maddaisy – cultivating innovation

Tag: enterprise-ai

  • McKinsey’s 25,000 AI Agents: First-Mover Advantage or the Industry’s Biggest Experiment?

    McKinsey now counts 25,000 AI agents among its workforce — roughly one for every 1.6 human employees. That ratio, disclosed by CEO Bob Sternfels at the Consumer Electronics Show and confirmed by the firm, makes the consultancy’s internal agentic build-out one of the most aggressive in professional services.

    The numbers have moved quickly. Eighteen months ago, McKinsey operated a few thousand agents. Today, through its AI arm QuantumBlack, AI-related work accounts for 40% of the firm’s output. The agents have saved an estimated 1.5 million hours on search and synthesis tasks. Non-client-facing headcount has fallen 25%, yet output from those teams has risen 10%.

    Sternfels’s stated ambition is to pair every one of McKinsey’s 40,000 employees with at least one AI agent within the next 18 months.

    Scale versus substance

    The scale is eye-catching. Whether it is meaningful depends on what you count as an agent and how you measure the return.

    McKinsey’s rivals are openly sceptical. EY’s global engineering chief has argued that “a handful of agents do the heavy lifting” and that value should be tracked through efficiency KPIs, not headcount. PwC’s chief AI officer has called agent count “probably the wrong measure”, advocating instead for quality and workflow optimisation. Their counterargument is clear: a smaller fleet of high-performing agents, rigorously measured, may deliver more than a vast deployment still being calibrated.

    The critique lands on familiar ground. As maddaisy examined earlier today, PwC’s 2026 Global CEO Survey found that 56% of chief executives still cannot point to revenue gains from their AI investments. The deployment-versus-outcomes gap is the central tension in enterprise AI right now, and McKinsey’s bet raises the question of whether the firm is racing ahead of the same problem — or solving it.

    From advisory to infrastructure

    The more consequential shift may be in McKinsey’s business model. Sternfels described a move away from the firm’s traditional fee-for-service approach toward a model where McKinsey works with clients to identify joint business cases and then helps underwrite the outcomes.

    This is a significant departure for a firm built on advisory fees and billable hours. It positions McKinsey less as a strategic counsellor and more as an infrastructure partner — one that brings its own AI workforce to bear on client problems and shares in the measurable results.

    QuantumBlack, with 1,700 people, now drives all of McKinsey’s AI initiatives. Alex Singla, the senior partner who co-leads the unit, has described the firm’s evolving recruitment profile: candidates who can move fluidly between traditional consulting and engineering, and who can work alongside AI rather than simply directing it.

    Boston Consulting Group is pursuing a similar direction, deploying “forward-deployed consultants” who build AI tools directly on client projects. But McKinsey’s scale of internal adoption — 25,000 agents embedded across the firm — gives it a data advantage that is harder to replicate. Every internal deployment generates operational insight into what works, what fails, and how agentic systems behave at enterprise scale.

    The governance question maddaisy has been tracking

    The timing of McKinsey’s announcement is worth noting against the backdrop of the agentic AI governance gap maddaisy covered earlier this week. Deloitte’s data showed that only 21% of companies have mature governance models for agentic AI, even as three-quarters plan to deploy it within two years. And a broader pattern has emerged across maddaisy’s recent coverage: enterprises are strategically confident about AI but operationally underprepared.

    McKinsey, as both a deployer and an adviser, sits at the intersection of this tension. If the firm can demonstrate that 25,000 agents operate reliably at scale — with governance, measurement, and accountability frameworks to match — it will have built the most persuasive case study in the industry. If the agents outrun oversight, the reputational exposure is equally significant. When an AI agent produces an analysis and the recommendation proves wrong, the liability question is not academic.

    What practitioners should watch

    For consulting professionals and enterprise leaders watching this play out, three things matter more than the headline number.

    First, the metric that matters is not agent count but outcome attribution. McKinsey’s 1.5 million hours saved is a process metric. The firm’s shift to underwriting client outcomes suggests it understands the need to move beyond efficiency and toward measurable business impact — the same gap that PwC’s CEO Survey identified industry-wide.

    Second, the talent model is changing faster than many firms acknowledge. McKinsey’s search for hybrid consultant-engineers, and BCG’s forward-deployed model, signal that the traditional consulting skill set is being augmented, not just supported, by AI fluency. Firms that treat AI as a productivity tool rather than a workforce design challenge will fall behind.

    Third, scale creates its own governance requirements. As McKinsey’s own Carolyn Dewar argued in Fortune, the real risk is not the technology but how leaders manage the fear and trust dynamics that surround it. Deploying 25,000 agents without the organisational infrastructure to govern them would validate every concern the firm’s rivals have raised.

    McKinsey’s wager is that first-mover scale in agentic AI creates a compounding advantage — more data, better workflows, stronger client proof points. The industry is about to find out whether volume leads to value, or whether a smaller, sharper approach gets there first.

  • The Governance Gap No One Is Closing: Why Agentic AI Is Outrunning Enterprise Oversight

    Three-quarters of enterprises plan to deploy agentic AI within two years. Only one in five has a mature governance model for it. That arithmetic should concern anyone responsible for enterprise technology strategy.

    The figures come from Deloitte’s 2026 State of AI in the Enterprise report, and they represent something more specific than the familiar story of AI adoption outpacing regulation. The challenge with agentic AI is not that rules do not exist — as maddaisy recently examined, the EU AI Act, Colorado’s AI Act, and a growing patchwork of global regulations are creating real enforcement deadlines. The challenge is that agentic systems demand a fundamentally different kind of oversight, and most organisations have not built the operational machinery to provide it.

    What makes agentic AI different

    Conventional AI systems — including the generative AI tools that have dominated enterprise adoption over the past two years — operate in an advisory mode. They suggest, summarise, draft, and classify. A human reviews the output and decides what to do with it. Governance for these systems, while imperfect, fits within existing frameworks: you audit the model, monitor outputs, and maintain a human in the decision loop.

