Every wave of technology predicts the same thing. The machines will replace us. They never quite do. Understanding why, and what actually happens instead, is the insight organisations need right now.
When the spreadsheet arrived, accountants were supposed to become redundant. When CAD software arrived, architects faced the same prediction. When digital cameras became universal, professional photography was supposed to collapse.
In each case, the prediction was wrong. Not for the reasons people assumed.
The technology did eliminate a large amount of work. It just was not the work that mattered most.
Accountants spent the majority of their time on manual calculation. Spreadsheets automated that. The time freed did not disappear. It moved upward, into analysis, interpretation and strategic advice that had always been the valuable part but could never be prioritised because the commodity work consumed everything.
The profession grew. The average accountant in 2000 was doing work that could not have been delivered cost-effectively in 1975, because the infrastructure to support it did not yet exist.
AI is following the same pattern, but faster, broader and at a greater scale than anything that came before.
The question is not whether it will change work. It will.
The question is this. When execution becomes cheap and widely available, what becomes more valuable as a result?
The answer is the human premium.
The task is not the job: the distinction most AI conversations are missing
The most important distinction in the current AI conversation is one that almost nobody is making clearly. A task and a job are not the same thing.
AI can increasingly perform tasks. It can draft, analyse, summarise, code and complete defined workflows with growing capability. But a job is not a single task. A job is a bundle of tasks, and the value of that job depends on which parts of the bundle can be separated out cheaply, and which parts cannot.
When the separable parts are automated, two things happen. The people in that profession have to move upward into the parts that cannot be separated. And demand for the non-separable parts often grows, because the scarce, high-value capability can now reach more people than before.
This is what happened to travel agents. It is what is happening right now across a far wider range of professions.
AI does not replace human value. It removes the commodity layer, and what remains becomes more valuable, not less.
What actually happened to travel agents: the AI workforce playbook
When online booking platforms arrived in the late 1990s, they eliminated a large number of travel agent jobs quickly. For a consumer wanting to book a return flight, there was no longer any reason to call a human. The platforms were faster, cheaper and available around the clock.
But the work those platforms automated was never the valuable part of what a travel agent did. It was simply the part that consumed most of their time.
Before online booking, a travel agent’s day was dominated by commodity transactions. Checking flight availability meant phone calls. Confirming hotels involved fax machines. A good agent might handle fifty transactions a week, the vast majority requiring no real expertise, just process and patience.
The genuinely valuable work, designing complex multi-destination itineraries, navigating group logistics, knowing which hotels had which specific qualities for which kinds of traveller, existed but could rarely be prioritised. There was simply never enough time for it when the commodity work filled every available hour.
After automation, something shifted. The agents who survived could no longer compete on the commodity transaction. That forced a choice that turned out to be commercially transformative. They specialised. They became genuine experts, in honeymoons, in luxury safari logistics, in complex family travel, in the particular requirements of high-net-worth clients who wanted every detail handled. They built direct relationships with premium hotel groups and airlines that gave clients access to upgrades and experiences no platform algorithm could offer.
They also did something that had been commercially impossible before. They charged planning fees. When the commodity booking was the core product, clients expected it to be free or very cheap. Once the commodity work was gone and only expertise remained, the model changed. Clients paid for the judgment, the relationships and the certainty that someone knowledgeable was responsible for their trip.
The result is visible in the earnings data.
In 2000, average travel agent earnings were 87 percent of the private sector average. By 2025, that figure had risen to 99 percent. US Bureau of Labor Statistics
The agents who remained earn more per hour than their predecessors ever did, because the technology stripped away the low-value work and left only the part that genuinely required expertise. The profession did not grow in headcount. But it grew dramatically in value per person.
Why radiology is a different lesson: AI unlocking scarce expertise
Travel agents show what happens when commodity work is automated and skilled professionals move upward. Radiology shows something different. What happens when AI unlocks a scarce skill and allows it to reach demand that previously could not be served at all.
Before AI entered diagnostic imaging, a specialist radiologist could review and report on between thirty and fifty complex scans per day. That was the practical ceiling. Reading a scan carefully, identifying anomalies, cross-referencing patient history and writing a clinical report takes concentrated time. There was no shortcut.
The consequence was not just slow processing in well-resourced hospitals. It was enormous unmet demand across the entire system. Rural hospitals in developed countries often had no on-site radiologist. Patients in many places waited weeks for results that could have directed treatment in days. In developing countries, imaging infrastructure existed without the clinical expertise to interpret what it produced. Millions of scans went unreviewed, or were reviewed by generalists who lacked the specialist training to catch what a radiologist would catch.
The bottleneck was not a shortage of radiologists in absolute terms. It was the number of scans any one radiologist could process in a working day.
AI changed that equation. It could pre-process images, perform initial pattern recognition, flag the cases most likely to require urgent attention and filter the straightforward from the complex. This did not replace what radiologists do. It changed the ratio of their work. Instead of spending equal time on every scan, a radiologist working with AI could focus their expertise where it was most needed. On the ambiguous cases. The complex presentations. The situations where specialist judgment was the difference between a correct and an incorrect diagnosis.
When that bottleneck was removed, all the latent demand became actual demand. The cases that had gone unreviewed. The hospitals that had lacked access. The patients who had waited too long. New subspecialties emerged. Teleradiology became viable at scale, allowing specialists to work across hospitals and clinics in locations where they could never physically have been present.
The profession did not shrink. It grew, because the AI did not take the radiologist’s work. It multiplied the number of patients the radiologist’s work could reach.
More than half of current work hours could be automated with today’s proven AI. Yet professions built on scarce human expertise consistently grow rather than shrink when AI enters them. McKinsey Global Institute, 2025
Both stories, travel agents and radiology, point to the same underlying dynamic. When technology automates the commodity layer of a profession, human value does not diminish. It concentrates. The remaining work is harder to do, more genuinely valuable, and in most cases in greater demand than the commodity work that preceded it.
What this means for the AI transition in your organisation
The pattern from travel agents, radiology, accountants and architects does not guarantee that every profession survives AI intact. Some jobs are more commodity than expertise, and those are genuinely at risk. But the lesson for most organisations is more nuanced than the standard narrative suggests.
The question worth asking for every role is not: can AI do this task? Increasingly, the answer will be yes.
The better question is: which parts of this job bundle cannot be cheaply separated from the human delivering them?
Those are the parts that become more valuable. Those are the capabilities to invest in.
There are ten specific categories of human contribution that consistently resist automation, not because AI lacks the raw capability in theory, but because their value is inseparable from human delivery. Understanding them in detail changes how organisations think about hiring, development and operating model design.
We examine all ten in the next article in this series.
AI is not making people less important. It is concentrating value in the human capabilities that machines cannot replicate.
Talk to Inovus
At Inovus, we help organisations unlearn pre-AI ways of working and deliver measurable outcomes through AI-first delivery. That means identifying where AI creates real value, building the human capability around it that makes the value sustainable, and designing operating models that put the right people in the right roles for the next decade rather than the last one.
If you are working through what this means for your teams or your operating model, we are happy to talk.
About the author
Filippo Di Pisa is CTO at Inovus, a data and AI consultancy backed by La Fosse Group, helping organisations unlearn pre-AI ways of working and deliver measurable outcomes through AI-first delivery.
Sources
McKinsey Global Institute, Agents, Robots, and Us: Skill Partnerships in the Age of AI, 2025.
US Bureau of Labor Statistics, Travel Agents Occupational Outlook Handbook and wage series.
Journal of the American College of Radiology, Artificial Intelligence and Its Impact on Radiology, 2024.
Boston Consulting Group, AI at Work: The People Imperative, 2024.