How AI is reorganising professional work
THE VAMPIRE'S BITE
You know the feeling. You used to spend three hours drafting a document, two hours in meetings about it, and the rest of your day on email. The drafting was hard, the meetings were tedious, and the email was soul-destroying, but together they made a day, and you went home tired in an understandable way.
Now you draft the document in twenty minutes with AI, and it's good. Maybe better than what you used to produce in three hours. Then you draft another, and another. By lunch you've done what used to take a week, and you're not tired in your body, you're tired in a place you can't quite name. The cognitive equivalent of having run a marathon while sitting perfectly still.
Steve Yegge nailed this when he described the AI-augmented work experience as a kind of vampirism: the company gets 10× the output, you get the same salary and a new kind of exhaustion.
CEOs and workers are both reaching for the wrong conclusions about what AI means for the structure of professional work. The real outcome of AI in professional services looks less like mass unemployment, and less like business as usual, than either side expects. The question at this point isn't whether AI changes professional work (spoiler: yes)... it's who ends up getting the value when it does.
Thinking was always the hard part of knowledge work, while execution was the holiday. Writing the brief was draining; formatting it was a break. Analysing the data was intense; building the spreadsheet was almost relaxing. AI removes the holiday.
When every mechanical task is instant, what remains is an unbroken stream of the hardest part of your job — judgement, planning, synthesis, detail review, decision-making… with no natural rest. Ten-times productivity is real, but it's 10× mental output, not 10× comfort; you're thinking at a rate that would have been spread across a team, while absorbing the cognitive load alone.
Yegge's insight that companies will try to capture the entire productivity gain is not just about individual burnout; it's about who controls the dial. A firm that discovers its senior associate can produce ten times the output has two choices: let the associate work a four-hour week for the same pay, or keep them at forty hours and capture the gain as profit. No company chooses the former, or even a middle ground. The competitive dynamics make it nearly impossible.
But running people at 10× for forty hours a week is equally unsustainable — they'll burn out or leave. It would be nice if dial could land somewhere in the middle, but that's wishful thinking given the collective track record of companies. But actually, there are two dials: the effort dial (the hr in $/hr in Yegge's post) and the value-captured dial (simplistically, the $). Yegge says you can't control the numerator in that equation, and that's been true in the past, but that's about to change.
Andrew Yang calls what's coming 'The Fuckening': the moment when corporate efficiency drives hit white-collar work with the same force that automation hit manufacturing. The playbook is straightforward: fire 15% now, use AI to maintain output, wait for the remaining staff to absorb the load, then fire another 20% when the AI improves. Markets reward headcount cuts; every earnings call that mentions 'AI-driven efficiency' gets a bump; and each individual CEO is stuck in a competitive dynamic where unilateral restraint is punished.
If you are old enough (like me) to remember defragging hard drives, this is similar. As employees master AI tools, some of them might set their hr dial to less than our notional 10x. They might prioritise sleep, time with families, coffee badging, etc. Arguing with them is an uphill battle, but if you just remove the right proportion of them, then the work consolidates onto the remaining contiguous hard drive sectors people.
They're not firing people, they're optimising.
But defragmentation doesn't reduce the amount of data on the drive; it rearranges it. Consolidating the same professional output into fewer people doesn't eliminate the work, it concentrates the load. And one thing we never did with hard drives was eliminate the "free" blocks after a defrag.
Eventually you hit the same wall Yegge describes, just with a smaller team: the four remaining associates doing the work of forty aren't more efficient forever; they're a single point of failure with no surge capacity, no institutional memory redundancy, and no one coming up behind them.
If everyone follows this playbook, there is no one left to buy anything. Mass white-collar unemployment compresses the professional middle class that buys consumer financing (home loans), legal services, software licences, and financial advice. The firms automating their way to efficiency are selling to the workforce being automated away.
The history of technology transitions (printing press, electrification, computing) warns against this approach: the winners were never those who eliminated the old workforce, but those who redirected the productivity gain into doing more. The printing press didn't make scribes more efficient; it made publishers possible. Electricity didn't just replace gas lighting; it created entirely new industries. Producing the same with lower costs is not a growth mindset. A growth mindset requires using efficiency gains to do more for your customers, not just cut costs.
