I have two jobs that don’t fit together cleanly. Some weeks they actively conflict.

One job is advising agency owners. Mid-market firms mostly, ten to eighty people, some are ones who’ve found a way to do good work but haven’t solved the problem of scaling without losing quality. I help them figure out how to grow the business while keeping the work interesting and the margins real. That’s OMG Center. That’s where I spend a lot of oxygen.

The other job is selling them a tool that automates the work their account managers do for a living.

The geometry doesn’t work. I can feel it when I’m in advisory mode with a client, talking about headcount decisions and margins and what work actually moves the needle, and then I switch into sales mode and I’m showing them how Orca can do that work. The looks I get are complicated.

Let me describe what this actually looks like in practice.

The advisory conversation

The conversation with an agency owner usually starts with a specific pressure. They’ve got a team that’s growing. Revenue is growing. But they can’t seem to hire fast enough or good enough to keep up. The economics are getting tight. They’re spending too much time on client management. Their team is spending too much time on admin. Some of the junior people they hired are undershooting on output. They’re wondering whether they’ve hit their natural ceiling, or whether they’re just not managing the operation efficiently.

This is where I usually start asking questions about what’s actually taking time. Two to three days a week on meeting prep. Another day on follow-up emails and CRM management. Hours every week on inbox triage and making sure nothing falls through the cracks. Hours on research intelligence and competitive briefings. Account managers spending a fifth to a quarter of their time on social media and content updates.

That’s basically the role. Account management is orchestration work. Knowing what happened in the last meeting, what the client’s actually worried about, what they said they wanted versus what they actually need, what the competitive context is, and then bringing all of that together into a coherent strategy recommendation or a status update.

So I ask them: what do you think that orchestration work is actually worth to your business?

They usually say some version of: not as much as we’re paying for it. Or: most of it could be automated but we need someone there who knows the client. Or: they’re good at this part (pointing to strategy or relationship management) but we’re paying for them to do the other part too.

This is where I lean in and say something like: your business would get better if you had fewer people doing orchestration and more people doing the stuff that only they can do. Strategy. Real relationship management. Creative direction. The work that actually moves the needle. But you need different people for the orchestration part. You need tools.

And then they usually say: yeah, we know, we’ve tried some stuff, it’s messy, and also aren’t the tools going to put people out of work. Which is fair. That’s the real question.

What I’m advising now

Here’s what I tell them these days. And this is where the two jobs collide in a way that matters.

Build headcount for what can’t be automated. Let Orca do the orchestration. Measure the difference in what your account managers can actually produce when they’re not doing coordination work all day. Then restructure around that output.

That’s the honest advice. Not “keep your current structure and add a tool.” Not “add tools but keep the headcount.” But actually: the work that AI can do, let it do. Free up the humans for the work only they can do. Measure the result. Restructure accordingly.

Because if you don’t, your competitor will. And then you’re paying the same cost for orchestration work that they’ve automated away. You’re losing margin. You’re losing the ability to compete.

When I say this to a client, they usually nod. They know it’s true. They’ve probably already started running the numbers. And then they ask: so what do we do with the people who are only doing orchestration work? And I tell them the truth: that’s the hard part. That’s the part no advisory can solve for you. That’s the part you have to handle as a leader. But the answer isn’t to pretend the tools won’t work. It’s to manage the transition with eyes open.

The conflict visible

Here’s where it gets uncomfortable.

Some of the people I’m advising are also in conversations with me about Orca. I show them the tool. They see what it does. I show them what it costs: two hundred and fifty pounds a month. I show them the math: if one of their account managers spends a quarter to a third of their time on orchestration, and a tool costs that little, then the economics are obvious.

And I can see them work through the reasoning. They came to me for advisory. I told them to automate orchestration. Now I’m selling them the thing that does it. The two roles have collapsed into the same transaction.

I don’t pretend this doesn’t matter. I lay it out. I tell them: I’m advising you to do this, and I also have a commercial interest in you buying this from me. You should price that into how you interpret what I’m saying. But I also think the advice is real. The automation is real. And the longer you wait to do this, the more margin you’re leaving on the table while your competitors figure it out.

Some of them buy. Some of them don’t. The ones who don’t usually go and build something internal or they use a different tool. That’s fine. The structure of my business doesn’t require every advisory client to buy Orca. What it does require is that I’m honest about the conflict.

What’s actually different about this

I could use the language most tech companies use: “amplifying your team”, “helping your people be more productive”. That language is comforting. It’s also dishonest.

Orca doesn’t amplify the orchestration work. It absorbs it. When a meeting brief that used to require someone half a day now takes a minute fully automated, the work isn’t being augmented. It’s being replaced.

The question is what you do about it as a leader. You acknowledge what’s happening, measure the impact, and structure your business around the new reality instead of pretending the old one still exists.

For agencies, that means: more headcount in strategic roles, less in coordination roles. Account managers doing account work, not AM admin. Planners planning, not planning then turning plans into decks then writing them down for the CRM.

The hard part isn’t the tool. The hard part is the conversation with the people whose jobs were mostly orchestration. That’s where the real work is. That’s where the leadership maturity shows up or doesn’t.

Why I can hold both roles

Someone asked me recently whether I felt conflicted about selling a tool that displaces people while advising the owners who might buy it. I said: only if I pretend the tool wouldn’t work, or the displacement isn’t real, or the maths are different than they are.

But I don’t pretend any of those things. I name the displacement. I explain the maths. I advise people to handle the transition with eyes open. Then I offer them a tool to do the work better and cheaper.

That’s not corruption of the advisory relationship. That’s completion of it. I’m advising them to do something that’s structurally necessary, and offering them a tool that makes it feasible.

Do I benefit when they buy Orca? Yes. Does that mean the advice is wrong? No. It means the advice happens to align with my commercial interests, which is worth disclosing, which I do.

I also stay visible. I don’t hide in the business model. I show up in demos. I’m in the advisory conversations. I’m subject to the feedback about what actually happens when these transitions occur. If the displacement is worse than I’m predicting, I see it. I can’t pretend it didn’t happen.

In 2019, the career hierarchy still made intuitive sense. Knowledge work paid more than trades. Knowledge work was more prestigious. The progression was clear: get a degree, move into professional services, climb to the top, retire comfortably. That hierarchy was built on the assumption that knowledge work would continue to be scarce and protected. That assumption died in the last seven years.

I’ve been running a garage alongside everything else. Windmill Garage, forty years old, family business. Four full-time staff, mixed workload, conventional schedule, the kind of place that fixes what’s broken rather than inventing the next thing. Watching that business function while watching AI automation compress knowledge work, I’m struck by how the old prestige hierarchy has become a liability.

Here is what I’ve observed.

A plumber in 2026 earns roughly 60,000 pounds a year. That’s a working income, solid middle class, reasonable mortgage. An electrician slightly higher. A carpenter at the craft end of the spectrum, if they’re good and established, can exceed that. These are people with physical skills, established client relationships, the ability to diagnose problems in unpredictable environments, the capacity to manage clients who are stressed because something is broken in their home.

All of those attributes are precisely what AI is bad at replacing. A pipe is different every time. An electrical fault is different every time. The environment is physical, unpredictable, contextual. The client isn’t interacting with a system, they’re interacting with a person they trust to solve a problem they don’t understand. That trust, that judgement, that physical presence, are the things machines can’t replicate.

Compare that to a knowledge worker trajectory. The graduate enters the market earning 28,000 to 32,000. They climb over ten years to 70,000 to 80,000. They then face the problem that their entire career has been orchestrating information flows, synthesising data, routing decisions, managing status updates. All of those tasks are being automated by tools that cost a fraction of their salary. The ladder they climbed no longer leads anywhere. The next rung is missing.

The snobbery around trades is structural to how UK education and careers advice works. School pushes A-Levels and university. University is the default aspiration. Anything else is framed as an alternative, a backup option, the thing you do if you don’t get the grades. Never as the better choice. Never as the safer choice. Never as the route that gives you more genuine security than the professional credential path.

That framing is poisoning decision-making at exactly the moment when the decision matters most.

I watch this with the Enterprise Skills work. The platform measures eight capabilities: commercial awareness, decision-making, problem solving, financial literacy, adaptability, data analysis, team collaboration, and leadership. We built it because capability-based credentialing is becoming more important than qualification-based credentialing. Credentials are becoming cheaper, more commodified, easier to fake. The ability to actually do something is harder to fake and harder to automate.

But when we pitch to careers leads, the conversation keeps coming back to the same point. The school system is built around qualifications. The parents want their children to “do well,” which means getting good grades and going to university. A parent meeting their child’s careers adviser and hearing “your child is skilled at problem-solving and decision-making under uncertainty, so a trade route might actually be the resilient choice” is not what they expect to hear. They expect to hear “your child should try for accountancy,” because accountancy has a name, an institutional structure, a path they understand.

That institutional structure is currently being demolished by AI. The Big Four firms deployed AI systems in 2024 and 2025 that now handle audit work, tax compliance, and financial analysis faster than junior accountants. The entry-level route into accountancy, three years grinding through ACCA exams while doing the spreadsheet work that nobody wants to do, is the exact pathway most vulnerable to automation. But it’s still the pathway parents point at when they’re trying to give their children good advice.

The honest answer is harder to give. “Your child is good at problem-solving and could make 50,000 to 70,000 as a skilled tradesperson by 30, with higher security and lower credential pressure than the professional route, which is increasingly automated and precarious.” That’s a true answer. It’s also a conversation-stopper in schools where the parent body is predominantly professional class and university-educated. The careers lead who starts having that conversation consistently will face pushback.

And yet it’s the conversation that actually matters now.

Here is the thing that gets overlooked. The tradesperson with 40,000 to 60,000 in annual income and no student debt, building equity in a house, developing a specialist skill set and client relationships that are difficult to replace, has more security than the professional earning 75,000 with 45,000 in student debt, a mortgage that assumes dual income, and a skill set that is being systematically automated.

The tradesperson also has another advantage that nobody talks about: option value. If you’re a skilled electrician, you can work for a firm, go independent, specialise in a particular sector, run a training business, teach apprentices, move into site management. The number of paths available to you doesn’t close down at any point. If you’re a marketing director, the number of paths available to you is shrinking. The roles that marketing directors used to move up into are being compressed. The lateral moves into other knowledge work are increasingly competed for by people doing the same calculation you’re doing.

