Executive Summary
- 60% of companies are delegating AI as a technology adoption problem — and handing it to the wrong person
- AI is a general-purpose technology: the first executives alive have ever had to navigate one at this speed and scale
- The J-curve is real — running parallel processes costs more before it pays off, and skipping it is dangerous
- The steam engine analogy reveals exactly why bolting AI onto existing processes produces marginal gains, not transformation
- Contextual intelligence and retention are becoming the primary determinants of competitive advantage
- Mid-sized, owner-operated businesses may be better placed than large corporates to absorb the J-curve and move faster
- Executives must set aside dedicated time, build age-diverse teams, and engage with AI personally — not just delegate it
The 60% Problem
When Harvard Business School Professor Joseph Fuller describes the current state of corporate AI adoption, he does not reach for optimism. In a recent conversation on the Arkaro Insights podcast, he offered a clear-eyed assessment: around 60% of companies are still treating AI as a technology adoption challenge — as if it were, in his words, “yet another glorified SaaS tool that can just be bolted on the side of an existing management process.”
The consequences of that framing are significant. Companies launching AI experiments without clean, well-tagged data and trained staff get disappointing results — and then misread those results as evidence that AI is not relevant to their business, rather than as evidence of a poorly designed experiment. The technology gets blamed. The management failure goes unexamined.
Fuller’s research reinforces this uncomfortable picture. Active utilisation of AI trails off sharply around the 80th percentile of organisational seniority. Senior executives who insist they have a clear AI strategy often cannot, when pressed, describe what they are actually doing with it. As Fuller puts it, the danger is not disingenuousness — it is that they genuinely believe they know what they are talking about.
The statistic that two Arkaro Insights guests have independently cited captures the gap precisely: 80% of CEOs say AI is core to their strategy, but only 15% of employees believe them.
A General-Purpose Technology — and Why That Changes Everything
Fuller’s central argument is that AI is a general-purpose technology — and that no executive alive has ever had to navigate the implementation of one before. The closest comparison is the arrival of alternating current in the late 1870s and 1880s, when George Westinghouse’s electrical infrastructure began to replace steam power in industrial facilities across the United States.
The parallel is instructive in ways that go beyond the obvious. When factories first adopted electricity, most of them made the same mistake companies are making with AI today. They bolted the new power source — the dynamo — onto their existing steam-driven infrastructure and captured modest efficiency gains. The real transformation only became visible when clever engineers realised that electrical current, unlike steam, does not care about verticality. It follows wires. It can go up, down, sideways, around corners. Steam had to rise, which is why the great textile and shoe mills of 19th-century Massachusetts were built four and five storeys high — to exploit the thermodynamics of low-pressure boilers. Twenty percent of every factory floor was occupied by manual elevators and work-in-process inventory, simply moving things up and down.
Electricity made that constraint disappear. The single-storey factory was not an incremental improvement on what came before. It was a structural redesign made possible by a new technology — and it required the willingness to abandon a perfectly functional, heavily capitalised existing asset. Many mill owners were not willing to make that investment. The shoe and textile industries migrated south, and eventually overseas. The mills became brewpubs.
The lesson for today’s executives is direct: AI will not transform your business if it is deployed to optimise processes designed for a different era. The question is not how to add AI to what you already do. The question is what you would design if you were starting from scratch.
The J-Curve: Why Transformation Costs More Before It Pays Off
Fuller’s concept of the J-curve describes what happens when companies pursue genuine AI transformation rather than incremental optimisation. The right approach, in his view, is to identify one, two, or three “main sequence” processes — those genuinely important to competitive advantage in your industry — and design a parallel, AI-native version of each from the ground up.
Running both processes simultaneously incurs double costs in the short term. The economics of adoption turn temporarily negative. The J-curve dips before it rises. But once the new process is stable and proven, and the old one can be shut down, the margin improvement can be substantial.
The alternative — flash-cutting from the existing process to an unproven one — is dangerous. And the lazy version — bolting AI onto the existing process to show the board something at the strategy off-site — produces the marginal gains of the mill owner who wired a dynamo to a five-storey building and declared the electricity problem solved.
The pressure to demonstrate short-term ROI makes this harder. Fuller describes a pattern he encounters regularly: an executive is charged with implementing AI and told to show a 15% return in year one, with the first instantiation launched within six months. The signals this sends are unambiguous — find the easy win, demonstrate activity, avoid the uncomfortable restructuring. This is not a transformational AI strategy. It is its opposite.
Who Can Stand the J-Curve?
The capacity to absorb the J-curve is not evenly distributed. Fuller draws a direct connection between ownership structure and the ability to pursue genuine transformation. Public companies, managed quarter to quarter, are structurally disadvantaged when it comes to multi-year process redesign. Private companies — particularly those that are family-owned or owner-operated — have a meaningful advantage: longer horizons, closer decision-making, and less pressure to manage perception rather than performance.
Fuller points to Koch Industries, H-E-B, and Cargill as examples of organisations that manage for positive operating cash without the anxiety of period-to-period income reporting. This gives them the patience that genuine AI transformation requires. It is a counterintuitive finding — that the largest and most resourced organisations may not be the winners of the AI transition — but it follows directly from the structural logic of the J-curve.
This argument connects closely to the themes explored in the Arkaro Insights conversation with Eric Ries, whose forthcoming book Incorruptible challenges the short-termism built into conventional capitalism — and what it means for organisations trying to invest for the long term.
