References for AI-Native Transformations in the Workplace

Beauchene, V., Duranton, S., Kalra, N. and Martin, D. (2025). AI at Work 2025: Momentum Builds, but Gaps Remain. Boston Consulting Group, June 2025. A BCG survey of more than 10,600 leaders, managers and frontline staff in 11 countries and regions. Frontline use of AI has stalled at about half of workers, and strong leader support lifts positive feelings about AI. Where companies are redesigning work comprehensively around AI, 46% of employees worry about job security, against 34% in less advanced companies.

Boston Consulting Group (2026a). What's Driving AI Value. BCG news page, 26 January 2026. BCG states its 10-20-70 rule for AI at scale: about 10% of the value comes from algorithms, 20% from data and technology, and 70% from people and processes. It is a rule of thumb from BCG's project experience, not a measured result.

Boston Consulting Group (2026b). AI at Work 2026: Strategy Matters More Than Tools (fourth edition). Slideshow, June 2026. A BCG survey of 11,749 people in about 14 markets, including Japan, India, Australia and the Middle East. In 'Reshape' and 'Invent' efforts, which change how work is done, 53% of employees save at least a day a week. In 'Deploy' efforts, which only give people access, the figure is 31%.

Deloitte (2026). State of AI in the Enterprise: The Untapped Edge. Deloitte AI Institute, January 2026. A survey of 3,235 director to C-suite leaders in 24 countries, run in August and September 2025. Only 30% are redesigning key processes around AI, and 84% have not redesigned jobs or the nature of work around AI. Just 21% have a mature governance model for AI agents, while 74% expect to use agentic AI at least moderately within two years.

EY (2026). EY survey: GenAI boosts productivity for 81% of Malaysian employees, but 68% say workloads have grown. EY Malaysia press release on the EY 2025 Work Reimagined Survey, Kuala Lumpur, 26 January 2026. A survey of 300 employees and 30 employers in Malaysia, across seven industries. 81% of employees who use generative AI report significant time savings, yet 68% say their workload has increased. Only 12% say they get enough AI training.

Gartner (2025). Gartner Says CHROs Must Become an Enterprise Leader of How Work Should Change in the AI Era. Press release, 27 October 2025. Tells HR leaders to sort AI work change into three types: augment, re-engineer and invent, and to find the top two to three redesigns for every function. A July 2025 survey of 1,973 managers found units that redesign work are twice as likely to beat revenue goals. A Gartner analyst says only one in five AI initiatives reaches measurable return.

IBM (2025). Cost of a Data Breach Report 2025: The AI Oversight Gap. Research by the Ponemon Institute; summary by L. Kessem, IBM Think, 30 July 2025. Based on 600 organisations that had a breach, surveyed by the Ponemon Institute. The global average breach cost fell to US$4.44 million. Organisations with high levels of shadow AI paid about US$670,000 more per breach, and 63% had no policies to manage AI use. The study is funded by a security vendor.

Tinkoff, D., Van der Veken, L. and Chui, M., with Balakrishnan, T. (2026). The state of AI in 2026: On the road to ROI. McKinsey & Company, 25 August 2026. A global survey of 1,719 respondents in 97 countries. Only 37% report any impact on EBIT (operating profit) from AI, about the same as in 2025, and the share of high performers stayed at 6%. Nearly three quarters of high performers have fundamentally redesigned workflows, up from 55% a year earlier, against about a quarter of other respondents.

Microsoft and LinkedIn (2024). AI at Work Is Here. Now Comes the Hard Part. 2024 Work Trend Index Annual Report, 8 May 2024. A survey of 31,000 knowledge workers in 31 markets, run by a vendor. It found that 78% of AI users bring their own AI to work ('BYOAI'), rising to 80% in small and medium-sized companies.

Challapally, A., Pease, C., Raskar, R. and Chari, P. (2025). The GenAI Divide: State of AI in Business 2025. MIT NANDA (Project NANDA), July 2025. Based on 52 interviews, a survey of 153 senior leaders at four industry conferences and a review of over 300 public AI initiatives. It claims 95% of organisations get 'zero return' from generative AI, but does not show how this was calculated, and calls itself preliminary. The figure is contested. It also found that workers at over 90% of surveyed companies use personal AI, while only 40% of companies say they bought an official subscription.

Niederhoffer, K., Rosen Kellerman, G., Lee, A., Liebscher, A., Rapuano, K. and Hancock, J. T. (2025). AI-Generated 'Workslop' Is Destroying Productivity. Harvard Business Review, 22 September 2025. Research by BetterUp Labs and the Stanford Social Media Lab defines 'workslop' as AI-made work that looks good but does not move the task forward. 41% of US workers surveyed said they had received it, and each case took nearly two hours of rework. The authors argue for clear norms rather than blanket AI mandates.

Ransbotham, S., Khodabandeh, S., Kiron, D., Candelon, F., Chu, M. and LaFountain, B. (2020). Expanding AI's Impact With Organizational Learning. MIT Sloan Management Review and Boston Consulting Group, October 2020. A survey of more than 3,000 managers in 112 countries, with 17 interviews. Only about 10% of companies got significant financial benefits from AI. What set them apart was learning between people and AI in both directions: AI learns from people, people learn from AI, and both are designed in.

Stanford Institute for Human-Centered Artificial Intelligence (2026). The AI Index 2026 Annual Report, chapter on the economy. Stanford University. The yearly AI Index pulls together data on AI from many sources. Its economy chapter reports that organisational use of AI rose from 55% in 2023 to 88% in 2025.

