More Efficiency, More Demand
The future of software engineers and data scientists is bright
According to the American Journal of Roentgenology there were 30,723 radiologists in 2014 in the US. At that time, before LLMs even, there were dozens of startups leveraging AI to read x-rays and images. Geoffrey Hinton, the Nobel Prize winning computer scientist who is often referred to as the “Godfather of AI”, famously asserted in 2016 that, “People should stop training radiologists now. It’s just completely obvious that in five years deep learning is going to do better than radiologists.” The logical expectation would be that the number of radiologists should begin to decline over time as they begin to get replaced by AI. People who went into that field would have a hard time getting jobs, they might have to be retrained, they certainly wouldn’t recommend it as a career to medical school students.
Contrary to what most people expected, by 2023 the number of radiologists in the US had grown to 36,024. There are even studies that project a 25.7% increase in the radiologist workforce by 2055 to 47,119. How could this possibly be happening? Wasn’t AI taking over these jobs?
To answer this we need to look back to a 19th century English economist named William Stanley Jevons. The eponymous Jevons Paradox is the economic principle that increased efficiency in resource use can lead to an overall increase in total resource consumption, rather than a decrease. Jevons noticed that improved steam engine efficiency resulted in higher coal consumption rather than less. He argued that as the efficiency of steam engines increased, their usage expanded, offsetting any gains made in energy conservation.
Jevons noticed something the rest of the world kept missing. When technology makes a task cheaper or easier, we don’t do less of it. We do more. The easier it becomes to extract value from a resource, the more we find reasons to use it. Coal was not conserved by better engines; it became the fuel of an expanding industrial world. This logic extends far beyond steam. Efficiency unlocks demand, and demand expands the ecosystem. That’s the real paradox.
AI in radiology followed exactly this pattern. As early machine learning models began offering preliminary reads or triage support, hospitals didn’t cut staff. They increased throughput. They ordered more scans. They screened patients who previously would not have been screened at all. This avalanche of demand didn’t grow because radiologists got worse at their jobs but because they got better tools. Faster preliminary reads made imaging cheaper for hospitals to perform. Automated triage made it easier to prioritize urgent cases. Clinics began scanning patients earlier in a diagnosis to catch issues sooner, often expanding the clinical pathways that relied on imaging. Even insurers found new reasons to approve imaging when early detection lowered long-term costs.
Efficiency did not erase the radiologist. It multiplied the radiologist’s workload.
And something else happened, something even more counterintuitive. As machines handled simpler, repetitive tasks, the complexity of the remaining human work actually increased. When a model flagged an anomaly, it was often the ambiguous cases that required a trained specialist to interpret. When an automated system produced false positives, someone had to understand why. When a patient’s case spanned multiple modalities or involved rare conditions, no algorithm could navigate the nuance. Radiologists became less like image readers and more like detectives, synthesizing more information, integrating clinical context, and handling the exact edge cases that automation struggled with.
In other words, automation changed the work but did not reduce its importance.
The underlying driver is bigger than radiology. It is the economic truth that when you make a knowledge-intensive task easier, the world finds new tasks that build upon that work. Instead of shutting down opportunity, efficiency creates room for more of it.
Jevons couldn’t have predicted neural networks, but he understood human behavior well enough that his insights echo across every technological revolution since. When a resource becomes more efficient, its value rises, and rising value drives higher consumption. This principle applies as much to compute cycles today as it did to coal in the 1800s.
The cost of inference drops, and suddenly millions of images can be scanned automatically for early warning signs. The cost of generating features or labels drops, and suddenly new datasets can be constructed for problems that had never been economically viable to explore. The cost of building a prototype drops, and suddenly teams can attempt ideas that previously would have died in a backlog.
Once efficiency improves, the frontier expands.
AI lowered the cost of interpreting medical images, so the world ordered more images. But this is just one example of a broader pattern playing out everywhere. Automated underwriting did not lead to fewer financial analysts. Document review software did not lead to fewer lawyers. Logistics AI did not lead to fewer supply-chain specialists. Each wave of efficiency produced a larger market, not a smaller one, because when a field becomes more capable, its relevance grows.
