Orientation before velocity

Abstract

The dominant question professional-services firms are asking about artificial intelligence is how fast: how quickly to roll it out, how to keep pace, how not to be left behind. This paper argues the question is premature. Before a firm can sensibly choose a speed, it must choose a direction — a bearing — and most have not. Drawing on the 2024–2026 management, information-systems, organisational-psychology and professional-regulatory literature, the paper makes four claims. First, that not choosing a bearing is itself a choice, and the one most firms are making by default: the evidence shows adoption accelerating sharply (organisation-wide use in professional services nearly doubling to 40% between 2025 and 2026) while the instruments of orientation — strategy, and even basic return-on-investment measurement — remain largely absent, and unsanctioned "shadow" use runs far ahead of any sanctioned decision. Firms are speeding up while measuring less. Second, that a genuine spectrum of legitimate postures exists — from principled non-adoption to wholehearted embrace — and that deliberate restraint is now not merely theorised but, for some firms, mandated by professional regulators. Third, that speed into the wrong orientation is not neutral but actively costly: the failure and return-on-investment literature attributes the conspicuous underperformance of AI programmes to managerial framing — to direction, not velocity. Fourth, that the decisive variables sit upstream of implementation, in a firm's identity, values and sense of what it is for, and that the 2026 organisational-psychology evidence now locates worker resistance squarely in these affective and existential registers rather than in usability. The paper proposes orientation before velocity as an organising doctrine, grounds it in established strategic theory — Boyd's OODA loop, in which orientation is the pivotal phase, and Wardley's doctrine, in which understanding precedes action — and is candid about where the evidence remains thin. The argument is offered evenhandedly: it is a case neither for AI nor against it, but for facing a direction on purpose before setting out.

---

1. The velocity trap

Walk into almost any professional-services firm in 2026 — a law firm, an accountancy practice, an advisory shop — and ask its partners about artificial intelligence, and you will hear a conversation about speed. Are we moving fast enough? Are our competitors ahead? Which tool should we buy this quarter? The anxiety is real and the questions are sensible. But they share a hidden assumption: that the direction of travel is already settled, and the only live variable is the pace. That assumption is almost always false, and it is the source of a great deal of wasted motion.

The navigator's discipline is the opposite. Before deciding how fast to sail, she decides which way she is facing. A fast boat pointed at the wrong star arrives nowhere worth arriving, only sooner. This paper takes that discipline seriously as a doctrine for AI strategy and gives it a name: orientation before velocity.

The contention is that a firm's bearing on AI — the settled, deliberate answer to "which way are we facing, and why?" — is logically and practically prior to every question of implementation, tooling and governance that currently dominates the discourse, and that the failure to settle it is the most common and most expensive mistake firms are now making.

As section 7 shows, "orientation before velocity" turns out to be a restatement, in a new domain, of a principle long established in strategic theory: that orientation is the decisive act, and that understanding must precede action for action to succeed.

What is new is not the principle but its urgency, and the evidence — much of it published within the last few months — that professional firms are violating it at scale.

---

2. The default bearing: not deciding is deciding

The first and most important empirical observation is that firms which have not chosen a bearing are not, in fact, stationary. They are being carried — and, on the latest figures, carried faster every year.

The pace of drift is now documented. The Thomson Reuters Institute's 2026 AI in Professional Services Report (9 February 2026), surveying more than 1,500 respondents across 27 countries, found that "organization-wide use of AI in professional services almost doubled to 40% in 2026, compared to 22% in 2025." Adoption, in other words, is no longer the laggard; it is galloping. What has not kept pace is orientation. The same survey found that "only 18% of respondents said they knew their organization was tracking return-on-investment (ROI) of AI tools." A firm that is doubling its AI use while four in five of its people cannot say whether anyone is measuring the return is not navigating. It is accelerating without instruments — no compass, no log, no fix on a star.

This is the velocity trap in its purest form: motion mistaken for progress.

The figures from a year earlier supply the baseline and the same diagnosis. The Thomson Reuters 2025 Future of Professionals Report (fieldwork February–March 2025, n=2,275) found only 22% of organisations reporting "a visible, defined AI strategy," and roughly four in ten "adopting AI without a strategy" at all. The trend across the two reports is unambiguous and, for the present argument, concerning: use is climbing steeply while strategy and measurement lag far behind. The gap between them is the space in which a bearing should sit, and mostly does not.