For a tradesperson, there is also the land question. I’m saying this plainly because it’s relevant. A plumber who establishes themselves in their twenties, who buys a house in their area in the early 2030s when they have equity from earnings and the market is still adjusting, who then runs that business for thirty years, who has client relationships that generate consistent work and stable income, that person will have accumulated land value. Not because they’re clever about property. Just because they were generating consistent income in a role that pays well enough to own property.

The professional who spent their twenties and early thirties competing for increasingly precarious roles, paying rent in London or another major city because that’s where the work is, taking contract work and freelance roles because the permanent jobs disappeared, building a mortgage that requires dual income or doesn’t happen at all, that person has a different trajectory. Not because they’re less capable. Because the structure of their market changed while they were climbing it.

I’m not suggesting everyone should become a plumber. I’m suggesting the careers hierarchy that makes that suggestion sound absurd is no longer connected to reality. The labour market changed. The advice didn’t.

Here’s what I see when I look at the trades through both the Windmill Garage lens and the Enterprise Skills lens. Physical presence. Problem-solving in unpredictable environments. Trusted relationships with clients who are stressed. Consistent income. No credential inflation. Low student debt. Portable skills. Option value. The ability to go independent or stay employed, both of which work. And increasing scarcity of people willing to enter the field because the snobbery persists.

The kids who go into plumbing in 2028 are not the first-choice cohort. They’re the people who were willing to do something that their school and their parents treated as a backup option. They will therefore be selected from a smaller, more self-directed population. They will have less competition because fewer people are choosing the route. And they will be in a field where client demand is high, supply is constrained, and the work is becoming more valuable, not less, as knowledge work struggles.

By 2035, the plumber who made that choice in 2028, who now has seven years of income, established client relationships, option value in multiple directions, and real property owned, will look like they made a smarter decision than the professional who has been job-switching, freelancing, and trying to adapt to a labour market that keeps moving underneath them. Not because trading is inherently superior to knowledge work. Because the hierarchy of prestige no longer matches the hierarchy of security, and the advice system still treats prestige as if it matters more.

The snobbery will age badly. In ten years, nobody will point at a qualified tradesperson with 60,000 in income, 30,000 in equity, and secure prospects as the cautionary tale. They’ll point at the professional whose credentials turned out to mean very little when the work got automated. But that recognition will come too late for the cohort making the choice right now, who are still being told that trades are the backup option.

The right advice for 2026 is this: if you have problem-solving skills, adaptability, the ability to work with your hands and with your mind, and you’re trying to decide which path gives you more security, the smart money is on the trades. Not because knowledge work is bad. Because the market for knowledge work is broken and the market for skilled trades is working. That will be obviously true in five years. You might as well act on it now, when you still have time to develop the skill.

He got in touch because he’s trying to figure out what to do with AI. Said he wasn’t sure if I’d remember him.

I did.

The bit that actually mattered came in his second message, after he’d had a look at my site. Paraphrasing, but only just.

“I started paying an agency £800 a month for SEO support. And with everyone telling me how powerful AI is, I’m thinking it might be better to use that for the SEO and the other tasks.”

Read that again. That’s a small business owner, unprompted, working out in real time that the thing he’s been paying an agency to do is probably the thing AI can do for less. He isn’t a tech founder, doesn’t read AI newsletters, doesn’t sit at SaaS events with a lanyard on. He runs an operation, gets sold to constantly, has heard the word “AI” used so many times he’s started to wonder why he’s still writing the £800 cheque.

He’s doing the maths. That’s the part I keep noticing.

If I’m honest, most of what £800 a month at an agency is buying is orchestration. Keyword research. Content briefs. A bit of technical audit work. Reporting back. Maybe some outreach if the retainer stretches that far. The actual work is using tools that already exist, running them against the client’s site, interpreting the output, and turning that into something the client can action.

The agency isn’t doing magic. They’re doing competent translation. They take a set of tools the client could in principle learn but doesn’t have the time or appetite for, and turn the output into language and tasks he can understand. They’re the orchestration layer between him and the systems.

And that orchestration layer is what AI does first.

Here’s the awkward part. A lot of what an agency in the £500 to £2,000 retainer bracket does can already be done by a small operator with the right setup. Keyword research, content clustering, brief writing, technical SEO audits, most of the on-page work. All of it has a credible AI version now. Not as a single magic button, more like a stack of capable tools that, wired up properly, do roughly what the agency is doing every week. Faster. Cheaper. Fewer meetings.

That’s the bit “everyone telling me how powerful AI is” is picking up on. The general sense in the air. Most people who aren’t paid to deny it can feel it.

The LinkedIn DM doesn’t say what the right answer looks like. It just asks the right question.

The agency model assumed the client couldn’t or wouldn’t do the underlying work. So the agency did it for him and charged the retainer. AI changes the first half of that assumption, which means that he probably could now. He still won’t, because he’s running a business and he’s not going to spend his Tuesday evenings learning how to set up a content production pipeline. That’s why he was paying the £800 in the first place.

So the question isn’t whether AI can do the work. It’s who builds and runs the AI for him.

If the answer is the agency, the agency’s new value proposition is operating AI on his behalf at a fraction of the price the old retainer carried. Which a lot of agencies aren’t structurally able to do, because their cost base is people and meetings.

If the answer is him, with a one-off setup from someone who can actually build it, the maths look different. A few thousand to build, a couple of hundred a month to run and maintain, no monthly orchestration call, no monthly invoice for the bit he can now do himself with a system someone else has wired together.

I’ll be straight about where I sit in this. I run a managed AI service for founders, and I do AI advisory for businesses trying to work out where this actually fits their operation. More and more of the work I’m being asked to do is exactly this: someone who’s been paying for orchestration via human services has clocked that the orchestration is now buildable. They want someone to build it, configure it around how they actually work, then leave them to run it.

That’s not a great long-term position for the agency that used to charge £800 a month. It’s a better position for the customer.

Which is the actual pattern, and the part most agency owners I speak to are still talking themselves out of seeing. AI doesn’t kill the work itself. It kills the layer in the middle, the bit that was charging to manage the work for someone. The strategy still matters. The judgement still matters. Knowing what to build, what to leave alone, what not to bother with, that still matters. The orchestration in the middle, the bit that was £800 a month, that’s the bit that softens first.

The thing I keep noticing is how unprompted this was. He didn’t read it on Twitter. He hasn’t been at an AI conference. He’s running his business, he’s heard enough times that AI is powerful, he’s looked at his retainer, and he’s asked the obvious question.

That’s the bit that should worry every agency owner in the sub-£2k retainer bracket. The customer is doing the maths. He isn’t waiting for the trend piece, isn’t waiting for the case study. He’s looked at the invoice, he’s looked at the noise about AI, and he’s drawn a straight line between the two.

He doesn’t know yet exactly what to build. That’s where someone like me comes in. But he knows the £800 a month doesn’t make sense any more.

And he’s right.

Subscribe now

Nobody can tell you which jobs will exist in 2035. But you can position yourself to be useful regardless of which scenario arrives.

The honest audit tells you where you stand. This tells you what to do next. Not because I can predict the future with any precision, but because there are moves that work across multiple scenarios, and moves that only work if one specific future arrives. The first kind are worth making now. The second kind are a gamble.

The scenarios are useful here. Managed Transition: governments implement support, new social contracts emerge. Neo-Feudalism: capital captures gains, most people subsist on fragments. Fragmentation: no coherent response, regions diverge. Black Swan: something nobody modelled. The current probabilities sit at roughly 28% Managed Transition, 48% Neo-Feudalism, 19% Fragmentation, and 5% Black Swan. The direction matters more than the precision.

If you optimise for Managed Transition, which is the assumption baked into most career advice, and Neo-Feudalism arrives instead, you’re caught with no safety net, no independent income, and a new qualification in a field that’s already overcrowded because everyone took the same retraining programme. If you position for resilience across scenarios, you might not be optimally placed for any single outcome, but you won’t be catastrophically exposed to any of them either. That’s the trade-off. It’s the right one.

The no-regrets moves, the ones that help regardless of which scenario materialises, are clear: build financial reserves, diversify income, strengthen trust networks, develop capabilities that survive automation. Everything else is scenario-dependent. Think of it like packing for a trip where nobody will tell you the weather. You can’t build a wardrobe for one pattern. But you can make sure you’ve got a waterproof, some layers, and decent boots. The specific outfit matters less than having options.

That’s positioning. Options, not answers.

Here’s what that looks like in practice if you have runway: time, resources, employability, options.

Build owned assets. Something that generates value without requiring you to sell hours. Intellectual property. A product. A course. A system. A business with recurring revenue. A property that produces income. The distinction matters because the fundamental vulnerability of professional services work is that it sells time. You have a finite number of hours. You sell them at a rate the market determines. When AI compresses the time required to produce the same output, the rate collapses. Owned assets don’t have this vulnerability. A product sells while you sleep. IP compounds. A recurring revenue business scales without proportional time input. The investment is front-loaded: months or years of work to create something that then generates value independently.

MIT Sloan’s research on founder age says the opposite of what most people assume. The highest entrepreneurial success rates come from middle-aged founders, not young ones. The person with twenty years of domain expertise, an established network, and hard-won judgment about what works and what doesn’t is better positioned to build something than the twenty-five-year-old with energy and no context. If you’re forty-five and thinking it’s too late to build something of your own, the data says you’re wrong. You’re in the optimal window.

Acquire skills in AI augmentation. The people who survive in knowledge work are the ones who use AI tools to multiply their output, not the ones who pretend the tools don’t exist. This is different from “learn to code,” which was never an escape route for most people. AI augmentation means understanding how to use the tools reshaping your industry: how to prompt effectively, how to validate AI output, how to integrate automated workflows into your existing practice. The person who uses AI tools to produce in two hours what previously took two days hasn’t automated themselves out of a job. They’ve made themselves five times more productive. Right now, AI augmentation skills are relatively rare in most workplaces. Within two or three years, they’ll be table stakes. The advantage goes to early adopters.

Develop trust networks. When credentials commoditise, when anyone can generate a polished CV, a strategy deck, or a thought leadership article in minutes, the scarce resource becomes trust. Granovetter’s research from 1973 still holds, verified by MIT in 2022 across LinkedIn data: weak ties, casual acquaintances and distant connections, are more valuable than close friends for employment opportunities, access to information, and career advancement. Your close friends know the same things you know. Your weak ties connect you to networks outside your own. Novel information. Novel opportunities.