    Agentic AI breaks that model. These systems are designed to plan, execute, and adjust autonomously — booking flights, approving procurement decisions, triaging customer complaints, or managing collections workflows without waiting for human sign-off. Oracle’s new agentic banking platform, launched in February 2026, illustrates the trajectory: domain-specific agents handle loan originations, credit decisioning, and compliance checks, with human oversight positioned as a “human-in-the-loop” role rather than a gatekeeping one.

    The distinction matters because it changes where governance must operate. With advisory AI, oversight happens after the model produces an output and before a human acts on it. With agentic AI, the system is the actor. Governance must be embedded in real time — monitoring agent behaviour as it happens, enforcing boundaries on what an agent can and cannot do, and maintaining audit trails that capture not just decisions but the full chain of reasoning and actions that led to them.

    The 21% problem

    Deloitte’s finding that only 21% of companies have a mature agentic AI governance model is striking, but the detail beneath it is more revealing. In Singapore, where deployment ambitions are among the highest globally — 72% of businesses plan to deploy agentic AI across multiple operational areas within two years, up from 15% today — the mature governance figure drops to just 14%.

    As maddaisy noted in its analysis of the broader Deloitte report, this fits a wider pattern: organisations are increasingly confident in their AI strategy but declining in readiness on the operational foundations needed to execute it. The agentic governance gap is perhaps the sharpest expression of this paradox — a technology that is advancing from pilot to production while the controls needed to run it safely remain in early stages.

    Half of Singapore respondents reported using a patchwork of public and internal proprietary frameworks to assess agent risk and performance. That is not a governance model — it is improvisation.

    Why existing frameworks fall short

    The AI Trends Report 2026, published by statworx and AI Hub Frankfurt, identifies three operational disciplines that are becoming foundational for reliable agentic AI: AI governance, DataOps, and what the report terms AgentOps — the operational layer for managing autonomous AI agents in production.

    AgentOps is a useful concept because it captures what most enterprise governance frameworks currently lack. Traditional AI governance focuses on model development: training data quality, bias testing, documentation, and approval workflows before deployment. That is necessary but insufficient for systems that learn, adapt, and take actions in production environments.

    Agentic systems require runtime governance: clear boundaries on agent autonomy (what decisions can the agent make independently, and which require escalation?), real-time monitoring of agent behaviour against expected parameters, kill switches for when agents drift outside acceptable bounds, and comprehensive audit trails that regulators can inspect after the fact.

    The EU AI Act’s requirements for high-risk systems — documented risk management, technical logging, human oversight mechanisms, and conformity assessments — implicitly assume this kind of operational infrastructure. But most organisations have not yet translated those requirements into engineering reality.

    Deployment is not waiting for governance

    The uncomfortable truth is that agentic AI is entering production regardless of whether governance is ready. Oracle is shipping banking agents now. Companies like AMD and Heathrow Airport are deploying autonomous agents in customer experience roles. Gartner predicts agentic systems will autonomously resolve 80% of customer service issues by 2028.

    Constellation Research offers a useful counterweight to the hype, arguing that agentic AI is “more of a feature than a revolution” and that the real measure of value is decision velocity — how quickly smaller decision trees and processes can be automated at scale. This framing is helpful because it reduces the abstraction. An AI agent rebooking a flight is not a paradigm shift; it is a process automation with a more sophisticated reasoning layer. But that reasoning layer is precisely what makes governance harder. The agent is not following a static script — it is making contextual judgements, and those judgements need oversight.

    What the governance gap actually costs

    The business case for closing the governance gap is not primarily about regulatory fines, though those are real. It is about operational risk. When an agentic system autonomously commits to a procurement decision, misprices a financial product, or gives a customer incorrect information with real-world consequences, the liability question is immediate and the reputational exposure is direct.

    It is also about scaling. Deloitte’s data shows that companies with stronger governance foundations are deploying agentic AI more successfully — they start with lower-risk use cases, build governance capabilities alongside deployment, and scale deliberately. Organisations that skip the governance step find themselves either slowing down when something goes wrong or, worse, not knowing that something has gone wrong until a regulator or customer tells them.

    What needs to happen

    The gap between agentic AI deployment and agentic AI governance is not going to close on its own. Three practical steps can narrow it.

    Define agent autonomy boundaries explicitly. For every agentic AI deployment, organisations need a clear specification of what the agent can do independently, what requires human approval, and what is prohibited. These boundaries should be codified in the system, not just written in a policy document. The Oracle banking platform’s “human-in-the-loop” architecture is one model, but even that needs specificity about when and how the loop engages.

    Invest in runtime monitoring, not just pre-deployment testing. The governance challenge with agentic AI is that it operates continuously and adapts to context. Pre-deployment audits are necessary but not sufficient. Organisations need real-time monitoring that tracks agent decisions against expected parameters and flags anomalies before they compound.

    Build audit trails as engineering infrastructure. When a regulator asks how an agent arrived at a specific decision — and under the EU AI Act, they will — the organisation needs to produce a complete chain of the agent’s reasoning, data inputs, and actions. This is not a reporting challenge; it is an engineering one that needs to be designed into the system from the start, not retrofitted after deployment.

    The agentic AI governance gap is not a future problem. It is a present one, widening with every new deployment. The organisations that treat governance as a technical discipline — building it into the engineering of their agentic systems rather than bolting it on as a compliance afterthought — will have a structural advantage as the technology matures. Those that do not will discover, as many enterprises have with earlier waves of technology adoption, that the cost of retrofitting oversight always exceeds the cost of building it in.

  • The AI ROI Crisis: Why 56% of CEOs Still Cannot Prove Value From Their AI Investments

    More than half of the world’s chief executives cannot point to revenue gains from their AI investments. That is not a fringe finding from an alarmist report — it is the central conclusion of PwC’s 2026 Global CEO Survey, which polled 4,454 business leaders and was released at Davos in January.

    The figure — 56% reporting no measurable revenue uplift from AI — lands at a moment when enterprise AI spending continues to accelerate. Budgets are growing, headcounts in AI-adjacent roles are expanding, and the consulting industry’s order books are thick with transformation mandates. Yet the returns remain stubbornly elusive for the majority. Only 12% of CEOs surveyed reported achieving both revenue growth and cost reduction from their AI programmes.

    This is not, as some commentary has framed it, evidence that AI does not work. It is evidence that most organisations have not yet figured out how to make it work — a distinction that matters enormously for practitioners and consultants navigating the current landscape.