Beneath the strategy gap lies a behaviour gap. Reporting MIT's Project NANDA State of AI in Business 2025 study, Fortune (19 August 2025) recorded that workers at more than 90% of companies were using personal AI chatbot accounts for daily tasks — frequently without IT approval — while only around 40% of companies held official large-language-model subscriptions. The space between those numbers is where most professional AI use now actually happens: outside any sanctioned decision, in private accounts, on terms no partner has read.

This is shadow AI, and its significance for the present argument is precise. A firm that has declared no posture has not thereby abstained. It has, in effect, adopted — silently, ungoverned, and on the worst possible terms — by allowing its least-considered behaviour to become its de facto policy.

The gap is rhetorical as well as operational. Andriole and Barsky, writing in the California Management Review (30 June 2025) and reporting their own survey of 151 executives, found that while 76% called technology "essential" to competitiveness, only 20% rated AI initiatives a high priority and 77% had "barely explored or completely ignore[d]" generative AI — inaction they characterise bluntly as "a conscious choice to navigate digital disruption with analog mindsets." (Their self-conducted, multi-industry sample should be read as illustrative; but it coheres with the better-sampled Thomson Reuters data.)

The composite picture is a paradox only on the surface. Firms simultaneously profess that AI is essential, decline to form a considered view about it, use it constantly without admitting they do, and now adopt it organisation-wide without measuring whether it works. The reconciling insight is that doing nothing is not neutral. In a fast-moving technological environment, the absence of an orientation does not preserve a firm's position; it surrenders the choice of direction to whoever or whatever is already moving — a curious junior, a handy chatbot, a competitor's press release, a vendor's roadmap. The first task of orientation is simply to notice that one is already facing somewhere, and to ask whether it is anywhere one would have chosen.

---

3. The compass rose: a spectrum of legitimate postures

If a firm is to choose a bearing rather than inherit one, it needs to know what the available bearings are. A persistent move in the adoption discourse — particularly in vendor-adjacent material — is to collapse the choice into a binary: adopt, or fall behind. The literature does not support that collapse. It describes a spectrum, and theorises the restrained end of it as a genuine, defensible position rather than as failure or sloth.

At the expansive end, Gartner's framing of "AI ambition" (AI-readiness materials, 6 February 2025) sets out graded tiers of boldness: Defend (low-cost, quick-win applications that confer no durable advantage), Extend (custom applications built for differentiation), and Upend (new AI-enabled products and business models, high risk and high reward). Whatever one makes of Gartner's commercial vantage — this is consultancy positioning, not peer-reviewed research — the framework treats level of ambition as a variable to be chosen, not a maximum to be approached. A firm may rationally elect Defend.

At the restrained end, the academic literature theorises deliberate disengagement as a discipline in its own right. Bhatt and Sargeant (arXiv:2402.18326, February 2024) introduce algorithmic resignation — the deliberate and informed decision to disengage from AI in particular settings, embedded not as a failure of nerve but as a feature of governance: restricting outputs, attaching performance disclaimers, declining to deploy where deployment cannot be justified. Ali, Ahmed and colleagues (arXiv:2510.03868, 2026) document mission-driven organisations practising avoidance under tension — withdrawing from AI opportunities that conflict with core values even when those opportunities are transformative. (The latter study is small and qualitative, and covers mission-driven organisations rather than professional-services firms; it is used here by analogy. But the construct it names is exactly the one professional firms need a word for.)

Between these poles sits the middle ground: restriction (a default-no posture with narrow, supervised permitted uses) and cautious adoption (a default-yes posture within defined limits). The decisive point is that four legitimate bearings exist where the discourse usually offers one — and that the most countercultural of them, principled and governed restraint, is no longer merely an academic construct. As section 6 shows, professional regulators are now actively endorsing graded caution, and in places mandating it. Restraint has been promoted from heresy to sanctioned option.

An evenhanded treatment must add that this cuts both ways. The evidence theorising restraint remains stronger on the coherence of the posture than on its medium-term payoff, and the firm that abstains carries a real and still under-quantified opportunity cost. Restraint is a defensible bearing, but it is not a free one.