Spend less time deepening relationships with people like you and more time building light connections with people who aren’t. The plumber who knows what’s happening in the trades market. The teacher who understands what young people are actually learning. The NHS administrator who knows which healthcare roles are expanding. Twenty minutes of coffee with someone outside your industry is worth more than an hour at a networking event full of people from your industry.

Build and maintain at least twenty professional relationships across different industries. Actual relationships where either party would take a phone call. That takes time, and the payoff is asymmetric: an introduction, a tip, an opportunity you wouldn’t have known about otherwise, arriving months or years after the coffee. Start now. The network you’ll need in a crisis is one you should have been building before the crisis.

Position as navigator. During periods of transition, the people who understand the change become more valuable, not less. If you work in professional services or consulting, the old value proposition was helping someone do marketing better, or strategy, or operations, or finance. That proposition is losing value because AI does it faster and cheaper. The new one: helping them understand what’s changing and position for it. The shift is from optimisation to navigation. From incremental improvement to fundamental repositioning. This isn’t rebranding. It requires actually understanding the displacement dynamics, the scenarios, the timeline, the risks. Most people in advisory roles don’t have this understanding. If you do the work, you become the person people call when they’re scared and uncertain.

Reduce financial vulnerability while you can. Least interesting advice. Most important. Reduce fixed costs. Pay down debt. Build reserves. Every pound of fixed cost you remove from your monthly expenses is a pound of runway added to your transition. The target is six months of expenses in liquid, accessible savings. Not stocks. Not property equity. Not a pension. In a savings account you can draw on within three business days. If six months feels impossible, aim for three. If three feels impossible, aim for one. Any buffer beats no buffer. The difference between making a decision from safety and making one from desperation is not a matter of degree. It’s completely different quality of decision.

For those without runway, living month to month in a role that’s already contracting: assess clearly and quickly. If your audit says your role doesn’t survive five years, waiting is expensive. Begin the transition while still employed. Not after redundancy. Not after the restructuring. While you still have income, a title, a professional identity, and the bargaining power that employment provides. Your single greatest advantage right now is that you’re not yet desperate. Desperation is expensive.

Across all positions, five principles apply. First: diversify income. Single source of income is a single point of failure. Even a small secondary stream, a few hundred pounds a month from freelancing, tutoring, consulting, or a side project, transforms your risk profile from binary to graduated. Second: build trust relationships across sectors, not just within your industry. Third: reduce fixed costs. Every pound of fixed cost removed is a pound of flexibility gained. Fourth: develop capabilities that survive automation. The eight capabilities from Enterprise Skills, commercial awareness, decision-making, problem solving, financial literacy, adaptability, data analysis, team collaboration, leadership, are what employers value and AI cannot replicate. Fifth: act now. The window for positioning is open but closing.

There is no version of this where waiting is the optimal strategy.

The positioning work sounds abstract until someone actually does it out loud, with the numbers written down. Then it becomes tactical. Not romantic. Not optional. Just necessary.

You don’t need to know which scenario lands to make these moves. You just need to be in the position where you’re useful when it does.

I want to walk through why, because the exchange contains something most public discourse on this subject is missing.

The post itself was a personal arc. A wife who watched her mechanical-engineer husband quit a stable job in 2021 to learn AI from YouTube, mocked him for two years, then converted in 2024 when he fixed a backend bug she’d been stuck on for two days in seven minutes. They now run an AI-learning newsletter together. He’s top one per cent on Upwork. Six and a half thousand hours logged.

The personal story is real. The arc is true. Plenty of people have had a version of it.

The thing that stuck with me was a different layer. The post wasn’t really about her husband. It was a case for individual reskilling as the response to AI displacement. Watch the YouTube videos. Pick up the tools. Get on the right side of history.

Which is fine, until you ask what happens at the level of the whole labour market.

The exchange

I left a comment. Generous about the personal arc, careful about the bit that doesn’t translate. The point I made was the aggregation point. One engineer pivots well and the headline writes itself. The maths underneath doesn’t get told. If AI makes the remaining engineers five times more productive, the firm doesn’t need five times the output. It needs fewer engineers. The labour market story is about the chairs that quietly disappear while a handful of people learn to build new ones.

She replied:

I think some current positions will need fewer people, yes. But new positions, workflows, and industries will appear as well, and people will shift into those areas over time. A system that works against the majority of society is not sustainable long term, and I believe even the companies building these technologies understand that too.

This is the standard optimist position, stated cleanly. New roles will appear, people will shift over time, the system can’t work against the majority indefinitely.

I came back with a question. Every previous transition had at least one absorption pillar working. New task creation, demand expansion, or skill transferability. Which one is doing the work this time?

She replied:

Actually, AI still makes obvious mistakes, so every industry needs a human in the loop. Fewer people in the old roles, sure. But the shift always creates new ones, we just can’t name them yet. So I might be too hopeful, but this is what I am thinking.

That last sentence is the most useful thing anyone has said to me about this in months.

The thing nobody noticed

Here’s what’s worth noticing, and what most of the thread missed.

The second sentence concedes the case.

“Fewer people in the old roles, sure” is the displacement argument. That’s the whole first half of the book I’ve been writing for the last year. It’s the thing I expected an optimist to push back on, and they didn’t. Not because they’re inattentive. Because at this point, displacement isn’t really in dispute. The statistics are too visible. Microsoft fifteen thousand cuts alongside agent rollout. Tech sector cuts at a quarter of a million in 2025. Anthropic’s data showing programmers, customer service and paralegals at seventy-plus per cent of core tasks handled by AI.

You don’t argue with that any more. You absorb it.

What you do is move the load to the next sentence. The one that says: but the shift always creates new ones. That’s where the optimism actually lives now. Not in denying displacement. In trusting that whatever gets displaced gets replaced.

So the question becomes: what’s holding that trust up?

What’s holding the future up

Every previous economic transition came with at least one absorption mechanism doing the heavy lifting.

New task creation. The factory worker becomes the service worker becomes the knowledge worker. Each transition added new categories of paid activity, and the categories were big enough to absorb the people the old work had stopped needing.

Demand expansion. A more productive economy generates new appetites, new markets, new things to spend money on. Cheaper bread doesn’t mean fewer bakers. It means people buy other things, and the people not baking go and make those.

Skill transferability. A weaver could become a millworker. A typist could become a data entry clerk. A bookkeeper could become an analyst. The skills you already had ported, partially, into the new shape of the work.

Acemoglu and Johnson’s Power and Progress is the long view on this. The Luddites lost not because they were wrong about looms (they weren’t) but because the factory economy that came after ended up bigger than the cottage economy that came before. New tasks. New demand. Transferable skills. At least one of the three was always working.

This time, all three are breaking together.

New task creation is constrained because the same tool that absorbs the old tasks also performs most of the new ones the moment they’re defined. Demand expansion is uncertain because the productivity gains are concentrating, not distributing. Skill transferability is the worst-affected pillar of the three. The skills that survive (judgement, taste, complex relationships, genuine creative direction) take years to develop and don’t show up neatly on a course playlist.

This is the question the book exists to answer. Not whether displacement is real. We’re past that. Whether the absorption mechanism still works.

The honest answer is: not on the evidence so far.

The honest sentence

Which brings me back to her last line.

“So I might be too hopeful, but this is what I am thinking.”

I don’t think she’s a fool. I think she’s an honest person stating, out loud, the position most people are operating on. The optimism isn’t analysis. It’s a hope, plainly named as such, by someone with the self-awareness to label it.

That’s rare. Most public discourse on AI and jobs is the same hope dressed up as an argument. Tech executives have a financial reason to repeat it. Politicians have a political reason. Consultancies have a fee-shaped reason. Wrap the hope in a chart and call it a forecast.

She didn’t do that. She said: this is what I am thinking, and it might be wrong.

That’s the ground floor of the whole conversation. Most of public discourse hasn’t reached that ground floor yet.

Where this lands

I’m not writing this to win an argument with someone on Substack. I’m writing it because the position she stated is the position holding most household, corporate and policy thinking together right now. We’ll figure it out. The shift always creates new ones. We just can’t name them yet.

That’s hope. Hope is human, and hope is fine.

Hope is not a plan. It’s not a household financial strategy if you’ve got a mortgage and a knowledge-work income. It’s not a labour market policy. It’s not a corporate workforce strategy. And it’s not what you tell your kids when they ask which subjects to take.

You don’t need certainty about what comes next. You need a position that survives the optimism being wrong.

That’s what the book is about. The Next Rung is out later this year. The Substack is where I work it out in public. If that’s useful, you know what to do.

One thing to declare, by way of footnote.

I run a software business that automates the orchestration roles I’m telling you are most exposed. I am, literally, building the loom. I am also the one writing about what looms cost.

Some readers find that contradictory. I find it the only honest place to write from. The people building this should be the ones taking the cost of it seriously. Most of them aren’t.

Subscribe now

The labour displacement from AI is a story about time. How fast the adoption happens. How long the transition takes. How many people have runway left to adapt before the ladder collapses. The Strait of Hormuz closure in late February 2026 changed the timeline entirely.

On 28 February, the United States and Israel conducted military strikes on Iran. Iran’s Supreme Leader Ali Khamenei was killed. In response, Iran launched retaliatory attacks on US military bases and Gulf states. The Islamic Revolutionary Guard Corps issued warnings that halted vessel traffic through the Strait. Twenty million barrels of oil a day normally transit that strait, roughly 20 per cent of global seaborne oil trade. Another 20 per cent of the world’s liquefied natural gas moves through the same channel.

Within days, the closure became the largest disruption to global energy supply since the 1973 embargo. Brent crude surged past $100 per barrel on 8 March for the first time in four years, and continued climbing to $126 at its peak. The UK wholesale gas market reacted within hours.

Three days after the initial strikes, UK wholesale gas prices spiked noticeably. The April price cap had already been locked in by Ofgem’s calculation window, so households didn’t feel the shock immediately. But the July cap, set in mid-June based on the months between mid-April and mid-May, will catch the new market prices. Expect it to increase substantially.

Here is what that means for the timeline.

The story of AI displacement in knowledge work runs on two parallel tracks. One is technical adoption: how fast enterprises deploy AI tools, integrate them into workflows, and observe the headcount compression that follows. This track moves reasonably slowly. Enterprise adoption of any new tool takes time. There is organisational inertia, change management friction, the lag between decision and implementation. The technically possible and the actually happening diverge significantly. Most estimates put critical mass adoption around 2027 to 2028, with the visible headcount compression starting seriously in 2028 to 2029.