    The pattern maddaisy has been tracking

    PwC’s data confirms a dynamic that maddaisy has examined from several angles over the past fortnight. Deloitte’s 2026 State of AI report revealed that enterprises feel more strategically confident about AI than ever, yet less operationally ready — a paradox the report’s authors attributed to weak data infrastructure, insufficient talent, and immature governance. Research published in Harvard Business Review, which maddaisy covered last week, found that AI tools were making employees busier rather than more productive, with efficiency gains absorbed by task expansion rather than redirected toward higher-value work.

    The ROI gap is what happens when these operational failures compound. Organisations adopt AI tools without redesigning workflows. They measure deployment (how many teams have access) rather than outcomes (what changed as a result). They invest in the technology layer while underinvesting in the organisational layer — the process changes, role redesigns, and measurement frameworks that turn a pilot into a production capability.

    A measurement problem masquerading as a technology problem

    One of the more revealing aspects of the PwC data is what it implies about how enterprises are tracking AI value. CEO revenue confidence sits at a five-year low of 30%, and this pessimism correlates with the inability to demonstrate AI returns. But the question is whether the returns genuinely are not there, or whether organisations simply lack the instrumentation to detect them.

    The answer, for many enterprises, is likely both. Some AI deployments are genuinely failing to deliver — deployed in the wrong processes, aimed at the wrong problems, or undermined by poor data quality. But others may be generating real value that never surfaces in the metrics that CEOs review. Time saved in middle-office processes, reduced error rates in document handling, faster iteration cycles in product development — these are real gains, but they rarely appear on a revenue line unless someone has built the measurement architecture to capture them.

    This is a familiar pattern in enterprise technology adoption. The early years of cloud computing saw similar complaints: organisations spent heavily on migration but struggled to quantify the business impact beyond infrastructure cost reduction. The value was real — in agility, speed to market, and developer productivity — but it took years for finance teams to develop frameworks that could track it. AI is following the same trajectory, with the added complication that its benefits are often diffuse, spread across many small improvements rather than concentrated in a single, measurable outcome.

    What the 12% are doing differently

    PwC’s survey found that the minority of organisations achieving both revenue and cost benefits from AI share common characteristics. They have invested in data foundations — not just data lakes and pipelines, but governance structures that ensure data quality, accessibility, and appropriate use. They have moved beyond isolated pilots to embed AI into core business processes. And they treat AI adoption as an organisational change programme, not a technology deployment.

    None of this is conceptually new. Consultancies have been advising clients on change management, data governance, and process redesign for decades. What is new is the speed at which the gap between leaders and laggards is widening. The 12% who have cracked the ROI equation are pulling ahead, using AI-generated insights to inform strategy, AI-automated processes to reduce costs, and AI-enhanced products to capture new revenue. The 56% who have not are still running pilots, still debating governance frameworks, and still struggling to answer the board’s most basic question: what are we getting for this money?

    The consulting industry’s uncomfortable position

    For the consulting sector, PwC’s findings create an awkward tension. The firms advising enterprises on AI strategy are, in many cases, the same firms whose clients cannot demonstrate returns. This is not necessarily a reflection of poor advice — the operational barriers to AI value are genuine and deep — but it does raise questions about what consulting engagements are actually delivering.

    As maddaisy noted when examining Capgemini’s recent results, CEO Aiman Ezzat explicitly framed the company’s direction as a shift “from AI hype to AI realism.” Generative AI bookings exceeded 8% of Capgemini’s total for the year, but the company’s 2026 revenue guidance fell slightly below analyst expectations — a reminder that even firms positioning AI at the centre of their strategy face questions about whether the investment is translating into proportional growth.

    The risk for consultancies is that the ROI gap erodes client confidence in AI-related engagements. If more than half of CEOs see no revenue benefit, the appetite for further AI spending — and the advisory services that accompany it — may tighten. The counter-argument, which PwC’s own data supports, is that the solution to poor AI returns is not less AI but better AI implementation. That is a consulting engagement waiting to happen, provided firms can credibly demonstrate that they know how to close the gap.

    What practitioners should watch

    Three developments will shape whether the AI ROI picture improves or deteriorates over the coming quarters.

    First, the regulatory environment is tightening. As maddaisy recently examined, the EU AI Act’s high-risk system requirements become enforceable in August 2026, and state-level legislation in the US is creating a fragmented compliance landscape. Compliance costs will add to the total cost of AI ownership, making the ROI equation harder to balance for organisations that have not already built governance into their deployment model.

    Second, the rise of agentic AI — systems that plan and execute tasks with minimal human oversight — will test whether organisations can capture value from more autonomous AI without losing control. Deloitte’s data showed a nearly fivefold increase in planned agentic AI deployments over the next two years, but only one in five companies has a mature governance model for autonomous agents. The ROI potential is significant; so is the risk of expensive failures.

    Third, and perhaps most importantly, watch for a shift in how organisations measure AI value. The enterprises that move beyond “did revenue go up?” to more granular metrics — cycle time reduction, error rate improvement, employee capacity freed for strategic work — will be better positioned to demonstrate returns and justify continued investment. The measurement framework may matter as much as the technology itself.

    PwC’s 56% figure is striking, but it is a snapshot of a transition, not a verdict on AI’s potential. The technology is not the bottleneck. Execution is. And for consultants and practitioners, that distinction is where the real work — and the real opportunity — lies.

  • Deploy First, Fix Later: How Poor AI Rollouts Are Engineering Burnout Into the System

    Over the past week, maddaisy has examined two dimensions of the AI-burnout nexus: the work intensification mechanisms that make employees busier rather than better, and the organisational frameworks that might address the people side of the equation. But there is a third dimension that sits upstream of both: the technology deployment itself.

    Before burnout becomes a management problem, it is often an implementation problem. And a growing body of evidence suggests that the way organisations roll out AI tools — rushed timelines, inadequate capability-building, and a persistent belief that go-live equals readiness — is baking unsustainable working conditions into the system from day one.