The CEO who fires 90% of the legal team and pockets the savings will be outcompeted by the CEO who keeps the team, augments them with AI, and offers clients ten times the legal output for the same fee. One is racing to zero differentiation; the other is building a new kind of firm. The same AI tools that enable the efficiency play enable the expansion play, and they either work or not for everyone. There is no technology moat around cost-cutting.
Organisational resilience also degrades catastrophically with extreme headcount reduction. Knowledge doesn't live in documents; it lives in people's heads, in the informal networks of who-knows-what, in the pattern recognition that comes from fifteen years of seeing similar problems. Fire those people and the knowledge walks out permanently, unless, of course, you ask them to "write down what they know into a digital twin" or some other similar AI-based knowledge extraction before they go. And that's when the conversation gets interesting...
THE LEVERAGE SHIFT
A lot of commentary about AI and professional services misses what clients are actually buying. When McKinsey wins a strategy engagement, the client is not purchasing 'McKinsey'; they're purchasing the specific partner who attended the pitch, whose judgement they trust, whose industry knowledge they need. (Aside: it's fashionable to dismissively call consultants "CYA insurance", but this oversimplifies and undersells what they do). The brand gets the meeting; the person gets the mandate. Every senior professional in every professional services firm knows this.
This has always been true, but it never mattered structurally, because the partner needed McKinsey—not for the client relationship, which was always personal, but for everything else: the research team, the slide production, the global office network, the back-office machinery of a large firm. The overhead was the price of doing business, and it was substantial enough that going solo was impractical for all but the most entrepreneurial.
AI handles the overhead now. The research that required three associates takes twenty minutes with the right AI tools; the financial model that needed a dedicated analyst is a conversation with Claude; the proposal document, the compliance check, the market sizing—all achievable by a single experienced professional with AI augmentation, producing output that meets or exceeds what the firm's team delivered (maybe not on the first try, but iterations are so much faster that it doesn't matter). The partner still has the client relationship, the domain expertise, and the judgement that the client is actually paying for. What the partner no longer has is a reason to share 60% of their billings with a firm.
Seems outrageous? Ask Zack Shapiro. Or even McKinsey, who have been shouting from every rooftop that they now "employ" 20,000 AI agents (every single one of them a live data breach, apparently).
The power dynamic in professional services has always rested on an asymmetry: the firm controlled the infrastructure that individual practitioners couldn't replicate.
A senior professional outside a firm faced multiple friction costs: marketing and business development, bookkeeping and financial reporting, IT infrastructure, compliance and regulatory filings, proposal writing, document production, practice management. Collectively they represented enough overhead that maintaining a solo practice was a full-time (or more!) administrative job on top of the actual professional work. "Better together" was a no-brainer: join a firm, share your revenue, and let someone else handle the machinery.
Now list those same functions and ask which ones AI handles credibly today. Marketing? AI drafts content, manages social presence, and personalises outreach, as well as helping you make sense of your social media analytics. Bookkeeping? Drop in a folder of scanned receipts, and tax lodgements are done. Proposals? Drafted and formatted in minutes. Document production? Better than most junior associates. IT support? You've never had such a patient helpdesk.
If you believe AI can handle your back-office functions for your firm, the same AI you're using to justify the headcount cuts to those functions that improve your margins, then you must also believe that AI can handle those same functions for a solo operator.
The experienced professional considering independence no longer faces a choice between 'do the work I love and share my billings' and 'do the work I love but also work a second full-time job as an administrator'; they face a choice between sharing their revenue with a firm that provides diminishing value, or keeping it and letting AI handle what the firm used to provide. This isn't theoretical—it's already happening, quietly, in every professional services firm with senior practitioners who can (ask AI to) do the arithmetic.
