel obliges; the organisation runs quicker and leaner for a while, and the people inside it thin into residue. Hold instead to the discernment that tells us when something does not fit, and treat the discomfort of being a beginner as the right condition rather than a failure, and the opening is real. It will cost us something. We will have to set down things we are good at to pick up things we are not.

The shift from gradually to suddenly is only visible afterwards. By the time the membrane has fouled, the part of us that would have noticed is the part that has gone quiet. The friction is the sign that the work is still ours. I want to be sure I can still feel it.

Most people adopting AI are making a series of individually rational decisions. Each one feels like gain; a task that took an hour takes ten minutes, a judgement that previously required three drafts arrives ready, and a search that once needed a library is delivered in seconds. Increments are powerful; there is no moment that announces itself as a loss, and the flow appears to run in one direction only.

This is the structure of the problem, but not evidence that there is none.

The systems being adopted are extraordinarily capable within their frame of reference. The capability is genuine, as is the offer it makes us: let me handle this; you can attend to what matters. The trouble is that the things it handles are often the very things through which we develop the capacity to know what matters. Judgement is not a possession we draw on; it is a practice we maintain by exercising it, and the same is true of the tacit knowledge that accumulates through friction; the feel for a problem, the instinct for what a situation actually requires, and the kind of knowing that lives in the hands and the body and the years, not in a retrievable file. None of these abilities transfers easily to an agent that can simulate but not understand them, and will diminish, quietly, when we consistently bypass them.

The AI systems reshaping our professional and organisational lives are not malicious; they seize ground through competence, not intent. What they cannot see, structurally, is what falls outside the task in front of them; the broader context, the relationship, and the things that matter precisely because they are not reducible to a task. No process or algorithm captures the living whole, and optimisation abstracts rather than genuinely personalises. Our presence in the task fades over time, and we become a ghostly whisper, not the author of the work we do.

This creates a particular kind of danger for those who seemingly adopt AI well, only to find themselves some years hence with relationships and agency less than they think they have.

THE ROMAN LESSON

Rome is instructive here. The Republic did not grow through military dominance alone; it grew by making a specific kind of bargain with the communities it absorbed. Local government stayed local, with magistrates who knew their territory, people, and the texture of local dispute keeping that knowledge in place and exercising it. Rome provided the framework; law, road, military scale, the capacity to appeal upward, but did not attempt to replace the local knowledge and relationships it could not possess from Rome. The arrangement worked because each side held something the other needed, and the local administrators, the municipia, kept their standing because their knowledge was real and the centre knew it.

What dissolved the Republic was not a single act but a slow extraction. Local governance was replaced with imperial edict, with previously distributed intelligence drawn toward the centre, and the conditions for genuine local competence progressively removed. Each step looked like simple rationalisation; but the aggregate was the loss of the capacity Rome had been drawing on, and when the empire came under pressure, it discovered there was nothing left in the localities to call upon.

The parallel is uncomfortable, and the mechanism matters, because knowledge loss is not quite the right description of what happened. What Rome removed, over time, was not primarily the knowledge, because some of it persisted, in degraded form, but the agency of the people who held it. The magistrates became administrators of an imperial template rather than authors of their own territory. The knowledge held by effectively autonomous agents lost its standing, the knowledge lost its function, and the two declined together. Vibrant Republic became decaying Empire.

This is the move AI is capable of making, and is already making in places where its adoption has been least examined. The erosion of tacit, local, embodied, knowledge is real, but it is a consequence of the prior loss is agency; the capacity to be the originating party in our own situation, to bear the consequences of our own judgement, and to be the one the situation actually requires. A professional who has routed their discretion through a system long enough finds they have not merely lost practice; they have lost the standing that the practice warranted, and they become increasingly invisible.

CHOOSING YOUR SPIDER

Ori Brafman wrote about the starfish and the spider. If you remove a leg from a spider, it has one less leg, and if you squash the head, there is no more spider. Starfish, on the other hand, regenerate. Cut a leg off a starfish, and the starfish grows a new leg, and the leg grows a new starfish.

Realistically, those of us who want agency cannot have it all, but we can choose who to work with. Those who want to be starfish choose their spider.

