Chapter 9 - Conclusion: Recommendations and Residual Questions

9.1 What this work has delivered

This chapter 💡 can do about it, 3/ tries to say what this work has not done.

The first-person voice is more present here than in the analytical chapters, deliberately: the conclusion is where the practitioner-researcher voice the methodology rests on speaks most directly. The recommendations answer to four of the seven cross-scenario findings from Chapter 8: the formation question, the sovereignty pressure, the third-box question, and the requirement of explicit legitimacy narration. Those four hold whichever trajectory unfolds, so the recommendations do too.

9.2 The central claim and the contributions

This work has advanced a single central claim, which I restate here in the form it has occupied throughout: Generic AI mediation, under current conditions, risks proletarianising the situated, answerable judgment on which the legitimacy of European infrastructure advisory rests, and on which its obligations to publics and futures depend.

Our eight chapters have assembled, stress-tested, and deployed the framework: the philosophical foundations (Chapters 2 and 3), the domain materialisation and political-economic analysis (Chapters 4 and 5), the comparative evidence and stage-by-stage core (Chapters 6 and 7), and the futures analysis (Chapter 8). Together they produce what the existing literature on AI in consulting does not yet provide 💡 : a philosophical account of what is at stake in the profession’s transformation, one that addresses formation, sovereignty, and the third box together.

The six contributions claimed in the introduction have been delivered as follows:

1. Stiegler’s proletarianisation argument applied to management consulting has been delivered. The analytical chapters develop it and show that proletarianisation as selective displacement of savoir-faire operates in the consulting profession, that the three-scale individuation argument (individual, firm, national-cultural tradition) gives the framework its conceptual unity, and that the application yields analytical purchase the literature on AI in consulting lacks. 💡
2. Jonas’s imperative of responsibility applied to AI-mediated advisory accountability in infrastructure is the framework’s second contribution. The argument about what advisers owe publics and future generations is grounded in Jonas. It shows that the third box, the constituency to which advisers answer, is institutionally available within the European public-interest tradition, that the answerability of advisory work to publics that bear consequences is the ground of legitimacy in the domain, and that this answerability is structurally at risk under contemporary AI mediation.
3. The ChMC competency framework mapped against AI substitutability across the three stages is provided here: to my knowledge the first systematic mapping of its kind, with the four ChMC competencies deployed analytically at each stage. The mapping tries to give the framework a stage-by-stage diagnostic that supports the differentiated recommendations developed in this chapter.
4. Pipeline rupture theorised as a cross-professional philosophical mechanism has been established across the adjacent professions. Chapter 6’s comparative evidence shows the mechanism at work in software and law, and engages the AI-washing critique directly: the framework’s argument addresses formation rather than headcount. Chapter 7 demonstrates the same mechanism in European infrastructure advisory. The cross-professional pattern is consistent enough to support the theoretical framing.
5. The dual-scale framing of sovereignty as identity-formation under technical mediation has been built and applied at the national-cultural and organisational scales. Chapter 5 connects the two through the same Stieglerian mechanism, and the connection is itself the analytical contribution: sovereignty operates at more than one scale, and the same dis-individuating pressure acts at each.
6. Four scenarios for European infrastructure advisory as philosophical thought-experiments are the contribution Chapter 8 provides. The scenarios (Commodity Advisory, Digital Twin Economy, Artisanal Advisory, Disintermediated Clients) are constructed as thought-experiments in the Williamsonian sense rather than predictions. Together they identify the cross-scenario findings that hold regardless of trajectory, and so ground the recommendations developed in this chapter.

I claim these six as original contributions. I’d say that their combination, working together through the framework this work has developed, is itself a philosophical contribution to the analytical conversation about AI in European infrastructure advisory.

Before developing recommendations, I’d highlight the failure mode that institutional responses to this argument most readily fall into. Debord described it as récupération: the spectacle’s capacity to neutralise critical challenge by converting it into spectacular form. The critique produces a representation of critique, which satisfies the demand for opposition without delivering it. In the advisory context, the failure mode takes a recognisable form. A recommendation that results in a responsible AI use policy, a human oversight requirement, or an editorial judgment protocol has been recuperated. The governance response produces the representation of a solution: the junior validates the model’s output; the senior approves the validated output; the policy specifies that human judgment must be applied; the firm records the application. Each step is a staged performance of the engagement the critique demanded. None of it restores the temporal structure of formative engagement.