What does AI efficiency actually mean for a professional services firm built on hourly billing? Hourly billing dies. If a task that took ten hours now takes one, you cannot bill ten hours with a straight face. The market won't allow it, because the first competitor to offer a fixed-price, AI-augmented alternative will make your hourly rate look absurd. The service has to be productised: a fixed-price deliverable, a platform, a subscription.
A firm that productises first operates at a fraction of the cost base of any competitor still billing by the hour; it charges high margins initially (even undercutting old-school competitors by half, it might be making more margin than them), takes market share, and builds a war chest for further AI investment. A firm that productises second discovers it's losing clients to a cheaper, faster alternative, and the AI investments it is now scrambling to make are being funded from declining revenue. The asymmetry creates an overwhelming first-mover incentive—every firm leader understands this. What fewer have reckoned with is the dependency it creates.
The knowledge that needs to get productised lives in your senior practitioners. They built the methodology; they know which shortcuts work and which produce garbage; they've seen enough edge cases to know where the standard approach fails. You need them to build the product version of the service, and those practitioners, for precisely the same reasons that make productisation possible, can walk out and do it themselves. Any time. Now.
This isn't a hypothetical leverage point, it's the same logic the firm is using to justify its own AI investment. If the firm believes AI can productise the service, the practitioners can believe it too; they're either both right or both wrong. You can't have it both ways.
A vampire firm that rolls out AI aggressively, trains its top people to mastery, and runs them at 10× productivity will lose its senior practitioners faster than any other firm. Faster than firms that adopted a more moderate pace, faster even than firms that didn't invest in AI at all, because those firms never demonstrated to their people exactly how much they were capable of producing independently. The compounding effect cuts both ways: the more clearly you show your senior people what they can do with AI, the more clearly you demonstrate that they don't need you.
The resolution, for firms that manage it, is sharing enough of the upside with key practitioners to make staying worthwhile: equity, profit share, genuine autonomy, reduced hours. But now this negotiation happens from a position where the practitioners have more leverage than they've ever had, and many leaders have never had to negotiate from that position—their entire career, the leverage asymmetry has been firmly to the firm's advantage. Firms that treat it as a standard employment negotiation, like the one they are used to, or even like a business-as-usual "scarce talent" negotiation (like paying more to hire COBOL programmers), will lose. It's now at least a conversation between equals.
WHEN THE COVENANT BREAKS
The social contract for educated professionals (study hard, get qualified, join an institution, build a career) has been violated before, and there are some lessons to be taken out of those events.
When formal employment has collapsed at scale, such as Russia in the 1990s, Argentina in 2001, Greece and Spain during austerity, South Korea after 1997, educated professionals have consistently refused to accept inactivity. They scramble, trade, improvise, emigrate. In some sense, they become entrepreneurs, but not in the way the previous section describes.
Russian engineers didn't open engineering consultancies; they became shuttle traders, hauling Turkish leather jackets to bazaars. Korean engineers didn't start technology firms; they opened fried chicken franchises. South Korea now has 87,000 chicken restaurants, more than double every McDonald's worldwide. Argentine professionals joined barter networks that collapsed within months. More often than not, the most capable displaced professionals chose to leave rather than start businesses: Greece lost an estimated 280,000–500,000 educated citizens; Argentina lost 800,000; Russia lost 80,000 scientists in half a decade. Brain drain and entrepreneurship draw from the same talent pool, and the evidence says exit wins over enterprise when the option exists (i.e. when the problem is localised to one country, and there is another place you can go where the problem isn't).
Only South Korea showed genuinely persistent self-employment growth through economic recovery, because the underlying labour market institutions changed permanently: lifetime employment ended, mandatory early retirement in one's forties became standard, and there was no 'normal' to return to. In every other case, self-employment rates fell as formal employment recovered (although most emigrants did not return).
Every one of those historical cases shares a defining feature: the entire economic ecosystem was degraded. Clients had no money, currency was unstable, contracts were unenforceable. Even doctors in Russia were paid in barter, not because demand for healthcare fell, but because no one had any cash. The informal economy emerged because the formal economy was genuinely broken.