The point is not that any particular bearing is correct — that is each firm's to determine, and the right answer for a sole-practitioner conveyancer is unlikely to be the right answer for a global advisory firm — but that the choice is real, plural, and the firm's to make on purpose.

---

4. Why the wrong bearing is worse than a slow one

The deepest justification for putting orientation before velocity is that speed and direction are not symmetrical. A firm recovers from being slow; it recovers far less easily from having scaled, fast, in the wrong direction. The 2024–2026 failure and return-on-investment literature supports this asymmetry, and locates the cause of failure in matters of orientation rather than execution.

The most-cited datapoint is also the most contested. MIT's Project NANDA reported that some 95% of enterprises were seeing no measurable profit impact from their AI deployments — a figure widely criticised in the technology press as imprecise and resting on a modest sample, and which should be handled with corresponding care. But the headline number is not the interesting part; the diagnosis of why is. Wolfe, Choe and Kidd (arXiv:2509.02853, September 2025) attribute the underperformance not to weak technology but to "paradigmatic lock-in that channels AI into incremental optimization rather than structural transformation." Their two-by-two of strategic patterns — individual augmentation, process automation, workforce substitution, and collaborative intelligence — finds that the first three "reinforce legacy work models and yield localized gains," and reframes AI transformation as "an organizational design challenge rather than a technological one."

Put in this paper's terms: most firms accelerate along the bearing they were already on, and are surprised when AI merely makes their existing shape more efficient rather than changing what they are.

The MIT figure does not have to stand alone. Deloitte's AI ROI: The Paradox of Rising Investment and Elusive Returns (22 October 2025), surveying 1,854 executives across Europe and the Middle East, found that while "85 percent of organizations increased their investment in the past 12 months," most reported achieving satisfactory return on a typical AI use case only "within two to four years" — against a typical technology payback of seven to twelve months — with just 6% reporting payback in under a year. Crucially, Deloitte's account of why returns are elusive is dominated by organisational rather than technical causes: benefits that are real but hard to monetise, siloed data, and — tellingly — that "adoption depends on people: how cultural resistance is managed." Two independent bodies of evidence, then, converge on the same conclusion from opposite directions: the binding constraint on AI value is not the capability of the tool but the orientation of the organisation deploying it.

The mechanism is sharpened by Navigating the AI Adoption Trap (Organizational Dynamics, Elsevier, 2026), which argues that "the core challenge is often not technological capability but the AI solutionism trap," and that AI failures are "predictable outcomes of managerial dynamics, including premature problem closure, escalation of commitment, and diffusion of responsibility as systems scale." Its proposed corrective, a "governance-in-use" playbook, is notable for prescribing the deliberate slowing of adoption: "concrete practices that slow adoption, reopen problem framing before deployment, make evaluation transparent, and legitimize decisions to pause, modify, or withdraw AI systems." Here orientation-before-velocity finds direct support in a peer-reviewed source: deceleration and the re-opening of framing — a return to the question of direction — is the remedy for adoption gone wrong.

The expensive mistakes are mistakes of framing, made upstream and then locked in by speed. Velocity does not correct a bad bearing; it compounds it. A firm is better served by an hour spent deciding which way it faces than by a quarter spent moving faster in a direction nobody chose — and, on the ROI evidence, better served by measuring whether it is arriving anywhere at all.

---

5. The dimension the tools cannot see: identity and feeling

If orientation is prior to velocity, what actually determines a firm's orientation? The conventional technology-adoption models — the Technology Acceptance Model and its successors — would point to perceived usefulness and ease of use. The recent literature suggests these models are looking in the wrong place, or not the whole place. The decisive variables are not about whether the tool works; they are about what the firm is, and how its people feel about becoming something else. The 2026 evidence on this point is striking, and it is the most important addition to this edition.

The theoretical scaffolding is not new. Orlikowski and Gash's "technological frames" (1994) established that the same technology means different things to different groups within an organisation, and that adoption turns on those interpretive frames as much as on the artefact; Weick's work on sensemaking locates organisational action in how people retrospectively construct meaning from events. What is new is a wave of 2026 scholarship applying these insights to generative AI and finding the affective stakes far higher than usability models admit.