The failure mode is already observable at the scale of the sector. Blangeois and Lebrument’s earnings-call study (Chapter 4) covers firms whose business, like advisory’s, rests on billing human effort 1 . Their calls tell a coherent “dual-front” story, outward to clients as innovation and inward as mastery of the firm’s own productivity, which the authors read as “a rhetorical performance”, a “discursive coping mechanism designed to project coherence, control, and legitimacy” while “deferring a direct confrontation” with what the technology does to the firm’s own model 1 . The performance, they conclude, can “create a buffer between the demands for external legitimacy and the complexities of internal operations” 1 . That buffer is récupération observed empirically: the representation of engagement, produced where firms speak to markets and decoupled from what happens to the work. In the narrative they reconstruct, the workforce appears as a number to be doubled, “from 40,000 to 80,000-strong” 1 , not as a pipeline to be formed. Inside the firm, the same decoupling reaches down to the engagement 💡 .

The junior’s validation is spectacular validation. They receive the model’s output and apply their current judgment to it, judgment formed by the same AI-mediated conditions the validation is meant to correct. No governance framework can specify the thing the argument requires: the time of real formative difficulty, the unassisted encounter with analytical challenge, the sustained engagement with consequence through which practical wisdom develops. We try to hedge the recommendations in this chapter against this failure mode 2 .

9.3 What this means for consultants

In their general form, the recommendations below hold for practitioners in any advisory context where AI mediation has begun to absorb the formative base of the pipeline. The obligations 💡 are available to advisory practitioners across contexts: strategy consulting, financial advisory, and management consulting generally, wherever formation has been disrupted in the ways Chapter 7 describes. The rest of this section develops these obligations in European infrastructure advisory, where the content of formation, the institutional setting, and third-box accountability give them their most demanding form. Readers in other fields should read the European-specific content as one instance and extract the general obligation accordingly.

The recommendations in this section track four of the seven cross-scenario findings, those named in §9.1. For individual consultants, they differ by ChMC stage.

Brussels, 2022. A Foundation-stage consultant is assigned to accompany a senior partner to a European technical working group on cross-border energy infrastructure governance. Her formal contribution to the session is minor. The partner considers leaving her behind: the preparation is substantial, the session complex, and she will not speak. He brings her anyway, with a specific brief: watch what happens when the official chairing it asks a question we have not prepared for, and watch how I handle what I do not know. The session runs three hours. The official asks, forty minutes in, a question about the interaction between the Energy Union framework and a specific bilateral treaty obligation that nobody in the room has anticipated. The partner handles it. In the taxi afterwards, they spend twenty minutes talking about it: what he did, why, what he could not say and why, what the question revealed about what the official actually needed. Three years later, now Applied-stage, she describes that conversation in a session on professional development as the most formative single afternoon of her career. She is not sure she could have had it via a briefing document. She is certain she could not have had it via an AI tool 💡 . When she staffs meetings now, she takes the junior who will not speak. 💡

The first recommendation is deliberate formation. Formation is the ground on which contributory expertise is reproduced across generations, and contemporary AI mediation acts on that ground in ways the productivity framings cannot see. Foundation-stage consultants should seek out work that develops contributory expertise through sustained engagement with consequence, even where AI-mediated alternatives would be faster 💡 . Applied-stage consultants are the ones forming Foundation-stage juniors, and that supervisory role asks more of them than AI-mediated review 💡 . Chartered-stage consultants carry the formation responsibility their position confers: they definitely should keep in mind what they teach the next generation, and what they refuse to delegate to AI mediation where formation is occurring.

The second is reflexive practice: the discipline of examining one’s own engagement with AI mediation as part of the work itself. For practitioners at any stage, this means a working awareness of when one’s own analytical contribution is being mediated, what the mediation is doing to the work, and what is being preserved or eroded in one’s own contributory grasp. Reflexive practice stops short of refusing AI mediation: it requires consciousness of what mediation is doing and what it is not doing, and the cultivation of judgment that survives mediation rather than being absorbed by it 💡 .

The third recommendation is refusal of unattributed mediation. Where AI mediation operates in advisory work, explicit attribution is a professional obligation. The Foundation-stage consultant whose draft has been substantially produced through AI mediation owes attribution to the consultant and the team who review it. The Applied-stage consultant owes attribution 💡 to the client who receives a deliverable so produced. And the Chartered-stage consultant whose advice was so produced owes attribution to the public who bears the consequences. Attribution is a matter of professional integrity, and it has institutional weight: without explicit attribution, the legitimacy of advisory work is structurally undermined in ways Chapter 6’s fourth finding identified.