We want more of what becomes powerful.
That is the future awaiting software engineers and data scientists. And to understand it, radiology is just the opening act.
If you zoom in on what actually changed in radiology during the past decade, you don’t see replacement. You see role shift.
The first shift was diagnostic breadth. When AI lowered the cost of initial analysis, clinics began testing earlier in the diagnostic chain. People who once would have gone straight to treatment plans now received imaging first. Preventative scans expanded across specialties. Orthopedic practices increased their use of imaging to track injury progression. Emergency rooms used triage algorithms to prioritize patients more effectively, and those triage systems demanded radiologists to make final calls.
The second shift was workflow complexity. AI didn’t simplify these systems. It layered new tasks on top of old ones. Models needed monitoring. Outputs needed validation. Pipelines needed tuning. Hospitals now needed radiologists who understood how to interpret automated flags or reconcile conflicting results across models. The cognitive overhead didn’t go down; it went up.
The third and most important shift was the expansion of the domain itself. Radiology used to be a bottleneck. AI widened the pipe. When you widen a pipe, you don’t just move existing volume faster; you increase total volume. Physicians who once avoided ordering certain scans because of backlog constraints suddenly had freedom to request them. Researchers could mine imaging datasets at scales impossible before. Medical schools expanded their imaging curricula, not reduced them.
The field didn’t shrink because the work did not disappear. The field grew because the work increased.
The same dynamic is unfolding in software, but with far larger stakes.
In every domain where automation becomes powerful, the pattern remains consistent. Human expertise becomes more valuable because the total volume of meaningful work increases. Early fears of automation nearly always assume a fixed amount of work being redistributed. But work is not fixed. Work expands when constraints are removed.
Look at manufacturing. Robotics lowered the cost of producing goods, but global manufacturing did not contract. It exploded. Factories shifted from manual assembly to oversight, programming, maintenance, and process optimization. Entire industries emerged around robot design, integration, and simulation. The total number of manufacturing workers shrank in some regions but the total number of manufacturing roles skyrocketed worldwide because demand soared for what could now be produced.
Finance tells the same story. Spreadsheet software didn’t eliminate accountants. It expanded the types of financial analysis companies could perform. Automated trading didn’t eliminate traders; it reshaped their skill sets and created entirely new fields in quant research, risk modeling, and algorithmic oversight.
Even programming tools themselves follow the pattern. The rise of high-level languages didn’t eliminate developers, it created more developers. The rise of cloud platforms didn’t reduce the need for operators, it created demand for DevOps, SRE, and platform engineering. Every layer of abstraction generates a new layer of complexity to manage. Every gain in simplicity spawns a new frontier of problems worth solving.
Efficiency doesn’t collapse a field. It deepens it.
And now AI is the newest, brightest efficiency engine ever built. The demand ripple it creates will dwarf what came before it.
Today, AI-assisted coding tools can generate boilerplate, translate languages, create tests, scaffold services, and even design early prototypes. This has triggered a familiar anxiety: if one engineer can now do the work of five, won’t companies simply need fewer engineers?
No. Companies will want to build more.
Lowering the cost of creation does not reduce the need for creators. It increases the demand for creation.
The effect is easy to see. Once you can build prototypes in hours instead of weeks, you don’t build fewer prototypes; you build ten. Once you can test three product directions at the cost of one, you don’t choose one; you explore all three. Once you can create software with smaller teams, entire industries that previously resisted digital transformation begin treating software as a core competency.
This is why the future for engineers and data scientists is not shrinking. It is expanding. The frontier of what is buildable will grow faster than the efficiency of any tool. Ambition scales to fill the available capability. Companies will ship more frequently. Iterate more aggressively. Explore more ideas. Use software in parts of their operations that once seemed too expensive or too specialized.
Engineers become amplifiers of organizational imagination.
Data scientists become orchestrators of automated insight.
The total quantity of problems worth solving rises when solving them becomes easier.
Efficiency creates abundance. Abundance creates demand. Demand creates new ecosystems that require more people, more roles, and more expertise than before. This is the engine that powers technological revolutions. It is not a smooth transition but a compounding one.