The construct doing the most work is professional identity threat. Shonhe and Min, in AI & Society (Springer, 2024), report a structural-equation study (n=413) finding that AI adoption generates a measurable threat to professional identity which shapes practitioners' intention to use AI — establishing identity, not merely usability, as a determinant. Their study found that a strong "AI identity" and the framing of AI as an explainable collaborator reduced that threat. (Two honest caveats: the sample comprised records- and information-management professionals, not lawyers, accountants or consultants; and that same study did not find organisational culture to be a significant driver — a useful corrective to any glib assumption that "culture explains everything.")

The 2026 work sharpens the point and raises the emotional register. Hermann, Puntoni and Morewedge, writing in Harvard Business Review ("Why Gen AI Feels So Threatening to Workers", March 2026), locate resistance in the frustration of three basic psychological needs — "competence (the feeling of being effective and capable); autonomy (the feeling of being in control of one's actions); and relatedness (the feeling of having meaningful interpersonal connections)." Their finding is that "when those needs are met, employees embrace gen AI as a helpful tool and copilot. But when they're not, employees feel threatened, at times even existentially, and balk at using gen AI."

Resistance, on this account, is not Luddism or inertia; it is a rational response to a perceived threat to selfhood, and no amount of tool-training touches it.

The existential framing is corroborated empirically. Shekhar and Saurombe (University of Johannesburg), in Frontiers in Psychology (17 February 2026), analysed 1,454 workplace-related comments using mixed computational and thematic methods and concluded that AI-induced workplace stress is "existential rather than technical": workers "question not just their competence with tools but their fundamental value as humans." They document professional identity erosion as creative work shifts "from executing the first idea" toward "curating, refining, and adding the crucial human touch (and catching AI's weird mistakes)," and a felt psychological-contract breach when tools marketed as assistive are experienced as replacement. (The study's source material is self-selected online discussion, which bounds its representativeness; but it triangulates closely with the Harvard Business Review account from an entirely different method.)

The significance for professional-services firms is hard to overstate, because professional identity is, for these firms, the product. A barrister's judgment, an auditor's scepticism, a consultant's craft — these are not incidental to the work; they are the work, and they are bound up with how practitioners understand themselves.

An AI posture is therefore never merely a procurement decision. It is, unavoidably, a statement about what kind of firm one wants to be: whether the technology sharpens the craft or hollows it out; whether one is willing to tell clients it is in use; how one feels about a junior's training years being compressed; whether the disagreement between the enthusiastic partner and the wary one is treated as noise to be managed or as a values question to be resolved.

This is the dimension that tool-first and governance-first approaches are structurally unable to see, because it is not a question about tools or controls. It is a question about orientation in the fullest sense — the firm's settled disposition toward what it is and what it is for. A firm that has not surfaced and reconciled this internal question has not chosen a bearing, however many tools it has bought; it has merely deferred the choice to whichever faction acts first.

---

6. The professional-services crossroads

The general argument lands with particular force in professional services, because these firms face a configuration of pressures that makes the orientation question both more acute and more avoidable — and because, uniquely, their regulators are now beginning to make orientation a matter of professional obligation.

The pressure to move is genuine, external, and increasingly incoherent. The Thomson Reuters Institute's 2026 report found that "more than half of both corporate legal departments and corporate tax departments want their outside firms to use AI on client matters," yet "less than one-third said they were aware whether their firms were doing so" — and, remarkably, that "40% of firm respondents said they have received orders both to use AI on matters and not to use AI on matters from various clients." Firms are being pushed simultaneously forward and back, by clients who themselves have not chosen a bearing.

The same report notes the arrival of agentic AI — already adopted by 15% of organisations, with a further 53% planning or considering it — which raises the stakes considerably, because an agent acts rather than merely assists, and a firm that has not oriented itself before delegating action is courting a different order of risk.

The pressure to pause is equally genuine and largely internal. The legal sector entered this period in robust financial health, with double-digit profit growth and the overwhelming majority of legal revenue still earned through hourly billing — what Thomson Reuters' market analysis calls "an almost absurd tension": a technology whose central promise is to compress the hours sits directly athwart a business model that sells them.

A firm cannot resolve that tension by buying a tool. It can only resolve it by deciding which way it faces: whether it is, at bottom, in the business of selling time or selling judgment — a question AI forces but does not answer.

What is genuinely new in 2026 is the regulatory overlay, which now does three things at once: it legitimises a cautious posture, it conditions speed on orientation, and in places it requires restraint. Four developments stand out.