A senior regulatory adviser in a UK infrastructure practice has, over fifteen years, developed a way of framing engagement preparation that does not appear in any methodology document: she consistently asks, for every engagement, what is the regulatory body trying to protect itself from? (not what it wants, not what the legislation mandates, but what institutional fear is shaping its behaviour right now). It is a question that produces different analytical entry points than standard stakeholder interest mapping, and it has been right often enough that her clients have come to expect it as a distinctive feature of the firm’s work. She has never written it as a framework and does not teach it through a slide, teaching it instead by working alongside people. She tested it once, informally, against an LLM’s stakeholder analysis for an engagement she had already completed. The model produced a correct and comprehensive answer. It had not asked what the regulator was afraid of. The question had not occurred to it. Since then she asks the juniors she works with to write that question at the top of their preparation notes, in their own words, before they open any tool. 💡

The fourth recommendation is to cultivate judgment that resists generification. The distinctiveness of the practitioner’s analytical voice is part of what makes contributory expertise valuable, and generic AI mediation presses on exactly that. So cultivate it, deliberately: the framings, vocabularies, and ways of seeing that one brings to engagements and that would not survive uncritical AI mediation. This is a professional commitment as much as a strategic one. The practitioner whose distinctive voice is preserved is making a contribution that the practitioner whose work is fully generified cannot make. In practice this comes down to knowing which parts of the work that voice comes from, and continuing to do those parts oneself.

The fifth, which Chartered-stage consultants carry most directly, is responsibility for the next generation. The formation pipeline is at risk under contemporary AI mediation, and what European infrastructure advisory is in twenty years depends on the formation pathways that current senior practitioners support or fail to support. Senior responsibility extends beyond the consultant’s own work to the practitioners who will succeed them: invest in the formation pathways within one’s firm, take the supervisory role seriously, and recognise that the firm’s capacity in two decades is being decided now. Juniors also learn what a senior is by watching one (Chapter 7, §7.4): a partner who checks the model’s draft, signs and forwards it is teaching that this is the role, whatever the training programme says.

These five recommendations differ by stage but cohere on the responsibilities constitutive of legitimate advisory practice. They are addressed to consultants who recognise the concerns named here and choose to take them seriously.

9.4 What this means for advisory firms

The same logic holds for any advisory firm whose formation pipeline is being eroded at its base by AI mediation. Preserving formation pathways, protecting organisational knowledge from extraction, cultivating a distinctive firm voice, attributing AI-mediated work explicitly, and engaging at industry level on the formation question are choices every advisory firm faces, whether in infrastructure, strategy, financial services, or management consulting broadly. This section develops these choices for European infrastructure advisory; the general institutional logic is available across the advisory landscape, applied with the gradient of third-box obligation that each sector provides.

Paris, 2024. An infrastructure advisory practice is evaluating three enterprise AI platforms for integration into its workflow. The evaluation criteria are capability, price, integration with existing systems, and (at the managing partner’s insistence) data governance. Her single non-negotiable: no client analytical content used in vendor training data, contractually enforced, with audit rights written into the agreement. One of the three vendors cannot meet the condition at any price. One meets it at a significant premium. One meets it at standard enterprise pricing but only for structured data, not unstructured client documents. The firm takes the expensive option. Three months later, the managing partner includes the data governance clause as a specific item in her annual presentation to the firm’s client advisory board: evidence that the firm takes the confidentiality of client strategic material seriously in an era when the boundary between workflow tool and data collector is not always clear. Two clients mention it unprompted in contract renewal conversations. Internally she puts it another way: the clause protects the client’s material and also the firm’s own way of reading a problem, which is what the training data would otherwise carry away. 💡

The first recommendation is to preserve formation pathways deliberately. Contemporary AI mediation acts on the pipeline through which contributory expertise is reproduced, and how a firm integrates AI into Foundation-stage and Applied-stage work decides whether that pipeline is preserved or eroded. Firms should make these choices explicitly: which engagements are routed through formation tracks that preserve formative engagement with the work, what supervisory practices are maintained for Foundation-stage consultants, what investments are made in Applied-stage development beyond AI-mediated review.

Productivity framings do not see these choices, which are consequential all the same: who a junior works beside is, in Ide’s model (Chapter 7), what decides whether tacit knowledge passes on 3 . Many firms are already cutting entry-level hiring without a real plan for what the cut will do to their partnership pipeline 4 . Commercial logic is dismantling the pathway without replacing it, and an institutional choice is the only counterweight available.

Garicano’s ‘AI Becker problem’ gives the economics 5 . Junior analytical work has historically cross-subsidised formation through the leverage model: the revenue juniors bring in pays for their training, which makes forming them rational for a firm even when a competitor may poach them. Automation removes the subsidy together with the formative work. No single firm then has an incentive to invest in formation, and the industry cannot correct this through the market alone; the recommendations to chartering bodies and policymakers are, economically, institutional answers to that failure.