AI displacement is categorically different: it occurs within a functioning, growing economy. The organisations shedding professional headcount are doing so because they're prospering. If they're among the first movers, AI automation means higher margins, not economic collapse. Clients still have budgets; contracts are enforceable; currency is stable. The demand for professional outcomes (legal advice, financial analysis, engineering solutions) hasn't fallen; what's shrinking is the demand for the labour hours to produce those outcomes.
This creates a two-speed economy: organisations that adopt AI effectively become more profitable, while the professionals they displace enter a market that still has solvent clients, functioning institutions, and real demand for their expertise. The displaced professional's starting position is better than the Russian engineer's.
Two caveats. First, firms that internalise capabilities with AI don't automatically become customers of the people they displaced; a law firm that replaces eight associates with two associates plus AI hasn't created demand for six independent lawyers, and the displaced six need to either find clients who weren't previously served by the firm model, or serve existing clients at a price point the firm can't match. Both are plausible, but neither is automatic. Second, the prosperous two-speed economy may be a window rather than a steady state: early AI adoption generates windfall profits, but competitive dynamics should eventually compress margins back toward equilibrium. The displacement of professional labour is permanent even if the super-normal profits are transient.
Every historical analogy listed above has a key difference to what is happening with AI: the changes were cyclical or transitional, but AI displacement is permanent. If document review, routine drafting, financial modelling, and diagnostic imaging are automated, there is no 'recovery' that restores those roles at previous volumes. The only historical parallel is South Korea—lifetime employment ended and never came back. The roles AI displaces won't come back either.
This permanence means that whatever entrepreneurial or independent response occurs should be more durable than any historical precedent. There is no 'normal' to wait for; the professionals who build independent practices won't be pulled back into firm employment by a recovering job market, because the job market for those specific roles isn't recovering. It's been permanently restructured.
This is not a difficult realisation to come to, which might push more people to make a durable lifestyle change (such as self-employment) earlier rather than hold out for better times, living off savings while looking for replacement jobs.
In the historical cases, the most capable displaced professionals left, a brain drain created by the asymmetry between a country in economic strife and alternative countries with stable economies. In a globalised AI economy, all countries will be affected by AI displacement, but there are still differences that can create asymmetries.
How easy it is to register a business, how onerous the regulatory requirements, how accessible the professional indemnity insurance, how welcoming the tax regime all vary enormously by jurisdiction. Countries that make independent professional practice frictionless will attract the talent that higher-friction jurisdictions lose. The existing brain drain patterns may reverse or redirect as political headwinds shift; a senior professional displaced from a New York firm has to ask whether rebuilding independently is easier in the US, in Australia, in Singapore, in the UAE.
The jurisdictions that get this right gain a compounding advantage: every experienced professional who arrives brings client relationships, domain expertise, a network that attracts more (and, of course, tax revenue). Brain drain becomes a recruitment strategy for countries that move fast and a downward spiral for countries that don't.
THE NEW APPRENTICESHIP
Everything argued so far benefits experienced, relationship-rich professionals disproportionately. They have clients, domain authority, and the judgement to quality-control AI output; the leverage shift runs in their favour. But what about everyone else?
Mid-career and junior professionals face a harder landscape: fewer firm positions, no client relationships yet, and the traditional institutional ladder, where you used to learn the craft by doing senior partners' work under their supervision, being dismantled by the same forces that liberate the seniors.
The firm bundled several things together: training, deal flow, quality control, and personal brand. You joined the firm and got all four; you learned by doing supervised work on real projects; you got exposure to clients you couldn't have reached independently; your mistakes were caught before they reached the client; and the firm's name on your CV signalled competence. That bundle is now decomposable—AI handles some of it, reputation platforms handle some, some goes to direct professional relationships—but decomposable is not the same as decomposed. What institutional scaffolding replaces the firm's apprenticeship function?
An AI-augmented senior professional operating independently can handle extraordinary volume, but not everything, and not all the time. They need surge capacity when a large matter lands; a second pair of qualified eyes on high-stakes work; someone to handle tasks that exceed AI capability but sit below the senior's optimal use of time; and presence in places they can't be, be it court appearances, site inspections, or client meetings in other time zones.
