Chapter 6 — What Software and Law Tell Us (plain-language version)
Why look at other professions
Consulting is still early in its encounter with AI tools. Software development and law got there first. In both cases, AI mediation has been operating at scale long enough to leave traces — in the published research, in what practitioners and researchers have written about what they’re observing, in how institutions are starting to respond.
This chapter reads those traces and draws out four findings that carry forward into the next chapter’s analysis of consulting itself. If the same pattern is visible across three different knowledge professions, that’s stronger evidence than one case alone.
The chapter also takes on two important counterarguments: the AI-washing critique (which says that some supposed AI displacement is really just ordinary cost-cutting dressed in AI language) and the substitutionist argument (which says AI tools will simply replace professionals, and the question is just how much of each job they’ll take). This chapter engages both directly.
The software case: the canary in the mine
Software developers had AI tools at scale before almost anyone else. The tools arrived from around 2022 onwards — code generation, automated test writing, boilerplate production — and the published evidence now includes a substantial body of research on what happened.
Three things stand out.
Selective displacement. AI tools absorbed specific components of software work — the routine code generation, the repetitive structures, the things that needed to be written but didn’t require judgment. Other components — the architectural decisions, the debugging of subtle interactions, the judgment about what the code should be doing — were much less affected. The displacement was real but partial.
Pipeline disruption. Junior developers have historically been formed by doing the routine work themselves — building things from scratch, making mistakes, having those mistakes corrected, slowly developing an instinct for how code behaves. The published research shows that this formative process is being structurally altered. Junior developers with AI tools produce competent code faster. But the evidence suggests that the formative engagement — the struggle, the failure, the building of judgment through difficulty — is being compressed. What they produce looks fine. What they’re developing is different.
The labour-market complication. Between 2024 and 2026, the technology sector saw substantial layoffs. Research into those layoffs found disagreement: some displacement was clearly AI-related; some was cost-cutting that firms were describing in AI terms because it sounded better than “restructuring.” This is the AI-washing problem. It matters because it means you can’t just look at job losses and conclude you know what AI is doing to the profession. The more important evidence is about formation, not headcount.
The law case: a more regulated profession
Law is interesting for a different reason: it’s a formally regulated profession. There are bar admission requirements, structured training frameworks, mandatory continuing education, ethical codes. All of that institutional architecture makes the formation question more visible — you can see it in the institutional debates that regulated professions have in a way that consulting, which is less formally structured, currently doesn’t.
The evidence from law follows a similar pattern to software. AI tools have absorbed document drafting, contract review, legal research, and due diligence. The displacement is selective, concentrating on components that are high-volume and relatively structured. The formation pathway through which junior lawyers develop contributory expertise — the sustained engagement with actual legal problems, the development of judgment — is being altered.
What law adds is institutional response. Bar associations, law schools, and regulatory bodies have started to develop explicit governance: disclosure requirements for AI-assisted work, training standards that address AI, guidance on professional ethics in AI-mediated practice, and — most importantly — early discussions about how to protect the formation pathway. Regulated professions can do this. Law is showing consulting what institutional responses might look like, a few years ahead.
A study of medical training (which has the same structure) adds sharp numbers: 55% of surgical residents in 2023 reported reduced skill development due to AI-assisted decision support. That’s not a theoretical projection. That’s an active empirical process. The medical literature uses the phrase “upskilling inhibition” — not the loss of skills you already have, but the failure to develop skills you should be building. That’s exactly the formation concern.
The AI-washing critique: why it doesn’t undermine the argument
Some people argue that the narrative of AI displacement is being used to disguise ordinary cost-cutting. The argument goes: firms are laying people off because interest rates went up, or because a growth phase ended, or because they overbuilt during the pandemic years — and they’re framing it as AI-driven because that sounds like strategic transformation rather than financial pressure.
This critique is real. Some of this has definitely happened. The labour-market data is genuinely complicated.
But the framework’s argument doesn’t depend on the labour-market data. The concern here isn’t how many jobs are disappearing. It’s what the work is doing to the people doing it. Even if every single layoff between 2024 and 2026 was entirely driven by cost-cutting with no AI involvement at all, the evidence on formation pathway disruption would still stand. Junior developers and lawyers and soon consultants are engaging differently with their work — and that engagement is what builds expertise over time.