The second is to make tooling decisions to protect organisational knowledge from extraction. Chapter 5’s extraction argument showed that the structural pathway through which firm knowledge can flow outward into vendor systems is real, and that how far it is realised in practice varies with the vendor, the contract, and the deployment configuration. Firms should take this pathway seriously when they choose tools: negotiate vendor contracts that explicitly exclude customer data from training where this is achievable, use deployment configurations that mitigate the risk, and treat the firm’s externalised knowledge as an asset worth protecting. These decisions are institutional rather than technical, and belong to the political-economic landscape Chapter 5 described.

The third is to cultivate the firm’s distinctive voice as a strategic asset. Generic AI mediation, Chapter 5’s organisational sovereignty argument showed, puts dis-individuating pressure on a firm’s own analytical content and, as it accumulates, puts at risk the way of seeing, the methodological reflexes, and the sectoral instincts that make one firm meaningfully different from another. That voice needs deliberate cultivation: senior practitioners whose analytical voice is recognisably theirs, institutional practices that reproduce the firm’s own content across generations, and explicit recognition that the firm’s value lies in the content Chapter 4 described rather than in the productivity AI mediation can provide. Such a voice is carried by the people the firm forms (§4.5): a style guide or a prompt library can record it, and on this work’s account neither renews it once formation stops.

The fourth mirrors the third recommendation to consultants: explicit attribution, at firm level. Refusing unattributed mediation has an institutional correlate: firms can develop attribution practices that make the use of AI mediation transparent to clients and to the firm’s own practitioners. These practices vary (disclosure on deliverables, attribution conventions in engagement reports, internal practices for Foundation-stage formation) but the commitment is the same: the legitimacy of the firm’s work depends on explicit narration of how the work has been produced, and how the firm narrates it is a choice the firm makes. It is also a choice about what disclosure costs the person who makes it. In Mizrak and colleagues’ survey (Chapter 7), concealment of AI use is driven by the fear of looking less competent rather than by an intent to deceive, and they read it as “a technology-related form of workplace silence” 6 . Of the interventions they compare, the one ranked first under every weighting is a non-punitive disclosure system, which pairs the expectation of transparency with “credible assurance that responsible disclosure will be evaluated fairly”; their practical forms of attribution include documentation proportionate to the AI’s contribution and, for consequential work, “short oral explanations, follow-up questions, scenario discussions, or requests to defend key assumptions” 6 . An attribution practice that penalises attribution produces its own representation and little else: the rule exists, and the use goes underground. 💡

Brussels, 2025. A European infrastructure advisory firm contributes to a professional body’s working group on AI and professional development (one of four firms invited to share practice experience with the body). The firm’s written contribution argues that the body’s current CPD guidance on AI literacy is addressing the wrong problem: it trains qualified practitioners in responsible AI use and governance compliance while leaving entirely unaddressed the question of how Foundation-stage formation functions when AI mediates the analytical work through which formation has historically occurred. The contribution does not receive a formal response from the working group’s secretariat. Three months after the submission, the body announces a consultation on its competency framework, with formation pathway review identified as a priority topic. The firm is not sure whether its paper had any effect, but it is sure the conversation has started, and that no single firm could have started it alone.

The fifth recommendation is to engage with the formation question at industry level. The formation pipeline is not a firm-level question alone. It runs across the profession, and each firm’s choices interact with the sector’s institutions: chartering bodies, professional associations, training and accreditation frameworks, regulators. Firms should work with these bodies: contributing to chartering-body standards that recognise AI-mediated practice, taking part in industry-level policy development on disclosure norms, supporting the structures through which the profession reproduces its capacity. The question cannot be solved firm by firm; it requires engagement at several levels at once. Any governance response here, a disclosure requirement among them, risks the récupération §9.2 described. The industry-level engagement this recommendation calls for is formative: investment in the formation pathways themselves rather than in the machinery of oversight 2 .