• The EU AI Act entered into force on 1 August 2024 and applies on a phased timeline: prohibitions on unacceptable-risk practices and AI-literacy obligations from 2 February 2025, general-purpose-AI and governance obligations from 2 August 2025, full applicability from 2 August 2026, and high-risk obligations extending to 2 December 2027 and 2 August 2028. For any firm touching the EU market, the regulatory clock is now running, and a considered posture is not optional.
• The Financial Reporting Council, in March 2026, issued what it describes as the first generative- and agentic-AI guidance from any audit regulator globally. Its posture is instructively permissive-but-cautious: it "supports firms to realise the significant benefits these technologies offer" while requiring them to "mitigate risks to audit quality," and it insists that "it is people … who are accountable for audit quality — this guidance does not alter that position." A regulator, in other words, conditioning velocity on orientation toward audit-quality safeguards and undelegable human responsibility.
• The Bar Council of England and Wales, in updated guidance of 26 November 2025, frames understanding as the precondition of legitimate use. Its Chair, Barbara Mills KC, stated that "the ultimate responsibility for all legal work remains with the barrister" and that "[t]he best-placed barristers will be those that make the effort to understand these systems and, if appropriate, use them as tools in their practice, while maintaining control and integrity in their use." This is orientation before velocity rendered as professional duty: comprehend the thing, and your own relation to it, before you deploy it.
• The New Zealand Law Society, in guidance of 24 February 2026, holds that "a lawyer is not absolved from responsibility for legal advice or defects in an end-product (such as a contract) because it is derived from Gen AI," warns that "inputting client details and legally privileged material into a publicly accessible/external Gen AI tool may also give rise to a breach of privilege and confidentiality obligations," and stresses the verification of outputs against hallucination. For many New Zealand firms this guidance does not merely permit a more restrictive bearing on public, consumer-grade tools; it effectively requires one.

These instruments do not point in a single direction, and they are not meant to. What they share is a structural insistence that the firm orient itself — understand the technology, locate undelegable responsibility, weigh confidentiality and quality — before it accelerates. The professional firm is precisely the firm that cannot responsibly resolve its speed before it has resolved its direction; in 2026, that is no longer only prudent counsel but, increasingly, a regulatory expectation.

---

7. A doctrine of orientation

Pulling the strands together yields a doctrine — and, this edition can now show, not a novel one but a well-pedigreed principle applied to a new problem. It can be stated in five propositions.

1. A firm always has a bearing. If it has not chosen one, it has inherited one by drift — and the inherited bearing is, on the evidence, the worst-governed one available.
2. There is a spectrum, not a binary. Abstain, restrict, adopt cautiously, lean in — four legitimate bearings, each defensible for the right firm, with principled restraint now regulator-sanctioned and in places mandated.
3. Direction dominates speed. The expensive failures are failures of framing, made upstream and locked in by velocity; speed cannot rescue a bad bearing and tends to entrench it.
4. The decisive variables are upstream. Identity, values and feeling — what kind of firm we want to be — determine orientation, and sit prior to any question of tools or controls.
5. Therefore: orientation before velocity. Decide which way you are facing, and why, before you decide how fast to go.

The principle has a lineage. In the strategic theory of John Boyd, the celebrated "OODA loop" — Observe, Orient, Decide, Act — is routinely mistaken for a recipe for speed; Boyd's own emphasis was the opposite. He held orientation to be the pivotal phase, the "big O," because it shapes how one observes, what one decides, and how one acts; an actor who orients better prevails over a faster actor who orients worse. Simon Wardley's doctrine makes the sequencing explicit and structural: its principles are arranged in phases, with situational-awareness and understanding ("Know Your Users," "Focus on User Needs," "Challenge Assumptions") in an earlier phase than action and speed ("Bias Towards Action"; "Think Fast, Inexpensive, Restrained and Elegant"), on the stated logic that "later phases require earlier phases to work." Richard Rumelt's account of strategy (Good Strategy/Bad Strategy, 2011) puts diagnosis — understanding the nature of the challenge — ahead of guiding policy and coherent action, and treats the leap to action without diagnosis as the hallmark of bad strategy. Across three independent traditions the same structure recurs: understand and orient first; move fast second. "Orientation before velocity" is the application of that settled structure to the particular challenge of generative AI.