AI Singapore (2026). AI Readiness Index (AIRI). Web page, first published 5 September 2022, last updated 31 July 2026. The current version, AIRI v3.2 (The AI Readiness Framework), is published by the AIRI Foundation under a CC BY 4.0 licence. AIRI scores an organisation on 15 dimensions grouped in five pillars (leadership and culture, ethics and governance, business value, data foundation, infrastructure and standards). It places the organisation on five levels, from AI Unaware to AI Catalyst, with AI Ready as the target for most. Singapore programmes are mapped to each level, so the result points to concrete next steps.

Gartner (2026a). Gartner AI Maturity Assessment (AI Maturity Model and AI Roadmap Toolkit). Gartner web page, accessed October 2026. Gartner's assessment places an organisation on five stages: Foundational, Emerging, Operational, Scaled and Transformational. It shows current and target maturity for each capability in a heat map, and highlights the biggest gaps. The public page gives no criteria for moving from one stage to the next.

Liu, X. and Siviero, A. (2026). Ambition Is Everywhere, Maturity Is Rare: Inside IDC's 2026 AI MaturityScape Benchmark. IDC blog, 11 August 2026. IDC benchmarked 1,900 organisations in 20 markets on strategy, governance, people and technology, across five stages from 'AI scramble' (ad hoc) to 'AI-fuelled organisation' (optimised). Only 3.1% reach the top stage and 61.3% sit in the bottom two. Of firms that called themselves 'thrivers', only about one in six scored in the top two stages.

KAIVANT (2026). KAIVANT: The AI Nativeness framework and assessment platform. KaivantO (organisation) and Kaivant-I (individual). Kuala Lumpur: KAIVANT. kaivant.org. KAIVANT asks whether AI is building or eroding the capability of an organisation and its people: augmentation or substitution. Kaivant-O measures two axes at once, Leverage Architecture (what the way of working already produces) and Organisational Capital (whether human capacity is being built or used up), linked by Adaptation Architecture, across nine dimensions. Scoring is non-compensatory: the weaker axis sets the limit, so a strong side cannot hide a weak one. Kaivant-I, free and private, asks the same question for an individual.

Sukharevsky, A., Krivkovich, A., Gast, A., Storozhev, A., Maor, D., Mahadevan, D., Hämäläinen, L. and Durth, S. (2025). The agentic organization: Contours of the next paradigm for the AI era. McKinsey & Company, 26 September 2025. McKinsey sketches the 'agentic organisation', where small human teams steer many AI agents, and humans sit mostly 'above the loop' to set direction and oversee outcomes. As first moves it suggests putting agentic AI on the top team's agenda, a CEO vision, a centre of excellence, upskilling, and rewiring one or two 'lighthouse' domains.

Microsoft (2025). 2025: The Year the Frontier Firm Is Born. Work Trend Index Annual Report, 23 April 2025. Based on a survey of 31,000 workers in 31 markets, Microsoft describes the 'Frontier Firm' in three phases: AI as an assistant, agents as digital colleagues, and agents running whole processes while people set direction. It introduces the 'agent boss', a person who builds and manages agents, and the human-agent ratio as a new design measure.

Weill, P., Woerner, S. L. and Sebastian, I. M. (2024). Building Enterprise AI Maturity. MIT CISR Research Briefing XXIV-12, 19 December 2024. A four-stage AI maturity model built on a 2022 survey of 721 companies: Experiment and Prepare (28%), Build Pilots and Capabilities (34%), Develop AI Ways of Working (31%) and Become AI Future Ready (7%). Firms in the first two stages perform below their industry average, while firms in stages 3 and 4 perform well above it, so the big financial jump comes between stage 2 and stage 3.

Agrawal, A., Gans, J. and Goldfarb, A. (2022). Power and Prediction: The Disruptive Economics of Artificial Intelligence. Boston: Harvard Business Review Press. The authors see AI as cheap prediction. Every decision has two parts: a prediction, which AI can take over, and judgement about what matters, which stays with people. Point solutions add AI to one step of the old way of working; the bigger gains come from system solutions, where the whole system is redesigned, as factories were once rebuilt around electricity.

Jesuthasan, R. and Boudreau, J. W. (2022). Work Without Jobs: How to Reboot Your Organization's Work Operating System. Cambridge, MA: MIT Press. Argues that the fixed job is no longer the right unit for organising work. It breaks jobs into tasks and recombines them to fit people's skills and new options such as automation and gig work. Examples include DHL and Genentech.

Lamarre, E., Smaje, K. and Zemmel, R. (2023). Rewired: The McKinsey Guide to Outcompeting in the Age of Digital and AI. Hoboken, NJ: Wiley, 20 June 2023. Written by leaders of McKinsey Digital, this 400-page book is a detailed playbook for building the capabilities a company needs for digital and AI transformation. It covers six areas: the transformation roadmap, talent, operating model, technology, data, and adoption and scaling, with diagnostics, checklists and cases.

Lee, K.-F. (2026). AI Native: The Mandate to Transform Your Company. New York: Crown Currency (US); London: Ebury Edge (UK). Lee argues that AI makes intelligence cheap, so companies must rethink how they run, compete and create value, and the CEO must lead this personally. He challenges the idea that AI is just another technology cycle or a route to cutting headcount, and gives advice on organising teams around AI agents.