Lovett, whose cognitive-commons account Chapters 7 and 8 drew on, gives a name to what the fifth recommendation asks of firms. He argues that human-resource development “must expand beyond organizational expertise optimization to encompass profession-level commons stewardship”, because expertise is “a collective resource that exists at the profession or societal level” and no single firm can regenerate it alone. He places the governance at three levels, “organizational, professional-association, and policy”, the levels this chapter’s recommendations also reach, from the firm through the chartering bodies to policy 7 . Stewardship is the right word for the firm’s part. A firm that protects its formation pathway is not only investing in its own future seniors; it is maintaining a stock the whole profession draws on, competitors included, and it should say so when it argues, as the Brussels firm did, for standards that make the stewardship shared rather than voluntary. 💡

Together, these choices shape whether the integrated three-scale system survives the transformation contemporary AI mediation has begun to produce. No single firm’s choices resolve the formation question or the extraction risk; the recommendations here identify what each firm can contribute within the constraints of competitive pressure and commercial logic. 💡

9.5 What this means for infrastructure clients

Infrastructure clients are a specific form of advisory client: organisations with substantial public-interest obligations, long time horizons, and a third-box accountability that sets them apart from clients in more commercial settings. The recommendations that follow apply most directly to clients in infrastructure, public-service, and regulatory contexts. In general form 💡 they apply wherever advisory clients are sophisticated enough to tell the deliverable from the judgment behind it. The European infrastructure content gives them their most specific form.

Vienna, 2025. A national rail operator has established an internal AI centre of competence for regulatory and strategic analysis. The team of five has spent eighteen months building a library of the operator’s own regulatory intelligence: past submissions, regulator correspondence, expert reports, outcome data from price control reviews going back to 1999. They are fine-tuning an internal model on this corpus. They are not attempting to replace external advisory input for major regulatory proceedings. They are attempting to be a more intelligent client: to arrive at advisory engagements having completed the preparatory analysis themselves, so that the advisory conversation can begin at the strategic question rather than the background briefing. The advisory firms who work regularly with the operator have noticed that the initial sessions run shorter and the questions arrive harder. One senior partner, after an unusually demanding opening session with the regulatory affairs director, tells a colleague: they have done our homework. Now they want to discuss the exam. Asked over coffee what the centre is doing to her own junior analysts, who no longer assemble the background briefings, the director has no answer yet; nobody has looked. 💡

The first recommendation is procurement norms that distinguish judgment from output. AI mediation produces interactional fluency far more readily than contributory expertise 💡 , and the value of advisory work rests on the contributory expertise that deliverables alone cannot demonstrate. Clients should procure for the judgment that advisory work brings, not only for the deliverable that demonstrates it, and build procurement frameworks that tell the two apart. These frameworks vary by subdomain and by client, but the logic is constant: clients who procure for judgment support the pathways through which judgment is reproduced, and clients who procure only for output accelerate the dis-individuating pressure.

Benjamin’s counsel/information distinction sharpens this recommendation: the client is procuring counsel (practical wisdom transmissible only from one Erfahrung to another) rather than information, which any adequately calibrated model can produce. The test is whether the deliverable bears the marks of the adviser’s sustained engagement with the specific regulatory tradition the project answers to, not whether it is technically correct: whether what arrives carries the authority of Erfahrung or only the adequacy of generated information 8 .

Procurement for auratic advisory work 💡 means procuring the adviser’s embeddedness in the tradition the work answers to: the marks left by sustained formation within a specific institutional context, rather than the delivery of accurate outputs. The test is the one just stated, and neither prestige nor firm size settles it 9 .

The second is disclosure norms that recognise the third box. Chapter 5’s sovereignty argument and Chapter 7’s stage-by-stage analysis both show that the third box (the publics who bear the consequences of infrastructure decisions, developed in Annex B) is structurally less visible under contemporary AI mediation, and that the consultation and submission processes through which the third box is heard require explicit disclosure to function. Clients should make the use of AI mediation visible to the publics whose interests are engaged: in public consultation processes, in regulatory submissions, in the documentation that infrastructure decisions are recorded in. Disclosure of this kind is how the third box’s standing is recognised in practice; it is not a compliance matter.

The third recommendation concerns in-house capability. The disintermediation dynamic Chapter 7 examined is partly driven by client decisions to build in-house AI capability, and those decisions shape what the wider professional ecology becomes. Clients should engage strategically with their own capability development: recognise that it affects an ecology they depend on, that the advisory expertise reproduced there has value they cannot reproduce internally, and that procurement and capability decisions together shape whether it goes on producing the expertise they themselves benefit from.

The fourth follows from the third: audit the formation consequences of in-house capability. As clients build in-house AI capability, the analytical work that has historically fed Foundation-stage formation can be moved in-house. Clients should examine whether their capability development is systematically compressing the scope of Foundation-stage engagement they commission: whether the engagements they retain for advisory firms are, as a consequence of in-house AI tools, increasingly the high-judgment tasks that Chartered-stage practitioners can do, with nothing left that develops Foundation and Applied practitioners. The client who builds excellent in-house AI capability while eliminating the scope of advisory engagement that sustains the wider ecology’s formation pipeline is building on ground that is being eroded beneath them 💡 .