Two clarifications guard the doctrine against misreading. It is not a counsel of delay; an orientation can be chosen quickly, and "lean in, fast" is a perfectly coherent bearing for a firm that has genuinely reckoned with the question. The doctrine objects not to speed but to unoriented speed. Nor is it a covert argument for non-adoption; principled restraint and wholehearted embrace are both points on the same compass, and the doctrine is studiously neutral between them. Its only commitment is to the priority of the question. The firm that faces a direction on purpose may sail as fast as it likes.

---

8. The sceptic's ledger

Intellectual honesty requires setting out where this argument is exposed.

First, the evidence is uneven in rigour. Several data points come from vendor or consultancy publications — Thomson Reuters, Gartner and Deloitte among them — which have a commercial interest in encouraging adoption; they are cited here for their survey figures and as evidence of how the field frames the question, not as disinterested validation. Several academic sources are conceptual preprints, not yet peer-reviewed. The widely-quoted MIT "95%" failure figure is genuinely contested, and is relied on here only in company with the independently-sourced Deloitte ROI evidence that points the same way.

Second, the identity evidence, though now broader, has its own limits. The strongest single quantitative study (professional identity threat, n=413) is drawn from an adjacent profession; the 2026 corroboration is welcome but mixed in provenance — a Harvard Business Review practitioner article (authored by serious scholars, but not peer-reviewed empirical work) and a Frontiers in Psychology study built on self-selected online discussion. The convergence across these different methods is genuinely persuasive; no single one of them is dispositive.

Third, the restraint case is stronger on coherence than on payoff. The literature theorises principled non-adoption well, and regulators now sanction caution, but the medium-term cost of abstention is under-evidenced — and claims purporting to quantify the upside of having a strategy, and to cast non-adoption as an "existential" risk, failed verification in the underlying research and are deliberately not relied upon. The cost of facing the wrong way is documented; the cost of standing still longer than necessary is asserted more than shown.

Fourth, coverage is incomplete. Direct empirical data specific to management consulting firms remains thin; an instructive 2026 study of generative AI among Big-4 auditors could not be accessed for verification and is not cited; and the Australian court guidance on AI (the New South Wales and Victorian practice notes), which would round out the regulatory picture, could not be confirmed to this paper's evidentiary standard and is omitted rather than paraphrased loosely.

These are reasons for humility, not dismissal — and the surviving critical literature cuts in the doctrine's favour. Narayanan and Kapoor's AI Snake Oil (Princeton University Press, 24 September 2024) supplies the scholarly grounding for treating marketed AI capability with suspicion, helping readers discern "where [AI] might be useful or harmful, and when you should suspect that companies are using AI hype to sell AI snake oil," and naming the "AI hype vortex" that makes unoriented speed so tempting. A firm that has read the sceptics and still chooses to lean in has done the orientation work. That is precisely the point.

---

9. Conclusion: find your bearing

The firms now asking how fast to move on AI have, for the most part, skipped a question. They have not decided which way they are facing — and because doing nothing is not neutral, many are already drifting, oriented by their own least-considered behaviour rather than by any decision they would defend. The most recent evidence sharpens the picture rather than softening it: adoption nearly doubling year on year while barely a fifth of firms measure the return; resistance that turns out to be existential rather than technical; clients issuing contradictory instructions; regulators, newly arrived, insisting that firms understand before they deploy. The expensive mistakes are mistakes of orientation, made upstream and locked in by speed; a real spectrum of legitimate bearings exists, including a now-sanctioned posture of restraint; and the variables that should set a firm's bearing — its identity, its values, its sense of what it is for — sit prior to every question of tools and controls that currently fills the air.

The remedy is not to move faster, nor to refuse to move. It is to face a direction on purpose. Boyd understood that the contest is won in the orientation, not the acceleration; Wardley, that the later moves only work if the earlier understanding is done first. A navigator knows it instinctively: she does not begin by asking how fast the boat will go. She begins by finding her bearing — and only then, having fixed on a star worth steering by, does she think about speed. Professional-services firms would do well to navigate the same way. Decide which way you are facing, and why. The velocity can wait; the bearing cannot.