Lowering the cost of software development expands the universe of companies that can build products. Startups that once would have needed fifteen engineers can begin with two. Enterprises that once outsourced their systems can build internal products with small autonomous teams. Governments historically slow to adopt technology can accelerate modernization. Nonprofits can deploy tools that previously required massive budget allocations.
The total amount of software in the world grows faster than the efficiency of producing software.
And every new line of code, especially the ones written with AI, requires engineers to maintain it, integrate it, secure it, observe it, and govern it. The more surface area the digital world acquires, the more talent the digital world demands.
This is not a fantasy future. It is the only future the economics of efficiency allow.
Of course, the work itself will change. But this shift mirrors what happened in radiology: a reconfiguration, not a reduction.
For engineers, the lowest layers of development will become handled by tools. But the higher layers, architecture, integration, security, compliance, reliability, performance tuning, edge-case reasoning, will become more central. When code becomes easy to generate, systems become harder to reason about. When development accelerates, complexity compounds.
The engineers of the future will spend less time typing and more time thinking. More time designing than debugging. More time understanding the domain than wrestling with syntax. AI will help assemble the building blocks, but someone still needs to decide what the building is, how it should behave, and why it matters.
For data scientists, the shift is equally dramatic. Automated feature engineering, model selection, and tuning don’t eliminate the need for expertise. They elevate it. Data scientists become the stewards of model behavior, the interpreters of causality, the designers of experiments, and the boundary-setters for what the system is allowed to infer. When models generate answers cheaply, the real value shifts to asking the right questions.
The shape of the work changes, but the importance of the work grows.
Every technological leap follows a similar emotional arc. First comes fear: machines will take our jobs. Then comes surprise: the field grows more than it shrinks. Finally comes acceptance: the machines did not replace us; they augmented us.
The core misunderstanding lies in the assumption that human labor and machine labor are substitutes rather than complements. In most complex fields, the opposite is true. Machines handle tasks humans find tedious or repetitive, freeing humans to handle tasks machines cannot comprehend. This is why radiologists became more indispensable after automation, not less. It is why accountants, lawyers, and financial analysts saw their fields grow rather than collapse. It is why programmers are more in demand today than ever, despite decades of predictions about abstraction and code generation.
People fear that AI will take away the work they know. They rarely see the work that will emerge because the shape of future demand is invisible to them until efficiency makes it viable.
But once the frontier expands, the opportunities reveal themselves.
The future awaiting software engineers and data scientists is not one of scarcity. It is one of abundance. More products, more systems, more models, more capability, more complexity, more entropy, more ambition. Every efficiency expansion creates new edges to explore. Every leap in tooling creates new fields of study. Every simplification creates new layers of nuance.
This is not a surge that will last a few years. This is a restructuring of what it means to build things.
AI will reshape the profession, but only in the sense that cars reshaped transportation or spreadsheets reshaped finance. Not by eliminating the field, but by expanding its scope. Not by reducing labor, but by elevating it. Not by shrinking opportunity, but by multiplying it.
The world does not need fewer people who understand systems. It needs far more of them.
It is tempting to assume that the future holds less work because tools are getting better. But this intuition runs against both history and economics. Better tools don’t compress human ambition. They inflate it. As tasks become easier, we don’t do fewer of them. We broaden the landscape until it feels hard again.
In radiology, greater efficiency created greater demand. The same is about to happen to software and data science, only on a much larger scale. AI will accelerate the craft, but acceleration does not eliminate the need for craftsmen. It enlarges the world they operate in.
The Jevons Paradox remains a paradox only until you realize it is not really about coal or engines or imaging. It is about the way humans respond to possibility. When a resource becomes powerful, we use more of it, not less. When a tool becomes capable, we rely on it more deeply. And when technology expands what is possible, we expand right along with it.
The future belongs to the people who can wield these tools, shape them, question them, and integrate them into systems that matter. Not because AI needs them, but because the world built on AI will.
The future of software engineers and data scientists is not diminishing. It is compounding.