For clients, the levers are procurement, disclosure, and capability decisions: the demand-side architecture of professional formation. The client who recognises that the advisory ecology it depends on is being altered by AI mediation (including by its own procurement choices) has both the information and the levers to engage that alteration.

9.6 What this means for European policymakers

The general form of the policymaker recommendation applies wherever public policy has an interest in the professional formation of advisory practitioners who serve public-interest clients: treat the formation question as a policy concern rather than a private firm matter, and use existing governance instruments to protect formation pathways from being eliminated without replacement. The European case is developed here because it is this work’s primary research site and because the EU offers the most developed instruments available for the purpose. The general logic applies in other jurisdictions and regulatory contexts, with only the available instruments changing from one to another.

The recommendations to European policymakers answer to the same cross-scenario findings, but they engage the policy instruments that set the wider conditions of European infrastructure advisory: the landscape Chapter 5 described.

The dominant operative framework for European AI governance 💡 has crystallised around principlist approaches synthesised most influentially by Floridi 10 : the five principles 💡 . The recommendations I develop in this section sit alongside this framework. The principlist apparatus addresses the procedural level (what AI systems ought to do) and does real work in the policy contexts where it is deployed. The recommendations I make engage the level the framework works at: what European public-interest infrastructure advisory is constitutively for, and what protects that content institutionally under contemporary AI mediation.

Paris, 2025. The national authority that oversees public investment publishes revised procurement criteria for AI-assisted advisory tools used in engagements that inform public infrastructure investment decisions above a defined threshold. The criteria require that such tools: use training data that does not include confidential French public-sector content without explicit consent; provide explainability of outputs on request from the commissioning public authority; and be subject to French data protection law for all data processed in the engagement. The criteria do not specify European ownership, French manufacture, or public-sector operation. They specify epistemic and accountability conditions: what data trained the model that informed the recommendation, and whether the reasoning behind the recommendation can be interrogated. A legal adviser reviewing the criteria describes them to a client as “the EU AI Act applied to advisory procurement, before the AI Act applies to it.” They are, in the vocabulary this work develops, a partial engagement with genuine sovereignty rather than procedural sovereignty alone: not asking whose tool it is, but asking what it knows and whether it can be held to account.

a. Learning from defence

The first recommendation is to extend the strategic-autonomy framing from defence to other infrastructure subdomains. Chapter 5 (§5.5) showed the sovereignty logic to be already institutionally available in defence. Policymakers could/should extend it wherever the European public-interest tradition is constitutive of a subdomain’s content. The extension does not require defence-grade arrangements, only arrangements fitted to how that logic applies in each subdomain.

Acemoglu, Kong and Ozdaglar’s model (Chapter 5, §5.6) ends in a policy of the same shape 11 : a moratorium in which agentic AI is suppressed 💡 to rebuild the depleted stock of general knowledge, then a permanent cap on AI precision. Institutions that protect the European public-interest tradition and procurement that limits the substitution of AI for formative engagement do the same two jobs in this work’s terms. The convergence should not be overstated, for the reasons §5.6 gives: their welfare-economic framework reaches the policy on its own assumptions, while this work reaches it from advisory’s answerability to the third box, the publics present and future who bear the consequences of its decisions.

Ide and Talamàs’s analysis of global knowledge work qualifies the recommendations in this section in one respect 12 . Their model shows that a regional restriction on autonomous AI in knowledge work provides minimal protection to lower-skilled knowledge workers in the restricting region, while global coordination of autonomy constraints redistributes gains more equitably. The procurement recommendations below are qualified accordingly: European procurement frameworks that specify non-autonomous AI deployment conditions in advisory contexts are a necessary institutional move, but how well they protect the distinctiveness of the European public-interest tradition is improved by, and on the distributional dimension may depend on, engagement in international standard-setting on AI autonomy in professional advisory work.

The importance of procurement

The second recommendation is to build sovereignty into procurement. The literature on AI sovereignty, as Chapters 1 and 5 noted, works mostly at the procedural level and at the national or supranational scale, and rarely reaches how procurement is institutionally configured. Mügge’s analysis shows that EU AI strategy documents have themselves reproduced this limitation, treating jurisdictional independence as a proxy for the epistemic sovereignty that would actually protect European professional distinctiveness 13 . European policymakers should develop procurement frameworks for advisory services in infrastructure that recognise the European public-interest tradition, engage the third box institutionally, and protect the formation pathway through which contributory expertise is reproduced. These frameworks would vary by subdomain, but each would start from the same premise: procurement is one of the key instruments through which European public-interest practice is reproduced or eroded, which makes it more than a procedural matter.