The AI-washing critique complicates the job-loss story. It doesn’t touch the formation story.
The Susskinds: a different question
Richard and Daniel Susskind, in The Future of the Professions, argued that professional services will increasingly be delivered through AI-mediated systems, with human practitioners displaced into narrower roles. It’s an influential account, and it’s about service delivery: who delivers what, and what professionals’ roles become as AI matures.
This thesis disagrees — not about what’s happening, but about which question is most important.
The Susskinds ask: how will professional services be delivered? This thesis asks: what is happening to the formation of the practitioners who deliver them? These are different questions. You can’t get from the Susskinds’ question to the formation question; they’re at different levels.
Three specific differences:
What counts as expertise. The Susskinds treat professional expertise mainly as the institutional architecture through which services have historically been delivered — and argue that AI can replicate enough of it to transform that architecture. This thesis treats expertise as a capacity built through sustained engagement with difficult, consequential work. You can’t replicate that capacity by replicating the outputs it produces. The lift delivers you to the sixth floor; it doesn’t build the legs that climbing would have built.
What the normative question is. The Susskinds are broadly neutral about the transformation they describe — they treat it as a question of adaptation. This thesis treats the European public-interest tradition as something worth preserving, and the formation question as a question about whether we’re preserving the conditions under which that tradition can continue to be exercised. Those are different postures toward the same set of facts.
What the analysis can see. The Susskinds’ analysis can see which professional services get replaced by AI tools. It can’t see what happens to the pipeline that produces the people who exercise judgment in the ones that don’t get replaced. That’s precisely the question this thesis is asking.
Four findings that carry forward
The chapter synthesises four findings from the comparative evidence. These carry into the next chapter’s analysis of consulting across the three career stages.
Finding 1: Selective displacement. AI mediation doesn’t replace whole jobs. It absorbs specific components — the routine, the structured, the high-volume, the things that are easiest to automate. The important thing is which components: they tend to be exactly the ones that form junior practitioners through engagement with them. The proletarianisation doesn’t abolish the worker; it changes what the worker does, and specifically what the worker is becoming through what they do.
Finding 2: Pipeline rupture regardless of headcount. The formation pathway is being disrupted whether or not employment levels are stable. A firm can have the same number of junior consultants it had five years ago and still have a ruptured pipeline — because what those junior consultants are doing has changed in ways that alter what they’re developing into. Headcount stability doesn’t protect against formation disruption.
Finding 3: Uneven upward value flow. AI mediation redistributes value within the profession. What used to be worth money at junior levels is now available more cheaply through AI tools. Value flows upward to senior practitioners (whose situated judgment can’t be replicated) and outward to the platform vendors who own the AI systems. The distribution isn’t neutral.
Finding 4: Explicit narration of professional legitimacy. When interactional competence — the ability to produce outputs that look right — becomes available through AI tools, the profession can no longer rely on outputs alone to demonstrate legitimacy. Legitimacy has to be explicitly narrated: who did this work, what AI was used, what was the human practitioner’s actual contribution. The legal profession’s institutional responses — disclosure requirements, attribution norms — are early examples of this shift.
Three independent lines of evidence point the same way
The framework in Chapter 3 is philosophical. But the same concern now appears in three independent places:
Empirical research on software and law shows the formation pattern operating in practice.
Cognitive science shows that AI-assisted work involves measurably weaker neural engagement than unassisted work — and that this persists even after AI assistance is withdrawn. The tools don’t just save time; they change how deeply the work is processed and retained.
Formal economic modelling (Acemoglu, Kong and Ozdaglar) shows mathematically that when AI substitutes for human effort in knowledge work, it can deplete the collective stock of expertise that makes future expert work possible — even as individual outputs continue to look competent. The formal model produces the same structural concern the philosophical framework names.
When a philosophical argument, empirical pattern, cognitive science, and formal economics all point in the same direction, that’s worth taking seriously.
What comes next
Software and law together show: selective displacement, formation disruption, uneven value redistribution, and the requirement of explicit legitimacy narration. Chapter 7 carries these four findings into European infrastructure consulting specifically — what the same pattern looks like across the Foundation, Applied, and Chartered career stages.