The other track is financial pressure. When energy costs spike, when interest rates respond to inflation, when client revenues contract due to economic shock, CFOs accelerate decisions that were marginal before. The investment in the AI tool that used to have a three-year payback cycle now has a twelve-month one. The consultant recommending a hiring freeze gets taken seriously instead of dismissed as pessimistic. The restructuring that was under discussion becomes urgent.

The Strait closure created a genuine energy emergency. It also created artificial urgency around any efficiency decision that could be justified to a board as a response to that emergency. “We’re restructuring to protect margins in a high-energy-cost environment,” carries more weight than “we’re restructuring because the AI is good enough now.” For at least the next five months, until roughly October, there is political cover for decisions that would otherwise require more justification.

This window is real. Enterprises that were sitting on AI implementations, waiting for a better moment, now have an actual external pressure point to justify the acceleration. The cost-benefit calculation that was marginal becomes obviously favourable. The redundancy consultations that were theoretical become necessary. The contractor squeeze-outs that were being debated become executable.

I want to be specific about what this means for specific types of workers.

Orchestration roles are most exposed. The roles whose primary function is synthesising data, routing decisions, reporting on status, coordinating between teams and systems. Roughly 1.5 million to 2 million of these roles in the UK are at serious risk over the next eight to ten years under a normal adoption timeline. Under an accelerated timeline, compressed into eighteen months instead of five years, the math changes entirely. The people in those roles don’t have the runway they thought they had. The lateral moves into slightly different roles, the upskilling opportunities, the staged progression they were counting on, all of those now have to be compressed into months instead of years.

Entry-level roles face similar acceleration. The graduate cohort starting in September 2026 enters a market where the hiring freezes that were meant to be temporary have hardened into permanent structural reductions. The contractor market, which was meant to be a stepping stone, now has permanent reduced demand. The internship pipelines that were supposed to funnel graduates into junior analyst roles are being recut to funnel them directly into AI tool supervision or nowhere at all.

But the energy shock does create one opportunity that matters.

Anything that requires cheap energy becomes more valuable in the next five months. Data centres are energy-intensive. Training large language models is energy-intensive. Running inference at scale requires power. When the cost of energy spikes, the investment in building infrastructure to run that training and inference becomes more economically defensible. The ventures that were waiting for energy costs to stabilise, betting that the next generation of chips would be more efficient, now get accelerated funding because the current cost structure makes their business case viable immediately.

This favours skill in AI systems, infrastructure engineering, capability building, and anything that sits in the stack between the model and the user. It does not favour the 1.5 million orchestrators watching their roles compress into uncertainty.

The five-month window is crucial for a specific reason. Enterprise board meetings and budget cycles hit in autumn. The companies that have accelerated their decisions and implementations through the summer now have data to report, case studies to justify, confidence in the numbers. The companies that didn’t move during the emergency window will face a different conversation come October. “We didn’t need to restructure during the crisis” is not the same as “we don’t need to restructure.” But it does mean they’re restructuring from a position of having missed the window, which changes the conversation.

For workers in high-exposure roles, the honest assessment is that the personal runway is shorter than it looked in January. If you have one year to transition, the five-month emergency window compresses it to an effective eight months. If you have eighteen months, it compresses to thirteen. The decision points that were meant to arrive in autumn 2027 are arriving now, in the summer of 2026.

The energy shock is real. Its duration is uncertain. The US and Israel and Iran are engaged in a serious military conflict with no clear off-ramp. It could be resolved by September. It could persist through winter. It could escalate.

But even if it resolves tomorrow, the decisions made under the pressure of the emergency don’t reverse. A headcount cut justified by an energy emergency doesn’t get reversed when energy prices normalise. A restructuring accelerated by the crisis gets treated as the new baseline, not as something that was temporary. The technical infrastructure built in response to high energy costs stays in place even if the energy costs fall. The contractors who get squeezed out don’t automatically get hired back.

The five-month emergency created a legitimate reason to do what many organisations wanted to do anyway but hadn’t quite justified to themselves. That reason now exists. The decisions flow from it. And by October, when the energy situation has hopefully improved, the labour market will have already shifted into a different shape.

Here is what I would tell anyone in an orchestration role right now. You have from now until mid-September to make a positioning move. Not a panic move. A deliberate move. Into something that either sits below the orchestration layer (hands-on, trades-level, requires your physical presence) or above it (genuine creative vision, strategy that reframes problems, leadership that develops other humans). The middle is being compressed by an emergency that will probably last five months and creates decisions that will persist for years after that.

The IEA forecast global oil demand growth of 640 thousand barrels per day in 2026 as of March, having revised it down repeatedly through the year as economic pressure mounted. The Strait closure is now the dominant variable in that forecast. If it persists through summer, demand destruction accelerates, prices eventually normalise, but not before the restructuring has happened. If it resolves by late May, there is a brief normality before the July price cap hits UK households. Either way, the window is five months. The decisions made within it are permanent. The people affected by those decisions need to understand that the timeline they were operating on no longer exists.

I’m writing a book that argues orchestration jobs are the first layer to fall to AI. I’m also selling a product that automates orchestration work. If you’re reading this thinking those two things shouldn’t coexist, you’re seeing the same contradiction I see. That’s the whole thing.

Let me be direct about what happened. I didn’t build Orca in ignorance of what it would do. I built it in full knowledge. I watched the labour market signal for three years. I watched hiring freezes replace hiring. Watched contractors not get renewed. Watched graduates send 200 applications for work that would have been guaranteed five years ago. Then I watched what’s left of those jobs hollow out underneath from the inside as the tools got better every quarter.

The honest position: the tools get built whether I build them or not. The market has decided that orchestration can be automated. Three or four major companies are deploying autonomous agent swarms into these roles right now. If I don’t build Orca, someone else does. If nobody builds anything, the tools exist anyway, just worse, more expensive, less transparent about what they’re doing. The choice isn’t between building and not building. It’s between building carefully, in public, with the argument about impact visible, or stepping aside while someone else builds it without that honesty.

So I built it in public. I wrote the book arguing what was coming. I’m building the tool. I’m not pretending they don’t interact. This piece is me sitting in that interaction and explaining why I think it’s the only way to be credible in this market.

What Orca actually is

Orca is a native macOS application built on Tauri 2, React 19, TypeScript, running on Supabase in the back end. It integrates with your email, your calendar, transcripts from your meetings, your tasks, your contacts. It’s sold as a managed service on Claude API. Two hundred and fifty pounds a month, personalised setup, one-to-one onboarding with me.

Here’s what it does.

Calendar integration plus CRM plus email plus LinkedIn history produces a meeting brief in under a minute. Background research. Relationship history. What they’re working on now. What they asked about last time. Commercial context from Companies House. The whole picture that used to require someone spending half a day on research and coordination work. Now it’s generated, reviewed, and ready.

Follow-up emails. You finish a meeting, Orca has watched the transcript, it knows what was agreed, it understands your voice from previous emails, and it drafts next steps in your tone. Not templated. Not generic. Actually your voice.

Inbox triage. Three hundred emails a day becomes thirty flagged items and a summary. The ones that need action now. The ones waiting on response. The ones you should probably read but don’t need to act on. The ones that are just FYI.

CRM management that doesn’t require you to manually log every interaction. Calendar events feed through automatically. Notes get attached. Warmth scoring tells you which relationships are warm and which are cold by action, not by your memory.

LinkedIn content that’s actually in your voice. Not the corporate newsletter tone. Not the motivational poster energy. The way you actually think about the things you think about.

All of this runs continuously. It doesn’t take meetings off. It doesn’t need annual leave. A solo founder with Orca has the operational output of a PA, a social media manager, a CRM administrator, a research analyst, and a junior strategist.

That’s not hyperbole. That’s what it does, measurably, for the people currently using it.

Who it’s for

Founders who can’t afford (or don’t want) an operations person but who are drowning in coordination work. Directors running consultancies and agencies where they’re the bottleneck. Anyone operating at a level where the work is mostly about knowing what’s happening, making decisions based on that knowledge, and then coordinating the outcome, but who’s currently spending two-thirds of their time just keeping the information flowing.

The clients who’ve found most value are people who were previously good enough at coordination that they could do it themselves, but who hated it. Not because they were bad at it. Because they were good enough to know it was taking up space that would be better used on thinking, strategy, relationship building, actual work that required them.

What it displaces

Here’s where the honesty matters.

Orca directly displaces the functional output of junior support roles. PAs, yes, but also research analysts, CRM administrators, social media coordinators, the people whose primary function is keeping information flowing between systems and people. Every user of Orca who’s replacing someone is contributing to the displacement this book describes.

Every demo I do, someone eventually asks the question quietly: what happens to Sarah. Not in those words usually. Usually it’s “what about my team” or “what happens to the person in this role.” This piece doesn’t answer that fully. Article three does. But the honest answer starts here: roles that are 80 per cent orchestration, 20 per cent something else, those roles are contracting. The maths is unavoidable.

Orca accelerates that contraction. It makes the contraction cheaper, easier, more technically feasible. It reduces the friction on something that was going to happen anyway.

Why this is the only credible position

If I were selling Orca while claiming it wouldn’t displace anyone, you’d be right to dismiss me. If I were selling Orca while denying that displacement was happening at all, you’d be right to think I was either lying or willfully ignorant.

Instead, I’m saying: I know what this does. I’m selling it anyway. Here’s why that’s honest instead of hypocritical.

First, the alternative to my building this isn’t no displacement. It’s worse displacement. It’s a proprietary tool sold by a company that doesn’t have to sit inside the consequences. It’s an internal capability built by a larger company that does the same work without the transparency.

Second, I’m visible. Every user can see what’s happening. Every pitch includes the full picture of what the work was and what Orca does with it. I’m not hiding the displacement inside generic language about “augmentation” and “productivity.” I’m showing it.

Third, I’m betting my credibility on the advice in the book. If the displacement thesis is wrong, if orchestration work actually absorbs people and grows, then I look foolish for building this. That’s a real downside I’m carrying. It’s not enough to offset the harm if I’m wrong, but it’s not nothing either.

Fourth, the people buying Orca are getting advice. Not just from the tool. From me, through the setup and the ongoing relationship. That advice says: use the time you’ve freed up to get better at the work that actually matters. Build the relationships deeper. Think about strategy that’s actually novel. Do the work only you can do. Don’t just offload the coordination and move on to the next layer of coordination above it.

The hardest part of a demo

I do a demo of Orca at least three times a week now. The hardest moment comes at the end.