Nonaka, I. and Takeuchi, H. (1995). The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation. New York: Oxford University Press. A classic study of how Japanese firms such as Matsushita create new knowledge. It explains how knowledge moves between tacit knowledge (what people know but do not say) and explicit knowledge (what is written down), through four steps known as the SECI model: socialisation, externalisation, combination and internalisation.

Senge, P. M. (1990). The Fifth Discipline: The Art and Practice of the Learning Organization. New York: Doubleday. The book that made the learning organisation famous. It argues that an organisation's lasting advantage is its ability to learn, through five disciplines: personal mastery, mental models, shared vision, team learning and systems thinking. It does not discuss AI.

Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö. and Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. A field experiment with nearly 1,000 high school maths students in Turkey. Open access to GPT-4 raised practice grades by 48%, but when access was removed, exam grades fell 17% below students who never had it. A version built to coach rather than give answers (GPT Tutor) raised practice grades by 127% and largely removed the harm.

Brynjolfsson, E., Li, D. and Raymond, L. R. (2025). Generative AI at Work. The Quarterly Journal of Economics, 140(2), 889-942. First released as NBER Working Paper 31161 (2023). A study of 5,179 customer support agents who were given a generative AI assistant. On average they resolved 14% more issues per hour. Novice and less skilled agents gained 34%, while the most experienced gained little.

Budzyń, K. et al. (2025). Endoscopist deskilling after exposure to artificial intelligence in colonoscopy: a multicentre observational study. The Lancet Gastroenterology & Hepatology, 10(10), 896-903. doi:10.1016/S2468-1253(25)00133-5. In four endoscopy centres in Poland, doctors started using AI to help spot growths during colonoscopy. When they later worked without AI, their adenoma detection rate fell from 28.4% in the three months before AI to 22.4% in the three months after. It is an observational study, not a controlled trial, of 1,443 colonoscopies done without AI.

Buechsenschuss, R., Koch-Bayram, I., Biemann, T. and Puranam, P. (2026). GenAI adoption increases the density of knowledge and collaboration networks: Evidence from a field experiment. INSEAD Working Paper 2026/22/STR, April 2026 (first version January 2026). A field experiment with 316 employees in 42 teams at a European technology services firm. Teams that got an AI assistant built on the firm's own knowledge became more central in collaboration and knowledge-sharing networks after three months. Specialists became sought-after knowledge sources, while generalists gained most in productivity.

Dell'Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K. C., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K. R. (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013. A field experiment with 758 BCG consultants. On tasks inside AI's capability, those with AI completed 12.2% more tasks, 25.1% faster and at more than 40% higher quality. On a task outside it, they were 19 percentage points less likely to get it right, which the authors call the jagged frontier.

Dell'Acqua, F., Ayoubi, C., Lifshitz, H., Sadun, R., Mollick, E., Mollick, L., Han, Y., Goldman, J., Nair, H., Taub, S. and Lakhani, K. (2025). The Cybernetic Teammate: A Field Experiment on Generative AI Reshaping Teamwork and Expertise. NBER Working Paper 33641. A field experiment with 776 professionals at Procter & Gamble working on real product innovation tasks. One person working with AI performed as well as a team of two without AI. AI also balanced ideas, so technical and commercial staff proposed less one-sided solutions.

Demirer, M., Horton, J. J., Immorlica, N., Lucier, B. and Shahidi, P. (2026). Chaining Tasks, Redefining Work: A Theory of AI Automation. Working paper, arXiv 2606.15960 (first version February 2026, revised October 2026). The paper models work as a chain of connected steps. AI can run a run of steps on its own, and a person checks only the end result, so whether one step goes to AI depends on its neighbours. Data on real workflows show that AI-run steps cluster together, so the gains depend on how AI-suitable steps are grouped, not just how many there are.

Dillon, E. W., Jaffe, S., Immorlica, N. and Stanton, C. T. (2025). Shifting Work Patterns with Generative AI. arXiv preprint 2504.11436, version 4, 13 November 2025. A six-month field experiment with 7,137 knowledge workers in 66 firms. Workers who used the AI assistant spent about two fewer hours a week on email. The study found no change in the amount or mix of tasks people did.

Henseke, G. (2026). From Exposure to Adoption: Generative AI in European Workplaces. arXiv preprint 2604.18849, version 3, 28 April 2026. Uses worker data from 35 European countries, where adoption of generative AI ranges from under 3% to 25%. Workers in more exposed jobs adopt more. Workers with more influence in their organisation adopt more still, especially in highly exposed jobs.

Humlum, A. and Vestergaard, E. (2025). Large Language Models, Small Labor Market Effects. Becker Friedman Institute Working Paper 2025-56, University of Chicago, May 2025. Links a 2024 survey of Danish workers in 11 exposed occupations to national registry data. Employer encouragement nearly doubles take-up of AI chatbots, from 47% to 83% for a typical worker. Two years after ChatGPT, adopters show no measurable change in earnings or recorded hours.

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I. and Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv preprint 2506.08872, MIT Media Lab, 10 June 2025 (revised 31 December 2025). 54 people wrote essays with an AI chatbot, a search engine or no help, while their brain activity was measured. People who wrote with AI from the start showed the weakest brain engagement. In a fourth session with only 18 participants, those who first wrote without AI and then got AI had better memory recall and more brain activity than those who switched the other way.

Lund, C., Euyang, T., Munyikwa, Z. and Fadaee, M. (2026). AI Exposure Scores: what they measure, what they miss, and what comes next. arXiv preprint 2606.23633, Cohere Labs, 22 June 2026. A review of the 2023 'GPTs are GPTs' exposure scores that are widely used in policy. It finds the scores are static snapshots, tied to one time and place, and that a list of tasks misses the connections between tasks and the judgement and relational work in a job. It calls for worker-centred measures and for treating workers as partners in the research.