    The capability gap that no one budgets for

    A detailed analysis published by CIO.com this month draws on workforce upskilling data from large-scale enterprise rollouts to make a blunt assessment: systems rarely fail because the technology does not work. They fail because the organisation has not built the infrastructure to support the people using them.

    The numbers are sobering. Organisations routinely experience productivity drops of 30 to 40% within the first 90 days of a major technology go-live when workforce capability has not been adequately addressed. Support tickets triple. Workaround behaviours — offline spreadsheets, manual reconciliations, shadow systems — proliferate as employees revert to what feels safe and controllable. Post-go-live support costs run 40 to 60% over budget. ROI timelines slip by six to 12 months.

    None of this is caused by employee resistance or technological failure. It is the predictable consequence of treating capability-building as a training event rather than a systemic requirement.

    The 10% problem

    Perhaps the most striking data point concerns training transfer. Research on technology implementation training suggests that only 10 to 20% of skills learned in formal programmes translate into sustained on-the-job performance. The issue is not poor course design. It is the assumption that a three-day training session two weeks before go-live can substitute for the repeated practice, pattern recognition, and feedback loops that genuine capability requires.

    This matters directly for the burnout conversation. When employees lack the capability to operate new systems confidently, every minor issue becomes a support ticket. Every exception becomes an escalation. The cognitive load of navigating unfamiliar tools while maintaining output expectations creates exactly the kind of sustained pressure that Harvard Business Review’s recent research identified as driving work intensification. The tools are not the problem — the deployment is.

    Where AI deployments differ from traditional IT rollouts

    Enterprise technology deployments have always carried capability risks. But AI introduces specific complications that amplify them.

    First, AI tools change the scope of work, not just the process. As maddaisy’s earlier analysis noted, employees using AI do not simply do the same tasks faster — they absorb new responsibilities, blur role boundaries, and take on work that previously justified additional headcount. A traditional ERP deployment changes how someone does their job. An AI deployment can change what their job is. Upskilling programmes designed around process training cannot address a shift that is fundamentally about role redesign.

    Second, AI output is probabilistic, not deterministic. An ERP system produces the same result given the same inputs. An AI tool might produce different outputs each time, requiring users to exercise judgement about quality, accuracy, and appropriateness. That judgement cannot be trained in a classroom — it is built through experience, and it demands a kind of cognitive engagement that is qualitatively different from following a process manual.

    Third, AI raises the visibility of individual output. When an AI tool enables someone to produce a first draft in minutes rather than hours, the speed becomes the new baseline expectation. Employees who are still building capability with the tool face pressure to match output norms set by early adopters or by the tool’s theoretical capacity. The result is the self-reinforcing acceleration cycle that researchers have now documented repeatedly.

    Governance as deployment infrastructure

    The CIO.com analysis proposes treating upskilling as a governance concern rather than a training administration task — embedding capability-building into the transformation workstream with the same rigour as data migration or integration testing. This means defining capability in behavioural terms tied to business processes, starting practice in sandbox environments as soon as process designs stabilise, and measuring performance data rather than training completion rates.

    One practical recommendation stands out: establishing a “performance council” distinct from the training team. This group — composed of process owners, frontline managers, and high-performing end users — meets weekly to review whether people are performing reliably in live operations. They examine error rates, support ticket patterns, workaround behaviours, and time-to-competency. Critically, they have the authority to pause rollouts when the data shows capability is not sticking.

    This is not a radical proposal. It is the kind of operational discipline that well-run technology programmes have always applied to infrastructure and integration. The fact that it needs to be articulated separately for workforce capability suggests how often organisations skip the people side of deployment planning.

    Connecting the implementation layer to the burnout evidence

    The burnout data that has emerged over recent weeks tells a consistent story. DHR Global reports that 83% of workers experience some degree of burnout, with engagement dropping 24 percentage points in a single year. Only 34% of employees say their organisation has communicated AI’s workplace impact clearly. Among entry-level staff, that figure falls to 12%.

    These are not just management failures. They are deployment failures. When organisations ship AI tools without building the capability infrastructure to support them — without adequate training transfer, without performance monitoring, without role redesign — they are not just creating a change management problem. They are creating a technical debt of human capability that compounds over time, manifesting as burnout, disengagement, and quiet resistance.

    As Deloitte’s 2026 AI report made clear, the gap between strategic confidence and operational readiness remains the defining feature of enterprise AI. The implementation evidence suggests that closing this gap requires treating workforce capability not as a soft skill or an HR deliverable, but as core deployment infrastructure — as essential as the data pipeline and as measurable as system uptime.

    What this means for the next phase

    The organisations that avoid engineering burnout into their AI deployments will share a common trait: they will treat go-live as the beginning of capability-building, not the end. They will budget for the 90-day productivity dip and plan for it rather than being surprised by it. They will measure whether people can perform reliably at scale under real business pressure, not whether they completed a training module.

    Most importantly, they will recognise that the fastest way to undermine an AI investment is not a technical failure — it is deploying capable technology to an unprepared workforce and then measuring success by adoption rates alone. The tools work. The question, as it has always been with technology transformations, is whether the organisation around them does too.

  • The Productivity Trap: Why AI Tools Are Making Employees Busier, Not Better

    The promise of AI in the workplace has always rested on a simple equation: automate the routine, free up humans for higher-value work. It is the pitch that has launched a thousand consulting engagements and underpinned billions in enterprise software investment. But new research suggests the equation may be running in reverse.

    A study published in Harvard Business Review this month, based on eight months of ethnographic research at a US technology company with roughly 200 employees, found that AI tools did not reduce workload. They intensified it. Employees worked faster, took on more tasks, and felt busier than before — despite the efficiency gains that AI was supposed to deliver.

    The researchers, Aruna Ranganathan and Xingqi Maggie Ye from UC Berkeley’s Haas School of Business, identified three distinct mechanisms through which AI was quietly ratcheting up the pressure.

    Task expansion: doing more with less becomes doing more with the same

    The first mechanism was task expansion. When AI tools made certain tasks faster, employees did not use the freed-up time for strategic thinking or creative work. Instead, they absorbed responsibilities that had previously belonged to other roles. Product managers and designers started writing code. Researchers took on engineering tasks. Workers assumed duties that, in the researchers’ words, “might previously have justified additional help.”