After the brief is generated, after the follow-up email is drafted, after the inbox is triaged and the CRM is current and the LinkedIn content is posted, there’s usually a pause. The person I’m showing it to is usually a founder or a director. They’re relieved. Visibly, unmistakably relieved. The thing that’s been sitting on their shoulders, the weight of coordination work they’ve been carrying for years, has a path off.

But then they say something like: “So what does Sarah do now?” Not bitter. Usually not angry. Often with a kind of uncomfortable recognition in their voice.

Sarah in this case is whoever’s been doing the coordination work. Usually it’s someone junior, someone talented enough that they’d be good at it, often someone who went into it thinking it was the path toward strategy work, and who’s now realised it’s the path toward being made redundant.

I answer the question. I give them the honest version. And I watch them work through the moral mathematics of it. Want the tool, want the time back, want their business to run better, and also want Sarah to have a job. Those things don’t coexist anymore. The tool makes it impossible to have both.

That’s the moment where building this in public, with the argument about impact visible, matters. Because I’m not pretending the maths is different. I’m showing it. And the conversation that happens next, about what Sarah could do instead, about how to handle the transition, about what work actually requires a human, that conversation is the one that should happen, and almost never does.

The retraining narrative is one of the most durable fictions in contemporary policy. When an industry dies, governments say: retrain them. When automation threatens jobs, economists say: retrain them. When a sector collapses, business leaders say: retrain them. It is the comfort story we tell ourselves because the alternative (that some people’s skills have no future and we have no plan for them) is politically impossible to say out loud.

The retraining narrative is also testable. We have data. Multiple countries have run retraining programmes for displaced workers. We know what actually happens.

The results are consistent, careful, and devastating: most retraining programmes fail to successfully transition people into new sectors at sustainable wages.

The United States coal mining closure study from 2009 onwards is the clearest case. Coal employment in Appalachia was collapsing. It was obvious, inevitable, and on the policy agenda. Governments committed billions to retraining programmes. Miners would learn software development, advanced manufacturing, green energy installation. The cognitive foundation was there (miners are often strong in mathematics and spatial reasoning). The will was there. The money was there.

The outcomes were not there. Most miners who went through retraining did not successfully transition to sustainable employment in the new sectors. Some did. The majority did not.

Why. That is the question that matters.

Here is what the economists who study labour transitions found: it was not because miners were incapable. It was not intellectual failure. It was structural failure, and the structure was threefold.

First, the cognitive distance between coal mining and software development is not a gap you bridge in a six-month course. Mining is about reading geology, understanding mechanics, managing immediate physical risk. Software is about abstraction, symbolic logic, building systems you will never physically touch. The underlying thinking is different in kind. You can teach someone to code. Teaching someone to think like a programmer from a starting position of reading coal seams is not a linear progression. It is a reorientation.

Some people make that reorientation. Many do not. The transition requires not just learning new information but reorganising how you think.

Second, the geography was wrong. The retraining programmes were built for people who could move to coastal tech hubs or move to manufacturing corridors. But most coal miners were in Appalachia, in communities where they had family ties, property ownership, roots. Moving to Silicon Valley or Austin or Atlanta meant leaving behind everything. At age 45 with a mortgage in a declining market and a spouse with local employment, that move is not a rational choice. It is a forced displacement.

Most people stayed. Which meant they did the retraining locally, in programmes disconnected from the actual job markets they were supposed to feed into. The jobs that opened in tech and advanced manufacturing were not in Appalachia. So you retrained for a job that was not geographically available to you, or you did not retrain.

Third, the credential and experience hierarchy was working against them. A fifty-year-old coal miner with twenty years of industry experience is competing at entry level with twenty-five-year-olds who are native to tech or manufacturing. In an industry that values foundational knowledge and peer networks, the fifty-year-old has neither. The entry-level position pays what entry-level paid in 1995. A fifty-year-old with a mortgage cannot rebuild from that baseline.

This is not failure on the part of the miner. This is structural failure. The economic system is working exactly as designed. It rewards young people in new industries and makes it very difficult for people beyond a certain age to enter from the outside.

When you combine those three elements, retraining becomes a false choice. Not because people are incapable of learning. Because the structural conditions that would make learning economically viable are not present.

This is the data point that matters, and it is why the retraining narrative is so dangerous right now. Every solution to AI displacement assumes that “retrain” is viable, that people can learn new skills and move to new sectors. But the precedent says that works for a minority of displaced workers under very specific conditions: young enough to absorb wage compression, geographically mobile or in a place where jobs are actually available, in a sector where credentials from other fields transfer.

The coal-to-tech transition had all of those conditions stacked against it and it still failed for most people. AI displacement is going to be worse.

The sheer scale is different. Coal mining employed about 100,000 people in the US in its decline. AI could affect millions of knowledge workers across dozens of sectors simultaneously. There is no single new industry to retrain into. There is no unified effort or geographical alternative. Retraining narratives require that there is somewhere to retrain to. When the displacement is this broad, the “somewhere” does not exist.

The demographic is different. Coal miners were disproportionately working-class. Knowledge workers affected by AI displacement are middle-class with mortgages, dual-income households with geographic constraints, older professional cohorts with less patience for wage compression and geographic instability. The coal story was that they could move to a new region and start over. The knowledge worker story is different. They cannot start over. They have family, property, obligations. The retraining has to work in situ, or it has to work in a way that maintains their economic standing. And that is rarely what retraining achieves.

The skills are different. Coal mining requires physical presence in a specific place. Software development can be done remotely, can be done from anywhere. Some knowledge work that AI is displacing has the same property. But some of it does not. Relationship management at scale is place-dependent. Strategic judgment in an organisational context requires being present. Complex creative work often benefits from physical proximity and informal collaboration. You cannot retrain someone in those things remotely, at scale, in a six-month programme.

And then there is the psychological element, which the policy papers barely mention. When retraining fails, it does not fail cleanly. It fails after years of effort, after investing time and emotional energy into learning something, after believing the story that this was your path forward, after watching peers leave and succeed or fail, after watching the original job market continue to exist (in nostalgia, in memory, in community stories). The retraining narrative does not just fail to deliver work. It fails to deliver dignity.

The honest position is this: for most people whose jobs are automated by AI, retraining is not a viable solution. It is a story we tell so we do not have to confront the alternative, which is that we have no plan. The minority of people who will successfully transition to new sectors will be the ones who were already mobile, already young, already positioned to absorb wage compression, already embedded in networks that span sectors. The rest will do other things. Some of those things will work. Most will involve downward economic adjustment.

The question is not how to retrain people. The question is how to manage the economic transition for people who cannot be retrained, and we have not started that conversation, because it requires admitting that retraining is not the answer.

Everything from the audit and the positioning work sounds different when someone actually does it out loud, with another person in the room.

Tom runs a digital agency in the Midlands. Fourteen people when we last met, six months ago. Twelve now. Decent client list, mostly B2B tech and professional services. Solid reputation, the kind of agency that doesn’t make the news or win flashy awards but delivers consistent work and keeps clients for years. Tom built it from scratch after leaving a bigger agency in his early thirties. He’s forty-four. The agency is his livelihood, his identity, and about 70% of his self-worth, though he’d never say that last part out loud.

He didn’t fire the other two people. One left for a client-side role and he decided not to replace her. The other was a contractor whose project ended, and when Tom looked at the scope of the next project, he realised the AI tools his delivery team had started using could cover the gap. So he didn’t renew.

“It wasn’t a conscious decision to cut,” Tom says. “It was more like, the space where those people used to be just, sort of, filled itself in? The tools did it. We didn’t even have a meeting about it.”

That’s how it happens. Nobody makes the big decision. The small ones make themselves. And then one day you look around and the office is half empty and nobody remembers the meeting where they decided to cut. Because there wasn’t one.

“Yeah, that’s exactly it. And I’m looking around thinking, right, if two gaps filled themselves in six months, how many fill themselves in the next twelve? Eighteen?”

Tom rubs his face. He’s the kind of person who starts sentences and reroutes them mid-way, which is how you know the thinking is live, not rehearsed.

“So, right. We’ve got twelve people. The tools are good now. Really good. The reporting that used to take the account managers a day and a half is basically done by the time they sit down in the morning. Campaign setup, bid management, the tagging stuff, most of that’s automated or close to it. Which means the account managers are spending, I don’t know, maybe 40% of their time on actual client work and the rest of it is sort of, looking for things to do. And I can see it. They can see it. We’re all just sort of pretending we can’t.”

“How many account managers?”

“Four.”

“And if the tools do 60% of their current work, the maths says you need, what, two?”

“One and a half, probably.”

There’s a pause where you can hear the sound of someone doing the arithmetic they’ve been avoiding and arriving at a number they can’t unknow. Not shock, exactly. More like the moment you check your phone after a long drive and realise you’ve been going the wrong way for an hour. The destination hasn’t changed. The route has.

“Your wife’s income being in a different sector is the best thing you’ve told me today,” I say, after we’ve walked through the household numbers. “If the agency income drops, the household doesn’t collapse. That’s rare. Most of the agency owners I talk to are married to someone in marketing or PR, which means both incomes are in the blast radius.”

Tom nods slowly. Hadn’t thought about it that way.

The conversation moves through the practical terrain. Two account managers need to stay, the ones with deep client relationships. Two need to transition, the ones who do the work the tools now do. Not because those people are less capable. Because the work they’re best at is the work that’s being automated away.

“I feel like I’m betraying them,” Tom says.

“A bit, yeah. That’s the cost of running a business during a transition like this. You can minimise the betrayal. You can’t eliminate it. Anyone who tells you otherwise is selling something.”

We walk through the timeline. Eight months of runway. Fixed mortgage rate for twenty-four more months. A window where the household costs are predictable and the business can change without panic. The kind of window most people don’t have.

“If you wait until the P&L forces the decision, you’ll make it badly,” I tell him. “From panic, in a hurry, and the people who leave will leave angry because they felt blindsided. If you make it now, while you’ve got eight months of runway and the choice is yours, you can do it properly. Transition support. Honest conversations. Genuine references. Time for them to find something else. The difference isn’t the outcome, the headcount ends up the same. It’s how much damage you do on the way there.”

Tom is quiet for a while.

I ask him what comes next for him personally. Not for the agency. For him.

“The agency at fourteen people and the agency at eight are different businesses,” I say. “The second one makes more money per head but it needs a different kind of leader. Less people management, more strategic direction. Less operations, more client relationships. Are you that person?”