Makridis, C. (2025). Organisational Transmission of AI: Managerial Influence on Generative AI Adoption. Economics and Policy Seminar, Cambridge Judge Business School, 12 November 2025 (seminar abstract). Using the US Gallup Panel (about 10,000 people a year, 2023 to 2025), Makridis finds that trust in leaders and clear communication from managers were the strongest predictors of whether employees took up generative AI, even after allowing for income, occupation and sector. Staff who adopted AI under trusted, clear managers were also more engaged.

Becker, J., Rush, N., Barnes, E. and Rein, D. (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. arXiv preprint 2507.09089, METR, 12 July 2025. A randomised controlled trial with 16 experienced open-source developers working on 246 real tasks in their own projects. With AI allowed, they took 19% longer, yet afterwards they believed AI had made them 20% faster. Before the study they had expected a 24% speed-up.

Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000). A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics, Part A: Systems and Humans, 30(3), 286-297. A classic human factors paper. It splits any task into four functions: gathering information, analysing it, choosing a decision and carrying it out. Each function can be automated to a different level, and the authors note that automation does not simply replace human work but changes it.

Raisch, S. and Krakowski, S. (2021). Artificial intelligence and management: The automation-augmentation paradox. Academy of Management Review, 46(1), 192-210. https://doi.org/10.5465/amr.2018.0072 A theory paper. It argues that automation and augmentation cannot be cleanly separated: they are linked across time and across parts of the organisation. Pushing too hard on either one sets off reinforcing cycles that harm organisations and society, so leaders should manage both together.

Reich, A., Wolfe, D., Price, M., Choe, A., Kidd, F. and Wagner, H. (2026). Safety first: Psychological safety as the key to AI transformation. arXiv preprint 2602.23279, 26 February 2026. A study of 2,257 employees at a global consulting firm. Psychological safety predicted whether people started using AI, but not how often or how long they used it once they had started.

Schmid, S., Schieder, C. and Weeger, A. (2026). Becoming Hybrid: Human-AI Collaboration Reshaping Professional Role Identity. A Systematic Literature Review Using the ADO Framework. ECIS 2026 Proceedings (European Conference on Information Systems), paper ECIS2026-2427, 26 May 2026. A review of 61 studies on how working with AI changes people's sense of their professional role. It finds five patterns: threat, distancing, reinforcement, hybrid identities and mixed outcomes, depending on whether AI is seen to support or challenge expertise and autonomy. Identity changes step by step, through three kinds of identity work: protective, developmental and integrative.

Selenko, E., Bankins, S., Shoss, M., Warburton, J. and Restubog, S. L. D. (2022). Artificial intelligence and the future of work: A functional-identity perspective. Current Directions in Psychological Science, 31(3), 272-279. https://doi.org/10.1177/09637214221091823 A theory paper from work psychology. It proposes that AI can strengthen or threaten how people see themselves at work. Which way it goes depends on how AI is used (to complement tasks, replace tasks or create new ones), how it is introduced, and whether others still value the person's contribution.

Shao, Y., Zope, H., Jiang, Y., Pei, J., Nguyen, D., Brynjolfsson, E. and Yang, D. (2025). Future of work with AI agents: Auditing automation and augmentation potential across the U.S. workforce. arXiv preprint 2506.06576, first version 6 June 2025 (version 3, 1 February 2026). Stanford's WORKBank database combines what 1,500 workers want with what AI experts say AI agents can do, across 844 tasks in 104 US occupations. It uses a five-level Human Agency Scale, from H1 (AI does the task alone) to H5 (AI cannot work without continuous human involvement). An equal human-AI partnership (H3) is the top worker choice in 45.2% of occupations (47 of 104).

Tamura, S. (2025). Generative AI and knowledge creation: New management issues revealed by the Survey on Standardization Activities (2021). RIETI column, Research Institute of Economy, Trade and Industry, Tokyo, 15 January 2025. A Japanese government research institute column based on a survey of firms in standards work. About 62% found the tacit knowledge gained by taking part in standards meetings useful, against about 57% for written standards. The author argues that knowledge which depends on non-verbal signals is hard for generative AI to produce, and suggests marking each task in a job description as suitable or not suitable for generative AI.

Uchihira, N. (2026). Tacit knowledge management with generative AI: Proposal of the GenAI SECI model. arXiv preprint 2603.21866, 23 March 2026. Also in AHFE Open Access proceedings, https://doi.org/10.54941/ahfe1007727 A Japanese researcher redesigns Nonaka's SECI knowledge creation cycle around generative AI. Its central idea, 'digital fragmented knowledge', treats tacit and written knowledge as one connected whole in digital space. The paper also sets out a system design and compares it with similar work.

Wu, J., Ziems, C., Chan, R., Chen, N. F., et al., Kan, M.-Y. and Yang, D. (2026). When does AI augment work? A workflow-level framework for human-agent collaboration. arXiv preprint 2609.12482, 11 September 2026. A paper by 21 authors, mostly from Singapore (National University of Singapore, A*STAR and the Ministry of Manpower) and Stanford, written after a workshop on collaborative intelligence and the future of work. It defines AI augmentation by six conditions in two layers: durable net value, meaningful human control, and clear accountability and recovery; then deepening learning, career pathways and job purpose. It tests the framework on AI-run social surveys.