They need juniors, but on flexible, project-based terms rather than as permanent employees with training obligations and fixed salary overhead (remembering that not having to support overheads was one of the drawcards of leaving a firm to begin with). This creates a natural economic relationship: the senior has client access and domain authority; the junior has time, energy, and a hunger to learn the craft in its new AI-augmented form. The senior is also an AI-augmented practitioner, and part of what the junior learns is how experienced professionals actually wield these tools—something no course or certification can teach. You learn it by watching someone do it on real work with real stakes.
Not everything a junior practitioner needs can be provided by the senior. Exposure to real projects and feedback, yes, but not the progressive credentialling that signals competence to the market, replacing the 'I spent five years at Firm X' credential, and not a path from executing under supervision to originating independently.
Credentialling requires mechanisms that validate competence to the market, not just to one or two seniors that you happen to have worked with. The path from supervised work to client origination requires stepping stones that the market recognises, not just the confidence of having done good work under someone's supervision.
Professional associations are an obvious institutional home for this—they already have credentialling authority, ethical oversight, membership networks spanning career stages, and, in most professions, mandatory membership. What they lack today is a matchmaking and quality-assurance function connecting independent seniors with project-based juniors.
Imagine a law society that operates a verified marketplace: senior independent lawyers post matter-specific briefs; pre-vetted junior lawyers accept engagements; structured feedback flows back to the credential system; and the association provides the professional indemnity umbrella that currently requires firm membership. Or a medical college that matches specialist consultants with registrars for project-based rotations; or an engineering institution that facilitates design-review partnerships between independent principals and early-career engineers.
This is a new institutional form—not the firm, not pure atomised independence, but a networked apprenticeship market intermediated by professional bodies. It replaces the firm's bundled apprenticeship model with something modular, flexible, and potentially more meritocratic: juniors build credentials through demonstrated performance on real work, not through years of seat-warming at a prestigious address.
AI itself can enable the matching: skills taxonomy, availability tracking, conflict checking, performance analytics; even in today's state of AI capability, if designed well it could dramatically lower the coordination cost that previously made this kind of distributed mentorship impractical. Professional organisations that move into this role become central infrastructure for the new professional economy; far more consequential than their current functions of early-career training, accreditation, networking events and CPD admin.
If professional associations don't fill this matching role, commercial platforms will. Upwork, Fiverr, and their successors are already moving upmarket into professional services, and we should probably care: professional associations have ethical obligations, disciplinary powers, and alignment with professional standards. Commercial platforms optimise for transaction volume and take rate. The quality of the resulting professional ecosystem depends significantly on which model wins.
This connects to the jurisdictional competition point above: countries whose professional associations move fastest to build these networked apprenticeship markets will attract both the independent seniors and the ambitious juniors; countries whose associations remain gatekeeping-focused and structurally conservative will lose both—to other jurisdictions, to commercial platforms, or to a degraded quality of professional practice.
CONCLUSION
The networked apprenticeship model requires someone to move first: either professional associations build the infrastructure before independent practitioners demand it in volume, or independent practitioners move in volume before the infrastructure exists to support them well, and both involve accepting significant risk on partial information.
Associations that wait for clear demand will arrive too late—the displaced professionals will have already solved the problem through commercial platforms, informal networks, or by leaving the jurisdiction entirely.
Independent practitioners who wait for institutional support will find themselves operating in a higher-friction environment than necessary, competing against those who moved earlier when client budgets were still adjusting to new price points.
First movers in both categories face obvious risks (associations that build infrastructure for a market that doesn't materialise, practitioners who jump before the safety net exists) but second movers face structural disadvantage: they're building or entering a market that's already been shaped by others' choices, and the question of who captures value may have largely been settled by the time the optimal decision becomes clear.
The coordination problem isn't solvable through patient deliberation; it requires someone, somewhere, to make a gutsy call and put a stake in the ground - hopefully, to keep vampires at bay.