Formation and policy

The third is to take up the formation question at policy level. The formation pipeline is not only a firm-level or sector-level question; it raises policy questions that European policymakers can engage: support for the chartering and accreditation bodies that reproduce professional capacity, educational policy that supports European public-interest professional formation, regulation that treats formation as a policy concern. The tradition has an identity of its own (Chapter 5, §5.2), and it lasts only as long as practitioners are formed inside it; a pipeline that runs through generic models trained mostly on framings formed elsewhere reproduces those framings instead, so formation policy is also sovereignty policy. This engagement is more diffuse than the procurement recommendation, but it matters. 💡

Integration

The fourth is to use the EU instruments that already exist. The AI Act’s high-risk system provisions (which apply to AI systems used in critical infrastructure sectors and impose conformity assessment, transparency, and human oversight requirements) provide a regulatory architecture for infrastructure contexts in which the formation dimension could be recognised 14 . The Green Deal’s sectoral governance for energy, transport, and water establishes regulatory frameworks within which formation-preservation requirements could be specified for advisory services in high-risk infrastructure decisions. Policymakers should use these instruments to carry the formation argument, rather than treating the formation question as requiring entirely new regulation. The instruments are already in place; the explicit link between high-risk AI deployment in infrastructure advisory and the formation pathway it disrupts is still missing.

Responsibilities

The fifth recommendation is to give the third box explicit institutional standing. The third box 💡 is the constituency at the centre of advisory answerability. Policymakers can give that standing institutional form through public consultation requirements, the institutions of regulatory practice, the explicit recognition of intergenerational responsibility in policy frameworks, and the engagement with publics that the European public-interest tradition has historically attempted but that contemporary AI mediation makes structurally more difficult.

9.7 The limits of what this work has done

The framework developed here is not philosophically complete. At least three limits need to be acknowledged honestly; they mark where our contribution stops:

The first limit is the single-domain character of the analysis. The framework’s content is built for European infrastructure advisory, and its analytical claims are made at that domain’s level. The analysis does not claim to be a general framework for AI mediation in consulting, in professional work, or in knowledge work. The comparative evidence in Chapter 6 supports the claim that the proletarianisation mechanism operates beyond infrastructure advisory, but the framework (the third-box constituency, the public-interest tradition, the dual-scale sovereignty argument) is specific to the European infrastructure context. Whether and how it extends to other domains is one of the residual questions below.

The second limit is methodological: this work is reflexive and conceptual rather than systematically empirical. It is reflexive practitioner research, without the interview, survey and longitudinal programmes a systematic study would run (Chapter 1, §1.5). The evidence drawn on is the comparative evidence from adjacent professions in Chapter 6, the observational and reflexive material of the analytical chapters, and the structural argument I develop. The framework’s empirical validation in the European infrastructure context is the first of the residual questions below. The contribution is conceptual and normative rather than empirical, as Chapter 1 said from the outset.

The third limit is the deliberate lightness of the philosophical apparatus. Chapter 3 developed the framework at a working level adequate to the analytical chapters, but its philosophical depth 💡 has been deferred. The lightness is deliberate but it is also a limit 💡 : the apparatus deployed here is sufficient for the analysis undertaken, but it is not the philosophically complete framework this work points toward. A subsequent work would extend it into the depth the analytical chapters indicate would be required.

9.8 Residual questions

Six residual questions emerge from this work that further research should take up.

The first is empirical validation of the framework’s predictions in the European infrastructure context specifically. This work’s central claims are reflective, conceptual and normative, but they include empirical predictions: that the proletarianisation mechanism operates in European infrastructure advisory in ways analogous to its operation in software and law, that the formation pipeline is being altered, and that the firm’s distinctive way of seeing is at structural risk under cumulative generic mediation. These predictions could be tested through systematic empirical work: interview programmes with practitioners across the ChMC stages, longitudinal studies of AI adoption in consulting firms, comparative studies across European infrastructure subdomains. The empirical validation would either confirm the framework’s predictions, refine them, or identify where they require revision. It is the most consequential of the six.

The second is comparative analysis across non-infrastructure consulting and non-European contexts. This work addresses European infrastructure advisory specifically, but the framework’s analytical apparatus could be extended to other consulting domains and to non-European contexts. Comparative work could identify which content of the framework is general (the proletarianisation mechanism, the sovereignty argument, the formation question) and which is specific to the domain (the third-box constituency in its European public-interest form, the European tradition). The comparative work would refine the framework’s claims about generality and specificity in productive ways.