Beer, M. and Nohria, N. (2000). Cracking the code of change. Harvard Business Review, May-June 2000, 133-141. Beer and Nohria describe two theories of change. Theory E is top-down and aims at economic value through structure, incentives, layoffs and restructuring. Theory O is bottom-up and aims at building the organisation's capability and culture through experiment and learning. They argue that lasting change combines both.

Bushe, G. R. and Marshak, R. J. (eds) (2015). Dialogic Organization Development: The Theory and Practice of Transformational Change. Oakland: Berrett-Koehler. An edited book that sets out dialogic organisation development. It sees organisations as realities that people keep shaping through conversation. Change comes from changing who talks with whom, about what, and the stories and images people use, with the consultant hosting the dialogue rather than prescribing answers.

Cummings, S., Bridgman, T. and Brown, K. G. (2016). Unfreezing change as three steps: Rethinking Kurt Lewin's legacy for change management. Human Relations, 69(1), 33-60. The authors compare what Lewin actually wrote about change in three steps with how textbooks present it. They conclude that Lewin never built an unfreeze-change-refreeze model; it took shape after his death, traced back to his 1947 article in the first issue of Human Relations. They ask why this simplified model became the foundation of change management.

Engelbart, D. C. (1962). Augmenting Human Intellect: A Conceptual Framework. Summary Report AFOSR-3223, Stanford Research Institute, Menlo Park, October 1962. The founding text on augmentation. Engelbart treats the human, their language, artefacts (technology), methods and training as one system, which he calls H-LAM/T, to be studied and improved as a whole. A change in one part matters only in how it works in the whole process. He proposes to go for quick gains first and use each gain to reach the next.

Heifetz, R. A. (1994). Leadership Without Easy Answers. Cambridge, MA: Belknap Press of Harvard University Press. Separates technical problems, where a known fix exists and an expert can apply it, from adaptive challenges, where no ready answer exists. Adaptive challenges need the people involved to learn, change priorities and adopt new ways of working. Leadership means helping people do that adaptive work.

Hughes, M. (2011). Do 70 per cent of all organizational change initiatives really fail? Journal of Change Management, 11(4), 451-464. https://doi.org/10.1080/14697017.2011.630506 Traces five published sources behind the popular claim that 70% of change programmes fail. Concludes that the claim is a widely repeated story with no valid and reliable evidence behind it.

Kotter (2026). Leading Change in the Era of AI. Live training course, launched 22 April 2026. Kotter Inc. A 2.5-hour live course in two modules that applies Kotter's 8 Accelerators for Change to AI. It covers AI tensions, 'Survive' and 'Thrive' responses to AI disruption, adoption barriers and short-term wins. It says nothing about roles, identity or decision rights.

Mollick, E. (2025). Making AI Work: Leadership, Lab, and Crowd. One Useful Thing (Substack), 22 May 2025. A model for organisational AI adoption in three parts. Leadership paints a concrete picture of AIenabled work and makes open use safe and rewarding. The Lab turns the best finds into shared practices and tests models on real tasks, and the Crowd, the whole workforce, discovers uses in its own work. It is well argued but not tested.

Pasmore, W., Winby, S., Mohrman, S. A. and Vanasse, R. (2018). Reflections: Sociotechnical systems design and organization change. Journal of Change Management, published online 6 December 2018. https://doi.org/10.1080/14697017.2018.1553761 Four leading practitioners trace sociotechnical systems design from the British coal mines to the digital age. They restate its core principle that the person should be seen as a complement to the machine, not as an extension of it. They expect future design work to include virtual partners, such as IBM's Watson, as equal contributors.

Prosci (2026a). AI transformation: A people-centric guide to leading change. Prosci blog, updated 7 August 2026. Prosci's public English guide treats AI transformation as a people challenge. It recommends clear communication, hands-on training, early involvement and structured models such as ADKAR at every stage. It mentions role ambiguity and governance, but says nothing about professional identity or decision rights.

Prosci (2026b). Prosci launches new AI integration program to help organizations turn AI investment into business results. Prosci news release, 19 August 2026. Prosci's announcement of its AI integration programme. It argues that with AI there may be no fixed future state, because new capabilities keep changing what employees can do. It cites Prosci's 2026 State of AI Adoption research, which found that motivation alone does not separate organisations making progress with AI from those that struggle.

Weick, K. E. and Quinn, R. E. (1999). Organizational change and development. Annual Review of Psychology, 50, 361-386. A major review that separates episodic change from continuous change, which is ongoing, evolving and cumulative. For continuous change, the authors replace unfreeze, change, refreeze with freeze, rebalance, unfreeze. To 'freeze' means to make current patterns visible through maps, schemas and stories.

Ethington, E. (2026). Automation, AI, and the Risk of Doing Things Efficiently That Should Not Be Done at All. The Lean Post, Lean Enterprise Institute, 5 March 2026. A lean practitioner argues that before you automate a task or give it to AI, you should ask whether the task should exist at all. He gives factory examples where automation sped up work that was not needed, and advises designing the value stream first and monitoring AI output, because its waste is harder to see.

Hoeppe, A. (2026). Use Business Capability Maps to Prioritize AI Investments. Gartner webinar, recorded 18 February 2026. A Gartner analyst webinar for manufacturing CIOs. It ranks business capabilities by their potential for improvement and innovation, links the top ones to AI use cases, then ranks use cases by business value, feasibility and dependencies.