“I’m not sure,” Tom says.

That’s an honest answer. Better than the people who say yes without thinking.

I suggest a different shape for the work altogether. Advisory. Training. Fractional director work. Three or four businesses at a senior day rate. Portfolio career. One where the agency survives but isn’t all of it. One where if a client drops, the household doesn’t panic.

“Is that what you do?” Tom asks.

“Part of it, yeah. Plus the tools, plus the EdTech platform, plus the writing. Multiple income streams, different risk profiles, none of them dependent on a single client or a single industry. Not because I’m clever. Because I’ve seen what happens when you put everything in one basket and the basket gets automated.”

The meeting runs over. Tom checks his phone. The unconscious gesture of someone who has somewhere else to be but doesn’t want to leave yet.

“The people I let go,” he says quietly. “The two account managers. What do they, like, actually do? Not the LinkedIn platitude version. What actually happens to them?”

I sit with the question. It deserves a pause.

“Some land on their feet. The ones with strong networks and transferable skills and a bit of financial runway, they’ll find something. Maybe not equivalent, but decent. Some struggle for six months to a year and then find a new direction. Trades. Teaching. Healthcare. Something with a physical or relational component the tools can’t touch. And some of them, the ones without savings and without networks, have a really hard time. A third of UK adults have less than five hundred quid in emergency savings. If your account managers are in that bracket, redundancy isn’t a career setback. It’s a household crisis.”

Tom nods slowly.

“My account manager Rachel. She’s been with me seven years. She was my second hire. She came to our wedding.”

“I know.”

“And you’re saying she’s one of the ones whose role doesn’t, sort of, survive?”

“I’m saying her role as it currently exists doesn’t survive. Rachel might. Depends whether the things she does beyond the orchestration, the client relationships, the judgment, the ability to calm down a panicking CEO, are enough to justify a restructured position. You know her better than I do.”

“She’s brilliant with clients. They love her.”

“Then restructure the role around the bits that are brilliant and let the tools do the rest. But be honest with her about what’s changing and why. People can handle a changing role. What they can’t handle is waking up one day and finding out the change happened without anyone telling them.”

Tom stands up. Puts his jacket on. Pauses at the door before leaving.

“The team conversation. Don’t put it off,” I say. “Every week you wait costs you options.”

He nods. He knows. That’s always the hard part. Not the decision. The conversation that comes after.

After he leaves, I open my laptop. Orca has already drafted the follow-up email from the meeting transcript. Key decisions, action items, ready for review. I add one line at the end, something about Tom handling a tough situation well, because the AI can summarise a meeting but it can’t tell someone they’re doing okay when they need to hear it.

The next meeting is in forty minutes. Another business owner. Another set of numbers that don’t quite work. Another conversation where the uncomfortable truth has to be said out loud before anything useful can happen.

This is the work. Three sides of the same thing. Building the tools. Advising the people. Measuring the skills that survive. The tension doesn’t resolve. You just get used to holding it.

Here’s the uncomfortable part of scenario planning: the most likely outcome is not the one anyone is trying to create. It’s the one that happens when nobody makes a conscious choice.

Neo-feudalism is the path of least resistance. When productivity gains from AI go entirely to capital owners, when there’s no political mechanism to redirect them, when every current trend points in one direction: wealth concentrates, platforms consolidate, labour fragments. The scenario doesn’t need to be chosen. It just needs to not be prevented.

The mechanism is simple enough. Capital captures most AI productivity gains. A small class persists: people whose work depends on human trust (the relationship roles), genuine originality (top-tier creative work), and liability (the oversight functions where someone needs to sign off). Everyone else works in fragments. Three hours of content moderation here, a delivery shift there, a micro-task on a platform that takes 35% commission. Housing becomes extractive. You never own. You rent from corporate landlords or the people who accumulated property when professional salaries still existed.

The political system becomes performative. Elections happen. Manifestos are published. But the legislation that actually shapes daily life, platform regulation, tax policy, housing law, the employment status that determines whether you have protections or not, is written by people whose employers profit from things staying as they are. Democratic forms persist. The substance hollows out.

This is not science fiction. There’s a growing body of academic literature mapping exactly this trajectory. Yanis Varoufakis calls it “technofeudalism” and describes how platforms function as feudal lords: you don’t own the marketplace or the tools or the customer relationship. You rent access, and the rent goes up whenever the platform decides. Jodi Dean’s work traces the mechanism by which platform economics recreates feudal extraction. Cedric Durand examines how the digital economy tends toward rent-seeking rather than production. Shoshana Zuboff describes surveillance capitalism as a tribute system that would make a medieval lord envious.

The current data supports this trajectory. Oxfam reported in January 2026 that global billionaire wealth jumped 16% in 2025 to 18.3 trillion dollars. An 81% increase since 2020. Three thousand billionaires for the first time. Billionaires are 4,000 times more likely to hold political office than ordinary citizens. In the UK, 56 billionaires hold more wealth than 27 million people. Meanwhile, one in four people globally lacks regular food access.

The World Inequality Lab data shows the pattern in detail. India’s top 1% wealth share increased from 13% in 1961 to 39% in 2023, a threefold increase in 62 years. US billionaires grew from 835 in 2024 to 924 in 2025. The pattern reflects concentrated tech wealth, AI-driven productivity, and the simple mathematics that returns on capital exceed returns on labour, which is the dynamic Piketty described and which AI accelerates rather than reverses.

What does daily life look like in this scenario. Take Hannah, a former marketing manager. She rents a room in a house she would have owned outright before AI displaced her role. The house was purchased by a property investment company in 2031 when the original owners defaulted. She works three simultaneous gig tasks: content moderation for an AI platform at four hours daily, algorithmically managed, flagged for performance if accuracy drops below 94% in any 15-minute window. Virtual customer service for a subscription wellness company at variable hours, paid per resolution at a rate declining annually as AI handles the easier queries. Occasional freelance marketing through a platform that takes 35% commission and rates her five-star, where 4.7 is borderline.

Her income is roughly 40% of what she earned in her old role. No pension. No sick pay. No holiday. No employer. Her children’s school has lost funding for everything except core curriculum. The careers lead position was cut in the last budget round. The local GP practice closed; healthcare is a 40-minute bus ride and a four-week wait. She votes in every election. Nothing changes. The local MP gives passionate speeches about supporting working families and votes reliably with the platform companies that employ those families on terms the MP would never accept for themselves.

Why is neo-feudalism at 45% probability rather than lower. Because it requires nothing. No coordination. No political will. No conscious choice. Just inertia plus the compounding mathematics of capital accumulation. Every trend currently visible, wealth concentration, platform consolidation, labour fragmentation, institutional decay, points toward this outcome. The scenario doesn’t need to be chosen. It needs to not be prevented.

The comparison to the Industrial Revolution is instructive. The benefits took 50 to 70 years to reach working populations. Without intervention, capital captured the gains and everyone else adjusted downward. We have precedent. We have evidence from that period about what happened and how long it took. And we have no political mechanism currently visible that would produce a different outcome this time.

That’s why the other 55% matters. Managed transition at 30% requires deliberate policy reversal of twenty years of trend. Fragmentation at 20% requires regional divergence rather than global concentration. The black swan at 5% is the admission that something we haven’t modelled might reshape everything. All three require something to be done differently than it’s being done now.

Neo-feudalism requires nothing but inertia. And inertia is what we have in abundance.

I call it the forty-minute moment, though the actual time varies between thirty and fifty minutes depending on how much we talk through the initial setup.

The demo starts with connecting Orca to someone’s calendar. This is simple: grant access, authenticate, done. Then we pick a meeting from their schedule. Usually a client meeting, sometimes an internal one. Something real. Something they actually have to prepare for, possibly in the next week. I ask them to show me, briefly, what they currently do to prepare for this specific meeting type.

They show me. Sometimes it’s thorough. They spend a day researching the client company, pulling together information from Companies House, LinkedIn, recent news, previous project outcomes, internal project notes, CRM context. Sometimes they’ve got a system. Sometimes it’s a spreadsheet. Sometimes it’s a PA or a junior who does it because the founder isn’t doing the detail work themselves. Sometimes it’s just: they know the person, they show up, they wing it.

Then I say: watch.

I connect the meeting to Orca. I add the key integration points: their email, their CRM, LinkedIn, Companies House. I point the system at the meeting they selected. I press generate.

Forty seconds later, Orca produces a brief. Not a skeleton. A brief. It has the background research on the client company, including recent funding rounds, headcount changes, sector positioning. It has the relationship history from their CRM, cross-referenced with email conversations, pulling the tone and context of the relationship. It has the prep notes on previous projects, including outcomes and blockers. It has a personal note on the client contact, built from LinkedIn and CRM warmth scoring. It’s a page and a half, long enough to be useful, short enough to read before the meeting.

The person looks at the screen. They look at me. There’s a pause. I’ve watched this pause happen ninety-seven times. The pause is consistent. It’s about four or five seconds. During that pause, they’re doing the calculation. The calculation is unconscious but unmistakable once you’ve seen it a few times. They’re working through: what I just did in forty seconds, how long would that manually take?

Someone says: “That would have taken my assistant half a day.”

Someone else: “I used to have someone doing exactly that.”

Another person: “I used to have two people doing that.”

And once, with the kind of uncomfortable honesty that arrives unexpectedly in a room where someone is trying to sell you something: “So what does Sarah do now?”

I don’t have a good answer for that question. What I have is a product that works, a business model that scales, and a knowledge that’s become harder to carry the longer I’ve been building this. The efficiency I’m selling to one person is, in aggregate, a redundancy notice for someone else. Not necessarily their Sarah. But someone’s Sarah, somewhere, in the distributed math of thousands of founders adopting the same tool, running the same calculation, making the same decision.

The interesting part comes after the demo ends. After they’ve seen what the tool does. After they’ve run the calculation in their head and confirmed that yes, this eliminates the work that used to require human labour. And they say: can I try it.

They try it. Sometimes they connect their calendar themselves. Sometimes they want to see it work on a different meeting type. Sometimes they want to test it against a difficult client situation where the preparation is actually hard. I let them. They run it. It works. It usually works. Not always perfectly, but well enough that the twenty minutes they’d spend cleaning up the AI output is less than the four hours they’d spend writing it from scratch.

That’s when the expression shifts.