The third is the philosophical depth that a Chapter 3 bis would develop, along the lines the third limit (§9.7) sets out. The development would be a contribution to philosophy of technology in its own right, and it would deepen the framework this work has developed at a working level.

The fourth is the operational specifics of the recommendations developed in this chapter. The recommendations to the four constituencies answer to the cross-scenario findings, but they are not operationally specific. The procurement architectures that recognise sovereignty, the chartering body standards that recognise formation under AI mediation, the disclosure norms that engage the third box, all of these are recommendations whose operational specifics would require further work, ideally in collaboration with the institutional actors whose decisions would shape them.

The fifth is longitudinal validation of the long-horizon predictions about Chartered-stage erosion. The framework’s most consequential prediction 💡 cannot be validated within the timeframe of any single research programme. It requires longitudinal research across decades, tracking how cohorts of consultants formed under different conditions develop the capacities at issue here, and how they come to understand themselves as practitioners, the question Ke et al. leave open for medicine (Chapter 6) 15 . This is the residual question with the longest time horizon, and it is the most institutionally consequential because the framework’s argument about what advisers owe publics and future generations depends on what the long-horizon prediction is correct about. This is not a failure of the framework. The imperative of responsibility operates unconditionally and it does not wait on empirical confirmation. The claim about 2046 is not an empirical prediction subject to falsification but a philosophical commitment to taking the formation pathway seriously as a condition for futures we cannot yet verify; it should be defended as such, and not dressed up as a forecast. 💡

The sixth is whether outcome-based contracting is compatible with answerability to the third box. The profession’s own foresight, in the HEC Alumni white paper (Chapter 2), expects proof-based, results-priced advisory to become the template for other services within ten years, so that “on paiera un avocat pour le taux de succès, un enseignant pour la progression, un médecin pour les résultats thérapeutiques”, with the firms that industrialise the approach becoming “les architectes d’un nouveau capitalisme cognitif” 16 . The forecast is offered as an opportunity. For public-interest work it raises a question this work has not answered: an outcome that can be contracted and proven is an outcome the two contracting parties have agreed to measure, and the third box is by definition party to no contract. A fifty-year asset is not a deliverable that clears in a results clause. Whether advisory legitimacy can be re-grounded in proof without being re-grounded in the client’s metrics alone, and what Chapter 6’s adjacent professions can teach about that (the teacher paid for progression, the doctor for therapeutic results), is a residual question that sits between this work and the next one.

These six residual questions identify the territory that further work should engage. I trust none of them undermines the contribution this work claims; and they sketch the research programme it belongs to.

9.9 A closing word

I write as a consultant. The itch that drove this work is one I am experiencing. The framework we have developed is offered as a contribution to making contemporary advisory transformation visible in terms that the productivity framings don’t and cannot capture. The recommendations we have drawn are addressed to the four constituencies whose decisions will shape the trajectory European infrastructure advisory follows in the next decade.

I do not know whether the trajectory will follow Commodity Advisory, Digital Twin Economy, Artisanal Advisory, or Disintermediated Clients 💡 .

The concerns named here are not idiosyncratic to one theoretical tradition. Magnifica Humanitas, Leo XIV’s 2026 encyclical on AI, arrives from within the Catholic Social Doctrine tradition at a related formulation: ‘when efficiency becomes the ultimate measure of value, human beings are tempted to see themselves as a project to be optimized rather than as persons called to relationship and communion’, naming from the natural-law tradition what the framework names from the Continental philosophy of technics 17 . That the same structural concern finds independent expression across independent intellectual traditions (philosophical, sociological, formal-economic, and doctrinal) is some reason to think it is not merely a theoretical artefact of the apparatus deployed here, though convergence is not itself proof: traditions can share a blind spot as readily as an insight.

The image from WALL-E 18 💡 names what is at the far end of the mechanism this work diagnoses: a profession, and beyond it a society, that has ceded its capacity for situated judgment and can no longer exercise it independently. The Stieglerian and Debordian analyses make this a structural possibility rather than a dystopian fantasy.

I hope this work can be one piece, one contribution to the wider conversation that is already underway and will continue beyond these lines. What we have assembled is offered to that conversation as one resource among others. The recommendations are addressed to constituencies whose decisions matter, because the trajectory is being shaped now, by the decisions practitioners and institutions are making and by how seriously those decisions engage what is at stake.

That engagement is the European public-interest tradition itself, and it is constitutive of legitimate advisory practice. The tradition has been reproduced through such engagement across decades. Whether it continues is not a question this work can close.

The work that follows will answer it… hopefully! 💡