---

References

Ali, S., Ahmed, N., et al. (2026). Avoidance under tension: how mission-driven organisations withdraw from value-conflicting AI opportunities. arXiv:2510.03868. [Small-N qualitative; non-professional-services; cited by analogy.]

Andriole, S., & Barsky, N. (2025, 30 June). The Great AI-Era Technology Disconnect: Leadership Words, Posture, and Inaction. California Management Review (Berkeley). [Authors' own survey, n=151, multi-industry.]

Bar Council of England and Wales. (2025, 26 November). Updated guidance on generative AI for the Bar ("Considerations when using ChatGPT and generative artificial intelligence"); statement of Barbara Mills KC. [Primary; professional-body guidance.]

Bhatt, U., & Sargeant, H. (2024, February). Algorithmic resignation. arXiv:2402.18326.

Boyd, J. R. A Discourse on Winning and Losing (the OODA loop); as analysed in F. Osinga, Science, Strategy and War: The Strategic Theory of John Boyd (Routledge, 2006). [Foundational; orientation as the pivotal phase.]

Deloitte. (2025, 22 October). AI ROI: The Paradox of Rising Investment and Elusive Returns. [Survey, n=1,854 executives, Europe and the Middle East; 24 interviews. Consultancy provenance.]

European Commission. Regulatory framework on AI (the EU AI Act): in force 1 August 2024; prohibitions and AI-literacy obligations from 2 February 2025; GPAI and governance obligations from 2 August 2025; full applicability 2 August 2026; high-risk obligations to 2 December 2027 and 2 August 2028. [Primary.]

Financial Reporting Council. (2026, 30 March). Generative and Agentic AI Guidance — "the first from any audit regulator globally." [Primary; regulator guidance.]

Gartner. (2025, 6 February). AI readiness and the AI Opportunity Radar: defining your AI ambition (Defend / Extend / Upend). [Consultancy framework.]

Hermann, E., Puntoni, S., & Morewedge, C. K. (2026, March). Why Gen AI Feels So Threatening to Workers. Harvard Business Review. [Practitioner article by academic authors; not peer-reviewed empirical work.]

MIT Project NANDA. (2025). State of AI in Business 2025, reported in Fortune (2025, 19 August). [Industry report; the related "95% of pilots fail" headline is contested; the shadow-AI figures survived that criticism.]

Narayanan, A., & Kapoor, S. (2024, 24 September). AI Snake Oil. Princeton University Press.

Navigating the AI Adoption Trap (2026). Organizational Dynamics (Elsevier), S0090261626000136. [Peer-reviewed; "AI solutionism trap"; "governance-in-use."]

New Zealand Law Society — Te Kāhui Ture o Aotearoa. (2026, 24 February). Generative AI guidance for lawyers. [Primary; professional-body guidance.]

Orlikowski, W. J., & Gash, D. C. (1994). Technological frames: making sense of information technology in organizations. ACM Transactions on Information Systems, 12(2). [Foundational theory.]

Rumelt, R. (2011). Good Strategy/Bad Strategy: The Difference and Why It Matters. Crown Business. [Diagnosis before action.]

Shekhar, A., & Saurombe, M. D. (2026, 17 February). [Existential dimensions of AI-induced workplace stress]. Frontiers in Psychology, 17. [Mixed-methods analysis of 1,454 online comments; self-selected sample.]

Shonhe, L., & Min, J. (2024). Mitigating AI-induced professional identity threat and fostering adoption in the workplace. AI & Society (Springer). [SEM, n=413; records/information-management professionals.]

Thomson Reuters Institute. (2025, April). Generative AI in Professional Services Report; 2025 Future of Professionals Report (n=2,275; fieldwork February–March 2025); State of the US Legal Market.

Thomson Reuters Institute. (2026, 9 February). 2026 AI in Professional Services Report (n>1,500 across 27 countries). [Vendor publication; survey-derived figures.]

Wardley, S. Doctrine (Wardley Mapping): phased universal principles, situational awareness preceding action; "later phases require earlier phases to work." [Practitioner doctrine, Creative Commons.]

Weick, K. E. (1995). Sensemaking in Organizations. Sage. [Foundational theory.]

Wolfe, C., Choe, J., & Kidd, C. (2025, September). Strategic patterns of AI transformation: paradigmatic lock-in and collaborative intelligence. arXiv:2509.02853.