    This is a pattern that will be familiar to anyone who has watched organisations respond to efficiency gains over the past two decades. The spreadsheet did not eliminate accounting departments — it gave accountants more to do. Email did not reduce communication overhead — it multiplied it. AI appears to be following the same trajectory, with one critical difference: the speed at which task expansion occurs is considerably faster.

    The study also identified a secondary burden. Engineers found themselves spending additional time reviewing their colleagues’ AI-assisted work, a form of quality assurance that had not existed before because the work itself had not existed before.

    The vanishing boundary between work and rest

    The second mechanism was the blurring of work-life boundaries. Employees began incorporating work into what had previously been downtime — lunch breaks, gaps between meetings, even waiting for files to load. Because interacting with an AI tool felt, as one participant put it, “closer to chatting than to undertaking a formal task,” the psychological barrier to picking up work during off moments effectively disappeared.

    This is a subtler form of intensification, and arguably a more dangerous one. The physical cues that traditionally separated work from rest — closing a document, leaving a desk, switching off a screen — lose their power when work can be initiated through a conversational prompt on any device. The distinction between being productive and being available collapses.

    The multitasking illusion

    The third mechanism was increased multitasking. With AI handling parts of each task, employees managed multiple active threads simultaneously, creating what the researchers described as “continual switching of attention.” Worse, because AI made fast output visible to colleagues, it raised speed expectations across teams. Employees felt pressure not just to work with AI, but to keep pace with the output norms that AI made possible.

    The result was a self-reinforcing cycle: AI accelerated certain tasks, which raised expectations for speed, which made workers more reliant on AI, which widened the scope of what they attempted. As one engineer told the researchers, they felt “busier than before” despite the supposed time savings.

    Not new, but newly documented

    It is worth noting that the HBR findings are not entirely without precedent. Research from the University of Chicago and the University of Copenhagen published last year found that AI chatbots saved workers only about an hour per week — and that the tools created enough new tasks to largely nullify even that modest gain. What Ranganathan and Ye have added is the ethnographic depth: eight months of direct observation, 40 interviews, and a detailed account of the mechanisms through which intensification occurs.

    For consulting practitioners, this distinction matters. The earlier studies quantified the problem. This one explains the pathways. And that makes it actionable.

    Connecting the dots: from strategy gap to people gap

    The timing of this research is significant. As maddaisy noted last week, Deloitte’s 2026 State of AI in the Enterprise report revealed a widening gap between strategic confidence and operational readiness. Forty-two per cent of companies consider their AI strategy highly prepared, yet fewer feel equipped to execute it.

    The burnout research suggests one reason that gap persists: organisations are measuring AI adoption by deployment metrics — tools rolled out, processes automated, tasks per hour — while ignoring the human cost of that adoption. The operational readiness problem is not just about data pipelines and integration architecture. It is about whether the people using these tools can sustain the pace that the tools enable.

    This also has implications for the AI governance frameworks now coming into force across Europe and beyond. Most governance discussions focus on algorithmic bias, data privacy, and transparency. Workforce wellbeing — whether AI deployment is creating sustainable working conditions — barely features. As enforcement mechanisms sharpen, that gap may become harder to defend.

    What practitioners should watch

    The HBR researchers recommend what they call “AI practice” — a set of organisational disciplines designed to counteract intensification. These include intentional pauses (structured breaks for assessment), sequencing (deliberate pacing rather than continuous output), and human grounding (protected time for dialogue and connection).

    These are not revolutionary ideas. They are, in essence, good management practices adapted for an AI-augmented workplace. But the fact that they need to be articulated at all tells a story about how many organisations are deploying AI tools without thinking through the second-order effects on their people.

    For consultancies advising on AI transformation, this research is a prompt to broaden the conversation. Deployment is not the finish line. If the tools make employees faster but not better — busier but not more effective — then the productivity gains that justified the investment may prove temporary, eroded by turnover, cognitive fatigue, and declining work quality.

    The question, as the researchers put it, is not whether AI will change work, but whether organisations will actively shape that change — or let it quietly shape them.

  • From Principles to Penalties: AI Governance Enters Its Enforcement Era

    For the better part of a decade, AI governance lived in the realm of principles. Organisations published ethics charters, governments convened expert panels, and everyone agreed that responsible AI mattered — without agreeing on what, precisely, that meant in practice.

    That era is ending. In 2026, AI governance is shifting from aspiration to obligation, from white papers to enforcement deadlines with real financial consequences. The transition is not sudden — it has been building for years — but the concentration of regulatory milestones in the coming months marks a genuine inflection point for any organisation deploying AI at scale.

    The regulatory calendar thickens

    Three developments are converging to make 2026 the year that AI compliance moves from a planning exercise to an operational requirement.

    First, the EU AI Act reaches its most consequential milestone on 2 August 2026, when requirements for high-risk AI systems become fully enforceable. These cover AI used in employment decisions, credit scoring, education, and law enforcement — areas where automated decisions directly affect people’s lives. Non-compliance carries penalties of up to €35 million or 7% of global annual turnover, whichever is higher. For context, that exceeds the maximum GDPR fine by a significant margin.

    Second, the United States is developing its own patchwork of state-level AI laws. Colorado’s AI Act, taking effect in June 2026, requires deployers of high-risk AI systems to exercise reasonable care against algorithmic discrimination, conduct impact assessments, and provide consumer transparency. California’s SB 53 mandates that frontier AI developers publish safety frameworks and implement incident response measures. Illinois, New York City, Utah, and Texas have all enacted targeted AI requirements of their own.

    Third, the federal government has entered the fray — not to simplify matters, but to complicate them. President Trump’s December 2025 Executive Order signals an intent to consolidate AI oversight at the federal level and potentially pre-empt state regulations deemed “onerous.” A newly established AI Litigation Task Force has been directed to identify state laws for possible legal challenge. But the executive order itself does not create federal AI standards, nor does it suspend existing state laws. As legal analysts at Gunderson Dettmer have noted, it is “a statement of principles and set of tools,” not an amnesty or moratorium.

    The practical result is a compliance environment that is fragmented, fast-moving, and increasingly consequential.