Gmyrek, P., Berg, J., Kamiń ski, K., Konopczyń ski, F., Ładna, A., Nafradi, B., Rosłaniec, K. and Troszyń ski, M. (2025). Generative AI and Jobs: A Refined Global Index of Occupational Exposure. ILO Working Paper 140. Geneva: International Labour Organization, 20 May 2025. The ILO scores each task in a job for exposure to generative AI. Three AI models scored the tasks first, but they disagreed. So the team surveyed 1,640 workers in Poland, and international experts then reviewed a sample of 608 tasks. It finds about one in four jobs worldwide is exposed to generative AI.

Google PAIR (2019). People + AI Guidebook, chapter 'User Needs + Defining Success'. Google People + AI Research, first published 8 May 2019, updated online since. Google's design guidance on when AI should take over a task and when it should support a person. Automate when a task is boring, repetitive, awkward or dangerous, or when people lack the skill. Augment when people enjoy the task, when personal responsibility matters, or when the stakes are high.

Jesuthasan, R. and Ramanathan, P. (n.d.). Reimagine the art of the possible with Work Design. Mercer insights article, accessed October 2026. Mercer's work design has three steps. Break jobs into tasks and projects. Decide the best way to do each task: automation, AI augmentation, gig workers or outsourcing. Then rebuild the tasks into new jobs and workflows.

Jobs and Skills Australia (2025). Our Gen AI Transition: Exposure (Analysis Paper A). Canberra: Australian Government, Generative AI Capacity Study, August to September 2025. Gives each task two separate scores from 0 to 1: how far generative AI can augment it and how far it can automate it. It adapts the ILO method to Australian occupations. It finds that augmentation generally outweighs automation, with automation potential mostly in routine roles. The scores show potential under full adoption, not actual use.

May, C. (2025). Künstliche Intelligenz trifft Lean: Generative KI zur Eliminierung digitaler Verschwendung im Lean Management [Artificial intelligence meets lean]. YOKOTEN, issue 06/2025. CETPM, Germany. In German. A German lean institute (CETPM) proposes new kinds of waste for the AI era. Cognitive waste is skilled staff spending time searching for information they cannot find. Strategic waste is daily work that does not serve the company's goals, and a third is data that is collected but never used.

Moon, J., Walsh, R., Di Leo, V. and Thelwall, A. (2026). Rewiring software delivery for the agentic era. McKinsey & Company, 28 May 2026. McKinsey describes software delivery in which AI agents do structured work along each stage, from requirements and architecture checks to code, tests and packaging. Work moves between stages in standard, machine-readable formats, and people step in only at defined review gates. Smaller human teams supervise the agents.

Ozenc, K. (2026). AI Design is Service Design. Design Meets AI (Substack), 5 August 2026. Ozenc argues that designing with AI agents is a service design task. Agents can sit 'frontstage', facing the customer, or 'backstage', out of view, much like staff in a service blueprint. However much work the agents do, accountability stays with a human and must be kept clear.

Pereira, S. and Davis, A. (2024). Flow Engineering: From Value Stream Mapping to Effective Action. Portland, OR: IT Revolution Press. Published 14 May 2024. A practical book that brings value stream mapping into knowledge work. Teams use five maps in short, fast workshops to talk about how work really flows, find hidden dependencies and agree a roadmap that the people doing the work own. It is not written about AI.

Pereira, S. (2026). Value-stream mapping is your missing AI superpower. The Lean Post, Lean Enterprise Institute, 4 February 2026. Argues that AI makes both the strengths and the problems of a value stream bigger, so teams should map the whole flow first and aim AI at the real bottleneck. Its warning example: a team celebrates 30% faster coding while missing that the testing (QA) queue has grown four times longer. It names six AI-specific problems, such as overloading bottlenecks and automating work that should be removed.

Poteri, J. (2025). AI and value stream mapping [Tekoäly ja arvovirtakartoitus]. Master's thesis, School of Science, Aalto University, Espoo, Finland. Issued 24 April 2025. The only formal extension of value stream mapping for AI that the research found. It adds two scores to each step, repetitiveness and significance, and uses them to sort steps into full automation, AI then human, or a human-AI hybrid. A small multi-agent AI system it built was good at drawing and analysing maps but unreliable with numbers in complex cases; the author suggests adding process mining.

Ulwick, A. W. (2016). Jobs to be Done: Theory to Practice. IDEA BITE PRESS. Free copy published by Strategyn. The founder of Strategyn sets out jobs-to-be-done and his Outcome-Driven Innovation method. It starts from the idea that people buy products and services to get a job done, and works back from that job and the outcomes customers use to judge it. It is about customers and products, not about AI or internal work.

van der Aalst, W. M. P. (2025). No AI without PI! Object-centric process mining as the enabler for generative, predictive, and prescriptive artificial intelligence. arXiv preprint 2508.00116, 31 July 2025. The founder of process mining argues that AI needs good process data. Process mining reads the event logs that systems leave behind to show how work actually gets done. He argues for object-centric event data, because forcing events that touch several business objects into one case distorts the picture.

Alibaba Cloud (2026). Ant Group builds enterprise-wide AI infrastructure with QwenWork. Alibaba Cloud Community blog, 13 October 2026. Ant Group chose Alibaba's QwenWork as one company-wide base for AI assistants and agents for its tens of thousands of staff. It runs privately inside Ant's own network, so data stays in-house, and it connects to Ant's AI security and governance systems, with auditable logs. No user numbers or results are given.