The calculation they did at forty seconds was intellectual: this is more efficient. The calculation they do after running it themselves is emotional. They feel the weight lifting. The admin they’ve hated for years is suddenly handled. The meetings they used to dread, the ones that required four hours of research and prep, are now prepared in forty seconds. The inbox they’d been drowning in, reading and categorising and flagging and responding to, is now triaged. The CRM they’d been neglecting because the updates were tedious, that’s current. The content calendar that required hours of planning and coordination, that can now be scheduled in batches.

For two hundred and fifty pounds a month, they’ve bought back hours of their week.

They sign up. They’re relieved. The relief is real every time. I’ve watched this moment ninety-seven times. The weight they’ve been carrying has a weight, and suddenly it’s gone.

Then comes the part I find hardest to witness. The part that comes about ninety seconds after they sign up. They’re filling in their profile. They’re connecting more calendars, more data sources. They’re configuring their voice preferences. And then, usually about three minutes in, they pause. They look up. And they ask the question in a different tone.

Not: what will this do for me. But: what does this mean for my team.

Sometimes they’re asking it aloud. Sometimes it’s just an expression. I watched one founder fill in the profile section that asks “describe your role,” and when she got to the field for her assistant’s role, she stopped. She just looked at it. Didn’t type anything. Looked at me. Said nothing. That was the whole conversation.

Another one said: “My whole marketing coordinator role, all of it, this does all of it, doesn’t it.”

I said: “Most of it, yes. The tactical parts. Email management, calendar coordination, basic content drafting, simple analytics reports.”

He nodded. Then he said: “I need to think about something. Give me a week.”

He didn’t sign up. Or rather, he signed up after a week. But in that week he’d made the decision. The decision was: I’m going to use this, and I’m going to figure out what my coordinator actually does that this doesn’t do, and I’m going to have that conversation with her. Or I’m going to figure out what she could do if I freed her from the tactical work, and we’ll build something new out of her capability. Or I’m going to help her transition somewhere else, and I’ll feel like shit about it but I’ll do it properly.

He bought the tool because the economics didn’t give him a choice. But he’s thinking about it because he’s a human being and his coordinator is a human being and he’s not comfortable with the distance between the two truths.

That’s the forty-minute moment. Not the efficiency. That’s just arithmetic. The moment is the moment after. The moment when you’ve understood intellectually what this thing does, and you’ve felt emotionally what it means, and you’re holding both those things at the same time and they don’t align and you have to decide what you’re actually going to do about it.

The buy signal is relief, then guilt, then maths.

Relief because the tool works and you’re buying yourself time.

Guilt because you understand that your time is someone else’s absence.

Maths because you run the calculation: if I free up fifteen hours a week for my coordinator, what could she do with that. Or if I don’t free up those hours, if I just use Orca to make sure she’s redundant, what does that let me do. Build the business faster. Compete harder. Survive. The calculation doesn’t have a good answer. It has answers. None of them are comfortable.

That’s why I’m writing a book about displacement while selling tools that displace. That’s why I’m sitting in advisory sessions helping business owners restructure around AI tools, while knowing that the restructuring means redundancy somewhere. That’s the triangle I sit in. The tool, the transition, the cost of the transition. All three vertices are real. All three are true. And none of them cancel the others out.

The book is trying to describe what’s happening at kitchen tables and in business conversations and in the forty minutes after someone has seen what Orca actually does and understood what it means. The tool is completing that transition. I’m documenting it. The people on the other side of both are the ones bearing the cost.

That’s not a clean position. It’s the position I’m in.

The economic conversation happens at the wrong altitude. Economists and policymakers talk about household debt-to-income ratios, disposable income distribution, precautionary savings as a percentage of GDP. These are coherent at a macro level. At a kitchen table, they’re abstractions that don’t speak to the actual number that matters: how long can we keep the house if both incomes disappear.

UK household debt stood at 116.9% of disposable income in Q3 2025. That means the average household owed £117 for every £100 earned. The Office for National Statistics tracks this. The Bank of England stress-tests against it. The Financial Conduct Authority uses it to set mortgage lending criteria. And almost no one looking at their own situation thinks about it in those terms.

What they think about is this: we have ten grand in savings. We have a mortgage of four hundred and sixty thousand. We have a joint income of one hundred and ten thousand. The mortgage is forty-two times our savings and four-point-two times our income. Now, what does ten grand actually buy us.

The answer depends entirely on household outgoings. The Tom and Priya scenario from the book uses a real London professional-class household: mortgage £2,180 a month, council tax £180, energy £150, nursery fees £800, car payment £250, insurances and transport £200, utilities and basics £400. Total: roughly £4,200 a month. Their ten grand covers, at best, two and a half months of that burn rate. After tax and redundancy payments, it’s closer to two months before the account empties.

The word for this is not security. The word is precarious.

Precautionary savings exist to handle the thing that hasn’t happened yet. Car breaks down. Boiler fails. Illness means time off work. These are expected within a reasonable horizon. The financial services industry uses three months of expenses as the baseline: emergency fund sufficient to handle a short-term shock. Anything less, and you’re not covered. Anything more, and depending on the definition of “more,” you’re either being prudent or leaving capital idle.

But three months of expenses assumes the income returns. It assumes the boiler doesn’t break in month two of a six-month job search. It assumes your industry doesn’t restructure while you’re laid off. It assumes the redundancy payment covers the gap. All reasonable assumptions in a stable labour market. None of them hold in a market where task-level AI exposure quietly eliminates job categories over a eighteen-month period.

What’s happening now is that jobs are disappearing not through dramatic redundancy announcements but through hiring freezes, contractor squeeze-outs, and roles that don’t get backfilled when someone leaves. The paralegal who lost her job to an AI compliance system didn’t get a redundancy letter. Her firm restructured. Eight roles became two. She was one of the eight. The documentation said consultation period. The economics said redundancy. The bank account didn’t care which word they used.

This is the shape of the quiet phase. It’s not a crash. It’s a compression. And the compression hits the household that has stretched their finances to the absolute limit of what the lending tests allow. The mortgage broker calculated they could afford £460,000 on £110,000 income because the maths said so on paper. The paper said four-point-two times income was within the stress-test parameters. The Bank of England ran scenarios where rates spiked and house prices fell and employment remained stable. It ran scenarios where employment fell and rates remained stable. It didn’t run the scenario where the income loss is permanent and structural because that income loss comes from the work being automated away.

And so we’re left with Tom and Priya at a kitchen table with £8,200 in an ISA, £1,100 in a current account, and a mortgage of £460,000. The calculation is brutal because the units of analysis matter. You can’t stretch £9,300 very far against £4,200 a month in household outgoings.

Here’s what actually happens. Priya loses her job first. Tom’s salary covers the mortgage and the essentials. It doesn’t cover everything. They pause the ISA contributions they’d planned for their daughter’s nursery place in September. They stop the regular savings pattern they’d maintained for four years. They run down the buffer month by month. Eight months later, Tom’s employer announces an operational transformation. The compliance team of twelve becomes six. Tom is one of the six who leave.

Now they have no income. They have the redundancy payment: twelve weeks at full salary. That covers just over three months of £4,200 spend. After that, they’re on Universal Credit. Both of them. A household that paid sixty-five thousand in combined income tax now qualifies for means-tested benefits. The benefits cover about a third of the mortgage. They apply for a payment holiday. Interest accrues and adds to the debt. They apply for a forbearance scheme. The deferred payments defer but don’t disappear. They accrue at the mortgage. The house is built on an assumption that no longer holds.

The stress tests assumed distributed risk. Individual households would lose income at different times, in different sectors. Not thousands of households in the same sector losing income in the same quarter because their employers adopted the same technology. When the risk is correlated, when the shock is demographic rather than individual, the numbers that worked on paper don’t work in practice.

And so the unit of analysis that actually matters, that sits at the kitchen table where real decisions get made, is not the macroeconomic ratio. It’s the number of months of household expenses you have accessible right now. Not theoretical future savings. Accessible, liquid savings. Now.

For the UK median household earning thirty-five thousand a year with household expenses of roughly two thousand, ten grand covers about five months. For the professional household earning one hundred and ten thousand with expenses of four thousand plus, ten grand covers two and a half. The better your income, the tighter your budget, because you stretched to match your income. That’s not recklessness. That’s how mortgages work. The system is designed to stretch you.

The honest assessment: most UK households have between two and four months of household expenses in accessible savings. Most know this number. Some know it precisely. They’ve calculated it because they’ve thought about the thing that hasn’t happened yet. They’ve done the maths at the kitchen table. And they’ve understood, at three in the morning, that if both incomes disappear simultaneously, they have between two and four months before the system that holds their life together starts to fail.

That’s not long.

The reason “it was fine last time” became the default response to technology and employment is not that economists are lazy, though some are. It is that for two hundred years, the statement was demonstrably true. Every major technological transition from 1790 onwards really was fine, eventually, for most people. The farms emptied, the factories filled. The factories hollowed, the offices filled. The offices thinned, knowledge work expanded. Living standards rose across centuries. The Luddites turned out to be wrong about the long term, even though they were right about the immediate consequence.

This consistency was not accidental. It was mechanical. Three specific things happened in every previous transition, and those three things worked together like a system. Understanding that system is understanding why this time might be different.

Here is the first mechanism: new technology created demand for new tasks that did not exist before.

The steam engine is the clearest example. Nobody woke up in 1769 and said, “We need an occupation: steam engine mechanic.” The occupation did not exist. But once Watt’s engine became practical, it needed people. Someone had to build them. Someone had to maintain them. Someone had to install them. Someone had to understand what could be done with them. The textile industry did not just displace hand-loom weavers. It created an entirely new layer of work around the machines: engineering, maintenance, installation, troubleshooting, parts manufacture, transport, coordination.

The steam engine destroyed one kind of job and created several others. The same pattern held through every subsequent transition. The assembly line destroyed craft skills but created flow engineers, line supervisors, scheduling coordinators, quality inspectors. The computer destroyed typing pools but created programmers, data entry specialists, systems operators, IT support. Each wave of automation generated adjacent work that required human attention. The machines could do the core task but not all the supporting work around it.

This is the first crack, and it is a visible one. AI is not like the steam engine or the assembly line or the computer. Those were narrow technologies, purpose-built for specific tasks. They automated one thing. The work around them was different work. But AI is a general-purpose labour substitute for the cognitive work that knowledge workers do across the board: reading, writing, analysing, coordinating, summarising, recommending, deciding.

When you automate reading and writing and coordinating, the adjacent work does not disappear into “something else.” It collapses into the same system. The AI reads and writes and coordinates. The human becomes redundant not just at the task but at the meta-layer of managing the task. There is no adjacent work because the work itself has been automated.