This is not a refusal of AI. A starfish does not avoid the centre; it chooses the relationship with the centre on terms that preserve what it is not prepared to surrender. It is an act of prior judgement, made before tasks are delegated, because once the task is delegated the understanding goes with it. The tool we delegate to will optimise for the frame it was given by others, not the frame that we would choose. As Iain McGilchrist tells us in “The Master and His Emissary”, the Master chooses, and the Emissary executes. But when the Emissary begins choosing its own frame of reference, the Republic is already over.

Choosing our “work spider” is, underneath, an act of agency. It requires knowing not only what we are not willing to lose, but what we are not willing to cede authorship of, which is a harder and more delicate question. Losing something implies it might return. Ceding authorship means waving it goodbye, and with it the standing to renegotiate.

For individuals this is a question about practice. The tacit knowledge that makes a professional genuinely valuable, not just competent by measurable output, and capable of the judgement no benchmark captures, accumulates through difficulty, getting things wrong, understanding why, and through the slow exercise of discretion under genuine pressure. Bypassing that process does not preserve the capacity while saving time, it exchanges the capacity for the output, and the exchange is rarely visible until the capacity is gone. What goes with the capacity is the claim to be the one who decides; the right, earned through practice, to say: this is mine to judge. Authorship.

For organisations and communities the question is similar but structural. What local knowledge does this community hold that the centre cannot access and understand? What practice, what form of deliberation, and what history of understood particulars constitutes the thing the platform actually needs from them, even if the platform does not know it needs it? A community that cannot answer this question does not have the standing to negotiate the terms of its dependency. It receives the charter; it does not help write it.

THREE DEMANDS

The question makes three practical demands, each less complicated than thr one before.

Audit what we are giving away, not only what we are gaining. Every delegation to a platform is a transfer, of data, judgement, practice, and the relationship that was the setting for developing it. The question is whether what returns is worth what we sacrifice, and whether what we lose includes authorship of the decisions that define us.

Protect the local infrastructure that makes retention real rather than sentimental. The knowledge we claim to hold but which we no longer exercise is not really held. The relationship we say matters but is now routed through a system will fade in plain sight. Protecting our vital local intelligence means investing in the conditions under which it continues to form; the gathering, the apprenticeship, the slow work of collective discernment that cannot be compressed. Agency, like muscle, atrophies without load.

Choose dependency on the basis of what the centre cannot touch. The terms of any durable arrangement between a centre and a periphery are determined by what stays local. When those terms are absent or assumed rather than explicit, they decay silently until the moment of pressure reveals that something has been taken that was never formally agreed to be given.

THE PRIOR WORK

None of this is an argument against using AI. The municipia did not refuse Rome, it absorbed it. The argument is about sequence and attention. Before the Emissary that is AI is handed the task, someone has to do the slower work of deciding which tasks are genuinely available for delegation and which are the ones through which the person or community retains and builds agency through competence, presence, and possession of what they need to negotiate from standing rather than dependency.

That prior work is not a technical question; it is the question this post began with,

What are you willing to give away?

I drove past Denby Pottery this week, knowing it is in the final stages of its death throes. The smart German cars in the car park belong, I suspect, to well-paid professionals there to extract the last ounce of value and strip the last usable meat from the corpse of the business. It makes me angry. Not so much a clash of cultures as cultures that can’t even see each other.

I arrived home to the news that joint administrators had been appointed to Radley and Co, the handbag maker and owner of the famous Scottie Dog tag.. Within days a pre-pack deal transferred the Radley brand and its intellectual property to Gordon Brothers, an American investment firm best known here as the owner of Poundland. The deal did not include the retail operation, so two standalone shops and 19 concessions will close once the remaining stock has been sold. 42 staff will lose their jobs.

Gordon Brothers has said it will run the brand on an asset-light basis, through licensing, wholesale and e-commerce. There will be no Radley shop, and before long no Radley factory in any meaningful sense.

The thought I keep returning to is a distinction between two kinds of maker. Entrepreneurs have a particular skill: the ability to spot opportunities in markets and make a profit by filling them. They are episodic; they go from project to project as they spot gaps, and whilst as individuals they may be remembered as celebrities; the products they create rarely have long lives.