Chen, M. Y. (2023). Not if but when: Why Japan's Panasonic Connect is going all in on AI. Microsoft Source Asia, 11 May 2023. Panasonic Connect gave an AI assistant to all 12,500 staff in Japan, from the CEO down, with very little training: staff were told to ask full questions and given examples. They asked 55,380 questions in the first month. The company called the productivity gains so far anecdotal.

Coacci, J. (2025). Duolingo CEO admits his controversial AI memo 'did not give enough context' and insists the company never laid off full-time employees. Fortune, 18 August 2025. Duolingo's April 2025 'AI-first' memo said it would gradually stop using contractors for work AI can handle, and add headcount only if a team cannot automate more of its work. After a public backlash, the CEO said he had not given enough context and that no full-time staff had been laid off. Staff are encouraged to use Friday time to experiment with AI.

Fintech Futures (2025). Maybank adopts Azure in $236m Microsoft tech transformation deal. 7 August 2025. Maybank signed a five-year partnership with Microsoft valued at about RM1 billion (about US$236 million). It includes moving to Microsoft's cloud and rolling out Microsoft 365 Copilot to all 44,000 employees. It is an announcement; no results have been published.

Kell, J. (2026). Hitachi America's CIO says the Japanese conglomerate's enterprise AI strategy isn't one-size-fits-all. Fortune, 5 August 2026. Hitachi, with nearly 290,000 staff, has no single company-wide AI system. It combines everyday assistants, job-specific options approved by IT with business leaders, and coding assistants, and each division manager is responsible for AI use and spending. Hitachi and the vendor Appian jointly report a 40% efficiency gain in sales and marketing.

Leaño, M. (2026). DBS's approach to AI at scale. Frontier Enterprise, 16 March 2026. DBS spent nearly a decade building the data, systems and governance behind its AI work. It now runs over 2,000 AI models in more than 430 use cases, and has cut AI time-to-value from about 18 months to two to three months. DBS reports about SG$1 billion of value from AI and data in 2025, a company figure.

MONOist (2024). 「製造業らしい活用」が増加 パナソニック コネクトの生成 AI 全社導入の経 過報告 [More manufacturing-style use: progress report on Panasonic Connect's company-wide generative AI rollout]. ITmedia MONOist, 26 June 2024. In Japanese. In its first year, Panasonic Connect's AI assistant was used about 1.4 million times by about 12,400 staff in Japan. The company estimates 186,000 hours saved, based on an average of about 20 minutes saved per use. Use moved towards higher-value work such as strategy and product planning and manufacturing questions.

Nolan, B. (2025). Shopify CEO tells employees to prove AI can't do jobs before asking for new hires. Fortune, 8 April 2025. Reports on Tobi Lütke's internal memo making effective AI use a basic expectation for everyone at Shopify, including executives. Teams must show why AI cannot do the work before asking for more headcount. AI use is added to performance and peer reviews.

OpenAI (2024). Moderna: Accelerating the development of life-saving treatments. OpenAI customer story, April 2024. Moderna ran an AI prompt contest to find its top 100 users, who became internal generative AI champions. Within two months of adopting ChatGPT Enterprise, staff had built 750 custom assistants (GPTs), and 40% of weekly users had built one. All figures come from the vendor and the company.

OpenAI (2026). MUFG aims to become AI-native with OpenAI. OpenAI customer story, 7 July 2026. About 35,000 employees at MUFG Bank can use ChatGPT Enterprise in a phased rollout that began in 2026. Staff must complete e-learning before they get access, and AI champions in each department encourage colleagues to try it. After training from OpenAI, staff built more than 1,800 custom assistants (GPTs) in four months.

Panasonic Connect (2025). パナソニック コネクト、「聞く」から「頼む」へシフト した AI 活用で年間 44.8 万時間の削減を達成 [AI use shifts from 'asking' to 'requesting', saving 448,000 hours a year]. Press release, 7 July 2025. In Japanese. Panasonic Connect reports that in 2024 its staff saved 448,000 hours with AI, 2.4 times the year before, from 2.4 million uses at an average of 28 minutes each. Staff moved from asking AI simple questions to handing it longer tasks such as coding, writing procedures and reviewing documents. About 49% of the roughly 11,600 staff with access used it each month.

Sadler, D. (2025). CBA reverses AI-driven job cuts, admits 'error'. Information Age (Australian Computer Society), 21 August 2025. Commonwealth Bank of Australia announced in July 2025 that 45 customer service roles would go, replaced by an AI voice bot. In August, after weeks of pressure from the Finance Sector Union, it reversed the decision, admitted an error and apologised to the staff.

Mousicos, N. (2025). Klarna reverses AI customer service overhaul. Tech.co, 20 May 2025. Klarna cut about 700 jobs from 2022 and said in February 2024 that its AI assistant handled 75% of customer chats. In May 2025, after reports of customer dissatisfaction and poor support quality, it began recruiting human customer service staff again. Its chief executive said customers should always be able to reach a human.

Abdullahi, A. (2025). Duolingo CEO clarifies AI-first plan after backlash, says AI won't replace employees. TechRepublic, 26 May 2025. After public backlash to his 'AI-first' memo, Duolingo's chief executive Luis von Ahn wrote that he does not see AI as replacing what employees do, and that hiring would continue at the same pace. The company set up training, AI advisory councils and protected time for teams to experiment with AI.

The Edge Malaysia (2025). DBS CEO sees job cuts of 4,000 temporary staff as AI replaces roles. Based on Bloomberg News, 24 February 2025. DBS chief executive Piyush Gupta said about 4,000 contract and temporary roles would go over the next three years as AI takes over their work. DBS had about 41,000 staff, of whom 8,000 to 9,000 were contract or temporary. He said permanent staff would not be affected.