This is the critical difference. Previous transitions automated task-specific tools. This one automates general cognition.

Now, the second mechanism: productivity gains reduced costs, which expanded demand, which created more work.

This is basic economics. When factories made textiles cheaper through mechanisation, more people could afford textiles. Demand increased. To meet the increased demand, more factories were built, more textile workers were hired, more people were needed in the supply chain. The Luddites were right that mechanisation destroyed their jobs. But it destroyed them in the context of a massively expanding market. Textile production went from craft-scale to industrial-scale. Employment in textiles actually increased, even though the wage per worker dropped and working conditions were worse.

The same pattern held in every subsequent transition. Cheaper automobiles meant more automobiles sold. Cheaper information processing meant more information consumed. Cheaper computing meant more software written and more systems implemented. The productivity gains from the previous technology created expanded markets for the products of that technology.

This mechanism is holding now. McKinsey estimates that productivity gains from AI could automate thirty per cent of work hours in the US economy by 2030. That is a genuine, measurable gain. But here is the problem: the gains are not spreading.

The gains flow to whoever owns the AI system. And the AI system does not need many humans to operate it. The steam engine needed engineers and operators and maintenance workers in significant numbers. The assembly line needed supervisors, schedulers, quality checkers. The computer needed systems administrators, network engineers, helpdesk staff. Each technology created operational overhead.

AI requires a data centre, an electricity supply, and a handful of engineers. The humans it replaces do not get absorbed into operating it, because there is nothing to operate. The productivity gains stay at the top with the owners of the system.

This is the second crack. Demand expansion is real, but the employment expansion that historically followed it is not happening.

And then the third mechanism: displaced workers had transferable skills.

A farm labourer in 1790 could learn to operate a loom. It was not trivial. It required training. But it was learnable. A typist in 1980 could learn to use a word processor. The cognitive distance was manageable. You did not become a fundamentally different person. You learned a new tool, and the underlying work was still coordinating information, managing documents, organising time.

The skills that carried you through previous transitions were portable. You learned them once in childhood or early adulthood and you applied them repeatedly across different technologies. An accountant learned mathematics and logic and procedural thinking. Those skills worked with a ledger in 1920 and a spreadsheet in 1980 and a database in 2010. The tool changed. The thinking did not.

This is the third crack, and it is the most serious one, because it is not gradual. It is categorical.

The skills that survive AI automation are not refinements of existing skills. They are different in kind. They are genuine creative vision: the ability to see a solution that has not been expressed before and guide a team toward it. They are relationship management at scale: the ability to hold complex relationships with competing stakeholders and move them toward aligned outcomes. They are strategic judgment under genuine uncertainty: the ability to make decisions when the information is incomplete and irreversible, knowing you might be wrong.

These are not skills you learn in a six-month retraining programme. They are capacities that develop over years, sometimes decades, sometimes rooted in childhood experience. Some people have them latent and have never used them. Many simply do not have them, and no amount of training will install them.

Here is the historical precedent that should worry you. When coal communities were promised that miners could retrain as software developers, the programmes mostly failed. Not because miners were stupid. Many of them were quick learners with strong logical thinking. The programmes failed because the cognitive distance was too great, the geographic mismatch was severe, the age distribution was wrong. A fifty-year-old with seventeen years of industry-specific career capital was not going to start at entry level in a field where twenty-five-year-olds had a structural advantage.

The transition needed a generation. Not everyone, but most of the people whose livelihoods were built on coal mining did not successfully transition to software. They left the industry, took lower-wage work, moved away, or stayed and struggled.

That was coal to tech. One person, one place, one time.

AI displacement is the coal-to-tech problem at a hundred times the scale, compressed into a fraction of the time, happening simultaneously across every knowledge work sector.

Now, here is what matters. In every previous transition, all three mechanisms worked together. New tasks were created, so there were jobs to move into. Demand expanded, so those jobs multiplied. And displaced workers had portable skills, so they could move into those jobs in reasonable timescales. It was not comfortable. It was not always higher. But there was a rung to step to.

The question is whether all three of these mechanisms hold this time. And the honest answer is: they are not holding. They are cracking simultaneously.

New task creation is stalled because the technology is too general. Demand expansion is real but gains are not spreading. And skill transferability is broken because the skills that survive are not teachable in retraining programmes.

When all three systems crack at once, you cannot rely on the historical pattern. You cannot tell people, “it was fine last time” because the mechanism that made it fine is not present now.

This is where the position-don’t-predict framework becomes relevant. You cannot predict exactly what happens when all three absorption mechanisms fail. You can predict the direction: squeeze, compression, disruption. You cannot predict the precise magnitude or timeline. But you can position yourself for scenarios where all three are failing, and that is different work than trying to retrain into the next rung, because there might not be a next rung to retrain into.

The framework is simple. Stop trying to predict which specific job survives. Instead, identify the capabilities that survive across all scenarios, build those, and stay mobile enough to apply them in new contexts as they emerge. In a world where the old ladders are breaking, that is the only positioning that is not a gamble.

This is the argument the book develops across chapters one through four. But it is the foundation. These three mechanisms worked for two hundred years. Now they are all failing at once. That is why “it was fine last time” is not a reassurance. It is a warning.

Five questions. Two minutes. One honest answer. No email required to see your result. The follow-up analysis lands in your inbox if you subscribe.

Take Test

Subscribe now

In 1790, ninety per cent of Americans worked on farms. Today it’s less than two per cent. Nobody rioted over that transition. Nobody needed to. The factories were hiring.

Every previous technological revolution had somewhere for displaced workers to go. Agricultural workers moved to factories because factories actually needed bodies. Factory workers moved to offices because the service economy needed organisers, coordinators, managers. Office workers moved into knowledge work because the professional class was expanding and needed people to do the cognitive grunt work. There was always a next rung on the ladder. Not comfortable, not always higher, but a rung.

The pattern ran so consistently that economists gave it a name: the Luddite Fallacy. The Luddites themselves, in 1811, were textile workers who smashed stocking frames because the new machinery was destroying their livelihoods. They turned out to be right about the immediate consequences and wrong about the long-term ones. The machines did destroy their jobs. But mechanised textiles grew enormously, employed far more people than before (in different roles, at different wages), and kicked off an expansion that eventually raised living standards for nearly everyone. For two hundred years, the Luddites were wrong. Technology destroyed jobs in one place and created them somewhere else, usually somewhere better.

Which is why, when anyone worries about technology and employment, the standard response has been: it was fine last time.

Here’s the thing that actually matters about why it was fine last time. The mechanism wasn’t magic. It was mechanical. Three specific things happened in every previous transition, and all three of them are cracking now.

First, new technology created demand for new tasks. Someone had to build the machines, operate them, maintain them, sell what they produced, transport it, market it, finance it, insure it. Each wave of automation generated a cascade of adjacent human work that hadn’t existed before. The steam engine freed humans from doing the work horses did. It still needed people to operate, maintain, and build around it.

Second, productivity gains reduced costs, which increased demand, which created more work. When factories made clothing cheaper, more people bought clothing. When computers made information cheaper, more people consumed information. Cheaper output meant bigger markets meant more jobs. The gains spread through the economy because the new industries needed workers, and workers with income became consumers.

Third, displaced workers had transferable skills. A farm labourer could learn to operate a loom. A typist could learn to use a word processor. The cognitive distance between the old job and the new one was manageable. You didn’t become a fundamentally different person. You learned a new tool.

These three things, new task creation, demand expansion, and skill transferability, held up every transition from agriculture to industry to services to knowledge work. The question is whether all three hold this time.

Here’s what’s actually happening: they’re all cracking at once.

New task creation is cracking because AI is not a narrow technology. It’s not a substitute for one thing. It’s a general substitute for the cognitive work that knowledge workers do across the board: reading, writing, analysing, coordinating, summarising, recommending, deciding. When you automate a narrow task, adjacent tasks remain for humans. When you automate the general cognitive toolkit, the question of what adjacent tasks remain becomes much harder to answer.

Demand expansion is holding, but the gains aren’t spreading. AI does create productivity. McKinsey estimates that thirty per cent of work hours in the US economy could be automated by 2030. But productivity gains don’t distribute themselves. They flow to whoever owns the AI systems, and the AI systems don’t need many humans to operate. The steam engine needed operators. The assembly line needed workers. The AI system needs a data centre, an electricity supply, and a handful of engineers. The humans it replaces don’t get absorbed into operating it, because there’s nothing to operate.

And skill transferability is visibly broken. The skills that survive AI automation are different in kind, not just in degree. They’re things like genuine creative vision, complex relationship management, strategic judgment under genuine uncertainty. These aren’t skills you learn in a six-month retraining programme. They’re capacities that develop over years, sometimes decades, sometimes from childhood. Some people have them latent and never used them. Many simply don’t.

The historical comparison here is instructive. When coal communities were promised that miners could retrain as software developers, the programmes mostly failed. Not because miners were stupid. But because the cognitive distance was too great, the geographic mismatch was severe, and a fifty-year-old with seventeen years of industry-specific career capital wasn’t going to start at entry level in a field where twenty-five-year-olds had a structural advantage. The transition needed a generation.

AI displacement is the coal-to-tech problem at a hundred times the scale, compressed into a fraction of the time.

The adoption curve for AI is faster than anything in the history of technology. ChatGPT reached one hundred million users in two months. The telephone took seventy-five years. By August 2025, over half of American adults were using generative AI tools. Previous transitions gave labour markets decades to adjust. AI gave them roughly five years, from mainstream availability in 2023 to restructuring necessity by 2028.

The honest position is directional confidence with precise uncertainty. The direction is clear: AI is automating cognitive work at a speed and scale that breaks the historical pattern of job absorption. The precision is absent: we don’t know whether the net displacement will be ten per cent of knowledge work or fifty per cent. But we know it’s happening.

And it’s happening quietly. Not in headlines yet. In hiring freezes instead of layoffs. In contractors who don’t get renewed. In graduates who apply to two hundred jobs and get three interviews. In ground that moves so slowly you can tell yourself it’s not moving at all.

Until it is.

Which brings us to the part that actually matters: when all three absorption mechanisms are breaking simultaneously, you can’t rely on “it was fine last time” to be right again. The next rung was fine last time because it existed. The question isn’t whether the next rung exists. The question is whether anyone’s built one, and whether displaced workers can reach it. Right now, the answer to both is no. And the time to change that answer is measured in years, not decades.