    The operational gap maddaisy has been tracking

    This regulatory acceleration arrives at an awkward moment for most enterprises. As maddaisy recently examined through Deloitte’s 2026 State of AI report, organisations report growing confidence in their AI strategy but declining readiness on the operational foundations — infrastructure, data quality, risk management, and talent — needed to execute it. The governance gap is a specific instance of this broader pattern: many enterprises can articulate what responsible AI should look like, far fewer have built the internal machinery to demonstrate compliance under real regulatory scrutiny.

    The EU AI Act, for example, does not simply require that organisations have a policy. It mandates documented risk management systems, high-quality training data practices, technical logging, transparency to users, human oversight mechanisms, and conformity assessments — all subject to audit. For companies that have treated AI governance as a communications exercise rather than an engineering discipline, the compliance gap is substantial.

    A global picture with local friction

    The regulatory picture extends well beyond the EU and US. ISACA’s recent comparison of the EU AI Act and China’s AI Governance Framework 2.0 highlights how the two largest regulatory regimes take fundamentally different approaches. The EU classifies risk through a technology-centric, tiered system. China uses a dynamic, multidimensional model based on application scenario, intelligence level, and scale — and requires pre-launch government approval for any public-facing AI system.

    South Korea and Vietnam have implemented dedicated AI laws in 2026. India is hosting its AI Impact Summit this week, the first major global AI event hosted in the Global South, signalling the country’s intent to shape governance norms rather than simply receive them.

    For multinational organisations, this creates a familiar but intensifying challenge: complying with the strictest applicable standard while operating across jurisdictions with diverging requirements. The sovereignty dimensions that maddaisy explored in Capgemini’s approach to European digital sovereignty are directly relevant here. Data residency, model provenance, and supply chain transparency are no longer optional governance aspirations — they are becoming regulatory requirements with enforcement teeth.

    What practitioners should be doing now

    The shift from principles to enforcement carries specific implications for consultants and technology leaders.

    Inventory first, comply second. Before addressing any specific regulation, organisations need a clear map of where AI systems make or influence consequential decisions. Many companies still lack a comprehensive AI inventory — they cannot tell regulators what AI they are running, let alone demonstrate how it is governed.

    Build for the strictest standard. The temptation to wait for regulatory clarity — particularly in the US, where federal pre-emption remains uncertain — is understandable but risky. Companies operating across state lines or internationally will likely need to meet the EU AI Act’s requirements regardless, given its extraterritorial reach. Building governance infrastructure to that standard provides a defensible baseline.

    Treat governance as engineering, not policy. The new regulations demand technical evidence: logging, bias audits, explainability frameworks, documented data provenance. These are engineering problems that require engineering solutions. A well-written ethics policy will not satisfy a regulator asking for an audit trail of model decisions.

    Watch the US federal-state tension closely. The December 2025 executive order has created genuine uncertainty about whether state AI laws like Colorado’s will survive federal challenge. But as compliance analysts have noted, state attorneys general retain broad enforcement authority under existing consumer protection and anti-discrimination statutes — even if AI-specific laws are curtailed. The enforcement risk does not disappear; it simply shifts shape.

    The year governance becomes a line item

    None of this is unexpected. The trajectory from voluntary principles to binding regulation follows the same path that data protection, financial reporting, and environmental standards have all taken. What is notable about 2026 is the speed and breadth of the convergence: multiple jurisdictions, multiple enforcement mechanisms, and penalties calibrated to be genuinely material, all arriving within the same 12-month window.

    For organisations that have invested in governance infrastructure, this is the moment that investment begins to pay off — not as a cost centre, but as a competitive advantage. For those that have not, the runway is shortening. The question is no longer whether AI governance matters. It is whether the operational machinery exists to prove it.

  • Strategically Ready, Operationally Stuck: What Deloitte’s 2026 AI Report Reveals About the Enterprise Gap

    Deloitte’s 2026 State of AI in the Enterprise report, based on a survey of 3,235 senior leaders across 24 countries, delivers a finding that should give every consulting practitioner pause: more companies than ever believe their AI strategy is sound, but fewer feel ready to actually execute it.

    The gap between strategic confidence and operational readiness is the defining tension in enterprise AI right now — and it has implications for how consultancies sell, deliver, and staff their AI practices.

    The preparedness paradox

    According to Deloitte’s survey, 42% of companies now consider their strategy highly prepared for AI adoption, up from the previous year. That sounds encouraging until you read the next line: those same organisations report feeling less prepared than before on infrastructure, data, risk management, and talent.

    This is not a minor statistical wrinkle. It suggests that enterprises have become more fluent in talking about AI — they can articulate a vision, identify use cases, perhaps even secure board-level backing — but the operational foundations needed to move from pilot to production remain weak. Worker access to AI rose by 50% in 2025, and organisations expect the number with 40% or more of AI projects in production to double within six months. Whether the infrastructure exists to support that ambition is another matter entirely.

    The skills picture reinforces the point. Insufficient worker skills were identified as the biggest barrier to integrating AI into existing workflows. The most common response — educating the broader workforce to raise AI fluency (53%) — is necessary but not sufficient. Far fewer organisations are redesigning roles, workflows, or career paths around AI. Education tells people what AI can do. Restructuring tells the organisation how to use it.

    Sovereign AI moves from policy to procurement

    One of the report’s more significant findings is the emergence of sovereign AI as a practical enterprise concern, not just a political talking point. In Singapore, 77% of businesses surveyed said that data residency and in-country or in-region compute considerations are now important to their strategic planning.

    This echoes a dynamic that maddaisy recently examined in the European context, where Capgemini’s CEO Aiman Ezzat outlined a pragmatic four-layer sovereignty framework while signing partnerships with all three major US hyperscalers. Deloitte’s data suggests that the same tension — between the desire for local control and the reality of global infrastructure — is playing out across Asia Pacific and the Middle East, not just Europe.

    As Computer Weekly reported, organisations in the Middle East are approaching a similar inflection point, with sovereign and agentic AI expected to define the next phase of digital transformation. The pattern is consistent: governments want control, enterprises want capability, and the consulting industry is positioning itself to broker the compromise.