The Geometry of Organisations — and the Coming Structural Shift
Fuller’s forthcoming research, developed in collaboration with Accenture Research, maps the structural geometry of six large industries, including healthcare delivery, banking, and software platforms. Most executives assume their organisations are pyramidal in shape. The research shows this is rarely true. What emerges instead are irregular block-tower structures, with disproportionately large layers at unexpected levels of the hierarchy.
When AI exposure is overlaid on these structures — mapping which roles are candidates for automation or augmentation — the shapes undergo dramatic transformation. The implications are significant. Many white-collar entry-level positions are highly exposed to being automated away. If those roles disappear, where does the management pipeline of the future come from? You cannot have someone with ten years of experience if you did not hire them nine and a half years ago.
At the same time, training budgets in most US companies have fallen by more than 50% over 25 years. What remains is largely compliance training. Companies that have relied on experiential, on-the-job learning are facing a structural problem: the half-life of many technologies being deployed is now shorter than the time it takes a worker to master them through experience alone. The on-the-job model is breaking down.
Contextual Intelligence — the New Competitive Moat
As AI takes on more routine cognitive work, what remains distinctively human is contextual intelligence: the accumulated, tacit understanding of how things actually work in a specific industry, market, organisation, or customer relationship. Your supplier in Spain takes holidays in July. Your contact at the Danish distributor prefers to close deals before June. None of this is in a system. It lives in people — and it is acquired through years of proximity.
Fuller’s argument is that retention becomes significantly more important in an AI-augmented world, not less. If the competitive advantage of your people lies in experientially derived contextual intelligence, a 70% annual turnover rate is not a human resources problem — it is a competitive strategy problem. The churn-and-burn personnel logic that characterises many low-wage sectors is directly incompatible with building the kind of organisational capability that AI transformation requires
What Should the CEO Actually Do?
Fuller’s prescription for the individual executive is both honest and practical. He uses the analogy of a cardiologist delivering unwelcome news: the lifestyle is in the process of killing you, the pains in your chest are not from sitting at your desk, and there will be no shortage of excuses for not changing. The market will eventually remove the excuses by making you unemployed.
The practical steps he recommends are these. First, set aside dedicated time — he commits to approximately four hours a week — to learn and experiment with AI personally. Not to delegate it. Not to read reports about it. To use it. Second, build age-diverse teams around AI initiatives. The standard pattern — appoint a team of directors aged 30 to 40, knowledgeable and promotable — produces a group with common title, salary, and experience, but none of the diversity of perspective that genuine AI problem-solving requires. The more productive model puts a 25-year-old new hire in the same team as a 62-year-old veteran district manager, deliberately blending technical competence with contextual knowledge.
Fuller’s illustration of what this looks like in practice is telling. Working with a CEO preparing for a quarterly analyst call, he helped design a process that harvested all historical analyst questions, synthesised themes by analyst, built synthetic analyst personas, generated likely questions based on forthcoming data releases, drafted answers, and then had AI evaluate those answers against answers given by competitors. None of this required a technical background. It required curiosity, a willingness to experiment, and a few hours of engagement.
Speaker Bio
Joseph Fuller is Professor of Management Practice at Harvard Business School and co-head of the Managing the Future of Work project, which he founded. A former CEO of global strategy firm Monitor Group, he advises leading organisations on AI adoption, workforce transformation, and the future of organisational design. His research is published through the Harvard Business School project site and through the American Enterprise Institute, where he is a fellow. The Managing the Future of Work podcast, which he co-hosts, has published approximately 300 episodes and is the largest future-of-work podcast in the world.
Connect with Joseph Fuller
LinkedIn | Twitter/X: @JosephBFuller
Managing the Future of Work — Harvard Business School
Project | Newsletter | Podcast | Apple Podcasts | Spotify | Amazon Podcasts | LinkedIn Page
Related Reading from Arkaro Insights
On AI adoption and transformation
Why AI Transformation Fails — Charlene Li on Leaders and the 90-Day Blueprint
Niels van Hove on Human-AI Collaboration in Supply Chain Planning
Rewire or Retire — Marco Ryan on AI Leadership and Digital Curiosity
AI and the Octopus Organisation — Stephen Wunker on Adaptive Structures
Why 95% of AI Implementations Fail — Neuroscience and the PEOPLE Framework
AI Implementation Failure — Barry Eustance on People-Centric Change
AI Strategy Implementation — Why Human Collaboration Matters
The Human Factor in AI Implementation — Lessons from Aviation
On organisational design and change
Strategy is an Adaptive Challenge, Not a Technical Problem
Emergent Strategy — Why Five-Year Plans Fail in a World of AI
About Arkaro Insights
Arkaro Insights brings leading thinkers and practitioners into conversation with B2B executives navigating complexity, change, and the limits of conventional management tools. As Willie Pietersen of Columbia Business School has observed: “There are not enough tools to learn how to be successful in a complex world.” That is the problem this podcast exists to address.
📺 YouTube: www.youtube.com/@arkaro
🎧 Audio: https://arkaroinsights.buzzsprout.com/
Are you navigating the challenge of AI transformation in your organisation? Arkaro works with B2B leaders in agriculture, food, and chemicals to translate strategic intent into practical, sustainable change. We do not just coach — we get on the pitch with you.