The Straits Times (2026). Interview with Lee Yan Hong, DBS group head of human resources, 24 August 2026 (updated 25 August 2026). Read via the AI Understanding news summary, 26 August 2026. DBS says more than 70% of its employees use AI and generate over 1.8 million prompts a month across 19 markets. More than 100 of the 500 people in its HR team have been trained as career advisers. The figures come from the bank and were not independently checked.

Wade, M., Trantopoulos, K., Navas, M. and Romare, A. (2025). How to Scale GenAI in the Workplace. MIT Sloan Management Review, 8 July 2025. Also published in MIT Sloan Management Review Middle East. Novo Nordisk grew its use of Microsoft Copilot from a few hundred users in January 2024 to about 20,000 in February 2025. It used champion networks, role-based playbooks and a campaign to counter so-called AI shaming, where some staff saw AI use as cheating. Staff saved 2.17 hours a week on average, based on staff surveys and usage data, and 15% of early users went inactive after three to four months.

Wei, W. (2025). From failure to success: The birth of GrabGPT, Grab's internal ChatGPT. Grab Engineering blog, 19 May 2025. In March 2023 a Grab engineer tried to build a support chatbot for the machinelearning platform team, and it failed. He then built an internal ChatGPT-style assistant over a weekend. It had over 3,000 users and 600 daily users within three months, and by May 2025 almost all Grab staff used it.

Zhang, Y. (2026). Chinese firms embrace AI but find adoption bottleneck is more organizational than technical. Yicai Global, 21 July 2026. A report from the World Artificial Intelligence Conference (WAIC) 2026 in Shanghai. At one Haier factory, order to shipment took 7.1 days while production took only 2.3 days; the rest of the time went through 12 approval stages. Taikang Insurance scaled its AI financial approval system only after it ran reliably for a month at over 99% accuracy, and warned that many projects fail because they start too big.

Cao, D. (2026). What Fintech Founders Get Wrong About 'AI-Native' Development. HackerNoon, 27 July 2026. Argues that 'AI-native' in pitch decks is going the way of 'cloud-native': stretched until it means nothing specific. Names the gap between what a pitch claims and what the architecture can show 'positioning debt'. Repeats a removal test credited to the investor CRV: strip out the AI, and if the core value survives, the product is only AI-enabled.

Edmondson, A., interviewed by J. Brazier (2026). HR leaders must prioritize experimentation over engagement, says Amy Edmondson. UNLEASH, 17 April 2026. The Harvard professor advises HR leaders to run structured experiments rather than let a thousand flowers bloom. Teams should keep asking what to try next and judge the options with care. She warns that AI makes the easy path more tempting and that 'work slop' signals weak systems thinking.

Glassman, C. (2026). Your Company Uses AI. That Doesn't Make It AI-Native. Here's the Difference. Entrepreneur, 25 August 2026. Proposes a removal test: imagine every AI model disappeared on Monday morning. A company that only adopted AI would run in roughly the same way, just slower and dearer. In an AI-native company, work design, staffing, review and measurement would stop making sense.

Grothaus, M. (2024). Your 'AI first' company is giving me AI fatigue. Fast Company, 13 April 2024. A tech writer argues that calling a company 'AI-first' has become virtually meaningless. Using ChatGPT for sales or marketing does not make a firm AI-first, any more than spreadsheets make it 'Excel-first'. He compares the rush to the NFT boom and its failed promises.

IEEE Communications Society Technology Blog (2026). Comparing AI Native mode in 6G (IMT 2030) vs AI Overlay/Add-On status in 5G (IMT 2020). IEEE ComSoc Techblog, 15 January 2026. Explains that in 5G networks AI was an add-on for optimisation on top of a human-designed system. In 6G, AI is meant to be built into network control and design from the start, following the ITU's IMT-2030 framework. 6G is still largely theoretical.

Mollick, E. (2023). Detecting the Secret Cyborgs. One Useful Thing (Substack), 18 June 2023. Mollick describes 'secret cyborgs': employees who use AI to boost their work but hide it. They hide it because of company bans, because AI-written work is judged differently once people know its source, and because they fear that showing their gains will lead to job cuts.

The Oxford Review (n.d.). Is the change curve a myth? The Oxford Review (evidence briefing blog). Accessed 25 October 2026. This blog post reviews the research on the change curve, which came from Kübler-Ross's work on grief. It concludes the curve is real but does not always appear: it shows up mainly when a change forces people to unlearn old habits and learn new ones. The supporting studies are small, such as Elrod and Tippett (1999), which covered only a couple of organisations.

Shingo Institute (2026). Lean-AI: A humanistic integration of Lean and AI receives the Shingo Publication Award. Shingo Institute, Utah State University, 23 June 2026. The Shingo Institute announced that Lean-AI by Adil Dalal and Elizabeth A. Cudney won its Publication Award. The book combines AI with lean methods such as value stream mapping, total productive maintenance and error-proofing (pokayoke), and presents AI that adds to human capability rather than replacing it, with people leading the change.

Wilberg, J. (2025). AI and wisdom: Notes from 2RCon. The Wider Angle (Substack), 12 December 2025. Notes by the host of a panel on AI and wisdom at the first Second Renaissance Conference (2RCon), with Dave Snowden, Igor Grossmann and George Por. Snowden warned that if people stop practising a task for five or six years, they may lose the ability to judge whether AI is doing it well.