    Agentic AI outpaces its guardrails

    Perhaps the most striking data point in the Deloitte report concerns agentic AI — systems designed to plan, execute, and optimise tasks with minimal human oversight. Usage is set to rise sharply over the next two years, but only one in five companies has a mature governance model for autonomous AI agents.

    In Singapore, the numbers are particularly stark: 72% of businesses plan to deploy agentic AI across several operational areas within two years, up from just 15% today. That is a nearly fivefold increase in deployment with governance frameworks that are, by the report’s own assessment, not yet fit for purpose.

    The use cases are real enough. Deloitte cites financial services firms using AI agents to capture meeting actions and track follow-through, airlines deploying agents for common customer transactions, and manufacturers using autonomous systems to optimise product development trade-offs. These are not speculative applications — they are in production. But the governance question is not academic either. When an AI agent autonomously rebooks a flight or commits to a procurement decision, the question of accountability, audit trails, and regulatory compliance becomes urgent.

    Accenture stops counting

    A useful counterpoint to Deloitte’s enterprise survey comes from the supply side. In December, Accenture announced that it would stop separately reporting its advanced AI bookings — a category covering generative, agentic, and physical AI — because the technology had become “so pervasive” it was embedded across nearly everything the firm delivers.

    CEO Julie Sweet framed this as a sign of maturity: AI is no longer a distinct workstream but a feature of all client engagements. Advanced AI bookings hit $2.2 billion in the first quarter of fiscal 2026, double the prior year. The company has reached its target of 80,000 AI and data professionals.

    There is a less generous reading, of course. Stopping disclosure also makes it harder for investors and analysts to track whether AI is generating new revenue or simply being relabelled within existing services. But taken alongside Accenture’s acquisition of Faculty, a UK-based AI firm, in February 2026, the direction is clear: the major consultancies are absorbing AI into their core delivery model rather than treating it as a separate practice.

    What practitioners should watch

    The Deloitte report’s most useful contribution is not its optimism about AI’s potential — the industry has no shortage of that — but its honest accounting of where enterprises actually stand. Three signals are worth tracking.

    First, the strategy-execution gap will drive consulting demand, but not the kind that involves slide decks and maturity assessments. Enterprises need help with the operational plumbing: data architecture, infrastructure modernisation, and workflow redesign. The consultancies that can deliver engineering alongside strategy will win the next phase.

    Second, sovereign AI is becoming a procurement criterion, not just a policy aspiration. For firms operating across multiple jurisdictions — which includes most enterprise consulting clients — this means every AI deployment now carries a compliance dimension that did not exist two years ago.

    Third, the governance gap around agentic AI is a genuine risk, not a theoretical concern. As autonomous systems move from pilots to production, the organisations that invested early in oversight frameworks will have a structural advantage. Those that did not will find themselves either slowing down or taking on liability they have not fully priced.

    The AI story in 2026 is less about whether the technology works — increasingly, it does — and more about whether organisations can build the operational, regulatory, and governance foundations to use it responsibly at scale. The Deloitte data suggests most are not there yet, but they think they are. That gap is where the real work begins.

  • Capgemini’s Sovereignty Playbook: Bridging Europe’s AI Ambitions and American Infrastructure

    In the space of a single week in early February, Capgemini signed sovereignty-focused partnerships with all three major US hyperscalers — Google Cloud, AWS, and Microsoft. Days later, CEO Aiman Ezzat used the company’s full-year results presentation to publicly dismiss calls for complete European tech autonomy.

    The juxtaposition was deliberate, and it tells a more interesting story than the headline financials. Capgemini is not just adding AI capabilities to its consulting portfolio. It is building a distinct commercial proposition around one of Europe’s most politically charged technology questions: who controls the infrastructure that enterprises depend on?

    The gap between rhetoric and reality

    European digital sovereignty has been a policy preoccupation for several years now, accelerated by concerns over US government data access, the dominance of American cloud providers, and the growing strategic importance of AI infrastructure. The European Commission has pushed for greater technological independence. Member states have launched sovereign cloud initiatives. The language of autonomy is everywhere.

    The reality, as Ezzat put it bluntly during the post-earnings call, is more complicated. “There is no such thing as absolute sovereignty,” he told journalists. “Nobody has it, because no one has sovereignty over the entire value chain required to deliver services.”

    This is not a controversial claim among practitioners, but it is a notable one for a CEO whose company is headquartered in Paris and whose chairman also leads the digital working group at the European Round Table for Industry. Ezzat has been discussing sovereignty with the European Commission in Brussels and at Davos. His position is informed, not casual.

    A four-layer framework

    Ezzat outlined what amounts to a practical sovereignty framework built around four layers: data, operations, regulation, and technology. His argument is that Europe has meaningful independence on the first three — data residency and governance, operational control over services, and regulatory authority through instruments like GDPR and the AI Act. The fourth layer, the underlying technology stack, is where US Big Tech dominance means full independence is neither achievable nor, in his view, desirable.

    Rather than pursuing autonomy at every layer, Capgemini’s approach is to offer clients “the right sovereignty solution based on the use case, the client environment, the government.” In practice, this means European-managed services running on American infrastructure — sovereign in governance and operations, pragmatic on technology.

    As maddaisy noted earlier this week in examining Capgemini’s full-year results, the company estimates that over 50% of service contracts will include sovereignty requirements by 2029, up from just 5% in 2025. That trajectory, if it holds, represents a structural shift in how enterprise IT contracts are structured across Europe.

    Three partnerships, one message

    The timing of Capgemini’s hyperscaler announcements was no coincidence. On 6 February, the company expanded its partnership with Google Cloud, establishing a Sovereign Cloud Delivery Practice and Centre of Excellence. Capgemini will operate as a Google Distributed Cloud air-gapped operator — meaning it can deliver fully managed services with total isolation from the public internet, suited to defence, intelligence, and critical infrastructure clients.

    On 9 February, a similar announcement followed with AWS, focused on sovereign-ready cloud and AI capabilities. Two days later, Capgemini formalised integrated sovereignty solutions with Microsoft. Three announcements in five days, each offering variations on the same theme: Capgemini as the European operator sitting between the client and the American cloud.