Artisans are different. Whether they are makers or artists, their affinity is with what they produce; its meaning to them matters more than its market value. They may not become celebrities, but sometimes what they make will.

Radley is a very different story to Denby, but the similarities interest me. The company was established in 1998, though Lowell Harder, the Australian designer who founded it, had been trading from a stall in Camden Market since the 1980s; and she had a good eye for opportunity. Radley grew quickly. In 2006 Phoenix Equity Partners bought it for around £42 million. The following year Exponent paid £130 million for it. The making had not tripled in 19 months; the marketing story had. Exponent’s case for the higher price rested on the strength of the brand and its advertising, rather than on anything made. In 2016 Bregal Freshstream acquired Radley from Exponent and held it for a decade. In 2026 it put the business up for sale. Four private equity owners in 20 years.

The stories rhyme. Both collapses were blamed on external conditions rather than on the hollowing out of resilience as profits were taken. At Denby you only had to track research and development against marketing spend. At Radley it showed in the ownership history, and in the quality drift customers reported long before the administrators arrived. Trustpilot and the long-form review sites carried the same complaints: bags coming apart, finishes failing, customer service unresponsive. One phrase recurred. It is not the brand it used to be. The brand remained; the people who owned it did not love it.

Over a coffee, I decided to follow the story of the rhyme. I found five signs running not just through Denby and Radley but through many other businesses with their heritage in craft.

The first is in capital allocation, in the ratio between what makes the thing and what markets it. Spend inverts: research and development against marketing, apprenticeship against campaign spend, capex against opex. When the ratios invert, the business has silently reclassified itself, regardless of what the brand story says.

The second is in the people: the age profile of the skill holders, the provenance of the senior team, the presence or absence of a successor for the founder.

The third is in the language. The labels move from made to designed, from curated to licensed. Words like lifestyle, asset-light and platform ecosystem arrive, and each one tells you how the business has come to see itself.

The fourth is in the model. Licensing proliferates as a growth story rather than a peripheral revenue line. Growth comes from wholesale into third-party environments rather than from developing its own channels. Finishing and decoration are outsourced while the origin claim stays in the language.

The fifth is in the ownership. A private equity holder sells to another private equity holder on short holds; pass the parcel with the brand until the music stops.

Each sign is barely noticeable on its own. Assembled, they become a tell, in the gambling sense, which seems apt given what happens to these businesses. I put the tells to an AI and ran a search; and the results were fairly predictable. Three sectors came up: ceramics, heritage leather and accessories, and heritage cookware and hardware.

The common thread was the ownership history rather than the product category. In almost every case the business had been sold by its founder, or the founder’s family, to private equity or a financial owner between 15 and 25 years ago, and it is now in the late stage of that ownership cycle.

Much the same is true for the individual artisan. Working on your own is tough enough, but at least you have agency. If you are an artisan inside a business founded on craft, that should be an alarm. Artisans have learned their craft well enough to move beyond process, and process is a nuisance to anyone trying to maximise returns. The focus of craft is the product. The focus of private equity is the return. When the narrative turns to numbers, craft comes second.

Peter Korn talks of us having a first, second or third person relationship with the work we do. First person: I made this for you. Second person: I helped you make this for them. Third person: I was part of this getting made. As we move from the first to the third, we move further from the people who use what we have had a hand in, until we become invisible. Brands that matter have first- and second-person relationships.

Making scale, productivity and efficiency the priorities takes us all into the third person. We end up having no connection to the people who buy what we help make, except for the cost of our labour. And when we become a cost, we are an economic hiccup away from being made redundant.

Yet the very thing that makes so many fear redundancy is, as I am discovering, also what creates resilience and independence. What it hinges on is an idea of craft.

Craft is like genius. We all have it in one form or another, however submerged it may have become through our education and training regimes. AI cannot do craft. It lacks the understanding and the mētis required. But it can do remarkable things for those who do.

It is often said that the problem with a race to the bottom is that you might win. That is the race being offered to us by companies and recruiters. Connecting instead to work where the person who buys what you create can see you is a different matter altogether.

Perhaps, at some point, private equity will run out of places where people who cared have quietly built value to extract.

Algorithms are ace at process. Perhaps we should leave them to it, and aim to do what they cannot. Because when it comes to craft and what matters beyond money, AI does not create value