    This is a positioning play with genuine commercial substance. For European enterprises navigating tightening regulation — particularly public sector organisations, financial institutions, and healthcare providers — the question is not whether to use cloud services but how to use them in ways that satisfy increasingly specific sovereignty requirements. Capgemini is betting it can be the answer to that question.

    Where AI and sovereignty converge

    The sovereignty proposition becomes more compelling when combined with Capgemini’s broader AI pivot. Generative and agentic AI bookings exceeded 10% of group bookings in Q4 2025, and the company has trained 310,000 employees on generative AI and 194,000 on agentic AI — systems designed to take autonomous actions rather than simply generate content.

    AI workloads are particularly sensitive from a sovereignty perspective. They involve large volumes of proprietary data, often require access to regulated information, and increasingly touch decision-making processes that organisations want to keep within controlled environments. A sovereign AI solution — where the model runs on infrastructure governed under European jurisdiction, operated by a European firm, but built on the technical capabilities of a US hyperscaler — addresses a specific and growing need.

    Ezzat framed AI itself with characteristic pragmatism in a separate interview with Fortune. “AI is a business. It is not a technology,” he said, warning leaders against treating it as a “black box being managed separately.” His caution against AI FOMO — “You don’t want to be too ahead of the learning curve. If you are, you’re investing and building capabilities that nobody wants” — suggests a company that has learned from watching the metaverse hype cycle play out.

    What to watch

    Capgemini’s sovereignty strategy raises several questions worth tracking. First, whether the 50%-by-2029 estimate for sovereignty-embedded contracts proves accurate, or whether it reflects the kind of optimistic forecasting that consulting firms are prone to when promoting a new service line. Second, how European competitors — particularly Atos, which has its own sovereignty ambitions, and smaller European cloud providers — respond to Capgemini’s hyperscaler-partnered model. Third, whether the European Commission’s own stance on sovereignty tilts toward the pragmatic Capgemini position or toward more aggressive technological independence.

    For consultants and practitioners, the practical takeaway is straightforward: sovereignty is moving from a compliance checkbox to a structural feature of European enterprise contracts. The firms that build credible delivery capabilities around it now — not just policy positions, but operational partnerships and trained workforces — will have a meaningful advantage as regulation tightens. Capgemini has placed its bet. The question is whether the market follows.

  • Capgemini’s AI Pivot: From Hype to Realism, with €22.5 Billion at Stake

    Capgemini’s full-year 2025 results, released on 13 February, offered one of the clearest snapshots yet of how a major IT consultancy is repositioning itself around artificial intelligence — and the distance still to travel.

    The French group reported revenues of €22.5 billion, up 3.4% at constant currency. Growth was modest by historical standards, weighed down by cautious enterprise spending across Continental Europe. But the fourth quarter told a different story: 10.6% growth, suggesting that client hesitancy may be starting to ease.

    From hype to realism

    CEO Aiman Ezzat framed the company’s direction as a shift “from AI hype to AI realism” — a phrase that carries weight given how frequently the consultancy sector has leaned on AI as a growth narrative without always delivering measurable outcomes.

    The numbers behind the claim are worth examining. Generative AI bookings exceeded 8% of Capgemini’s total for the year, rising above 10% in Q4. The company has trained 310,000 employees on generative AI and 194,000 on agentic AI — the emerging category of AI systems designed to take autonomous actions rather than simply generate content.

    These are significant investments in capability. Whether they translate into proportional revenue growth remains the open question. Capgemini’s 2026 guidance of 6.5% to 8.5% revenue growth fell slightly below the 7.2% analyst consensus, suggesting the market expected more from a company positioning AI at the centre of its strategy.

    The sovereignty opportunity

    Beyond AI, Capgemini is betting on a second structural trend: digital sovereignty. The company estimates that over 50% of service contracts will include sovereignty requirements by 2029, up from just 5% in 2025. For a European-headquartered firm, this represents a genuine competitive advantage over US-based rivals.

    Ezzat was notably pragmatic on the sovereignty question, dismissing calls for full European tech autonomy. “There is no such thing as absolute sovereignty,” he told journalists. “Nobody has it, because no one has sovereignty over the entire value chain required to deliver services.” Instead, he advocated for finding “the right sovereignty solution based on the use case, the client environment, the government.”

    This positions Capgemini as a bridge between European regulatory ambitions and the practical reality that most enterprise infrastructure still runs on AWS, Google Cloud, and Microsoft Azure. The company has signed partnerships with all three US hyperscalers to deliver what it calls “sovereign” AI solutions — European-managed services running on American infrastructure.

    The uncomfortable parts

    Not everything in the results paints a straightforward growth picture. Net income declined 4.2% year-on-year to €1.6 billion. The stock has fallen approximately 29% so far in 2026, caught in a broader selloff of companies perceived as vulnerable to AI-driven disruption — an ironic position for a firm making AI the centrepiece of its strategy.

    There is also the matter of Capgemini Government Solutions, the US subsidiary the company announced it would sell following public backlash over a $4.8 million contract with US Immigration and Customs Enforcement. It is a reminder that the consulting business operates in political as well as commercial environments, and reputational risk can force strategic decisions that have little to do with market fundamentals.

    France, Capgemini’s home market, contracted by 4.1% — a concern for a company headquartered in Paris. The bright spots were North America (7.3% growth), the UK and Ireland (10.5%), and Asia Pacific and Latin America (13.8%). Financial Services led sectors with 9.2% growth, accelerating sharply to 20.4% in Q4.

    What to watch

    Capgemini’s results matter beyond its own balance sheet because they reflect broader dynamics affecting the consultancy and IT services sector. The shift from selling AI as a concept to delivering AI as an operational capability is where the next wave of value — and differentiation — will come from.

    Three things are worth tracking. First, whether GenAI bookings continue to accelerate past the 10% threshold reached in Q4, or whether that was an end-of-year surge. Second, how the sovereignty proposition develops as European regulation tightens — this could become a meaningful differentiator for European-headquartered firms. Third, whether the “Fit-for-growth” restructuring programme, which will cost an additional €200 million in cash outflow, delivers the operational efficiency the company is banking on.

    The consultancy sector has spent the better part of two years talking about AI. Capgemini’s latest results suggest it is now moving into the phase of proving it can deliver.