If I were starting my career today, I suspect I would be asking a question that many people are quietly wrestling with right now.
How seriously should I take AI?
The answers seem to be everywhere. Some people speak about AI as though the future has already been decided and every profession is about to be rewritten. Others dismiss it as another technology bubble driven by venture capital, hype cycles, and unrealistic expectations. Many people are trying to occupy a position somewhere in the middle, careful not to sound overly optimistic or unnecessarily skeptical because nobody wants to be remembered as the person who got it completely wrong.
The more I thought about the question, the more I realized that I was looking for answers in the wrong place. Most conversations about AI focus on predicting what will happen next. History suggests prediction is rarely where the useful lessons come from. People are notoriously bad at forecasting the future, especially when a new technology begins changing how industries operate. Looking backward has often proven more useful than trying to look forward.
Every generation encounters a technology that forces professionals to make a decision before the outcome becomes obvious. Some dismiss it. Some resist it. Some become evangelists. A smaller group decides to learn enough about it to understand where it helps, where it fails, and how it might change their craft.
The names of the technologies change. The decision rarely does.
Long before software engineers debated AI models, workers were debating mechanization during the Industrial Revolution. The transition was not simply a story about machines replacing people. Entire industries were reorganizing around new capabilities. Skilled workers who had built careers around existing methods suddenly found themselves confronting tools that could produce more output at lower cost. Some resisted. Some adapted. Some attempted to ignore the change altogether.
Looking back, what stands out is not that everyone who embraced mechanization won or that everyone who resisted lost. Reality was more complicated than that. What stands out is that the people who took the time to understand the changing landscape generally found themselves in a stronger position than those who spent their energy arguing that the landscape should not change in the first place.
The same pattern appeared when computers entered the workplace.
Today it feels absurd to imagine an office operating without computers, but there was a time when that future was far from obvious. Typewriters worked. Filing cabinets worked. Existing workflows were familiar. Learning to use a computer required time, effort, and a willingness to abandon habits that had served people well for years.
What is interesting is that the people who learned computers did not need certainty about the future. They did not need to know that computers would eventually dominate every office. They only needed to recognize that understanding the new capability created options. If computers became important, they would be prepared. If they did not, the skills acquired along the way would still be valuable.
That idea keeps appearing throughout history.
The internet provides perhaps the most useful example because it highlights something many people still struggle to separate: the difference between a technology and the excitement surrounding it.
The late 1990s were filled with predictions about the internet. Some were visionary. Many were ridiculous. Investors poured money into companies with weak business models and little evidence that they could ever generate sustainable profits. Valuations climbed to extraordinary levels. Optimism became speculation and speculation eventually became a bubble.
What happened next? The bubble burst, companies disappeared, investors lost fortunes, predictions failed. Yet something interesting happened after the collapse. The internet continued spreading through society. Online commerce continued growing. Digital communication continued expanding. Search engines became essential. Entire industries emerged on top of infrastructure that was still in its infancy during the dot-com era.
The people who concluded that the internet was useless because internet stocks crashed misunderstood what was happening. The speculative layer collapsed. The underlying capability remained.
I think many conversations about AI make a similar mistake.
People often ask whether AI is a bubble as though answering that question automatically tells us whether learning AI is worthwhile. History suggests those are different questions. Railroads experienced speculative bubbles. Telecommunications experienced speculative bubbles. The internet experienced one of the most famous bubbles in modern history. In each case, investors and markets often overestimated the short-term impact while correctly identifying a capability that would matter enormously over the long term.
Even if parts of the AI ecosystem are experiencing excessive hype today, it does not automatically follow that the underlying capabilities lack value. The more useful question is whether the technology changes how work gets done. Everything I have seen so far suggests that it does.
Software engineering offers its own collection of examples.
The shift from waterfall methodologies to agile development was not simply a change in project management. It represented a different philosophy about uncertainty. Traditional planning-heavy approaches assumed uncertainty could be reduced primarily through analysis and prediction. Agile introduced the idea that uncertainty could often be reduced more effectively through shorter feedback loops and continuous learning. Many organizations resisted the shift. Others embraced it without understanding it. Over time, the industry gradually settled into a more balanced understanding of where those ideas worked and where they did not.
Cloud computing followed a similar path. There was a time when many organizations questioned whether serious businesses would ever trust infrastructure they did not physically own. The concerns were not irrational. Security, compliance, and operational control were legitimate considerations. Yet the economics of cloud computing eventually became too powerful to ignore. The engineers who invested time learning cloud technologies before they became mandatory accumulated advantages that compounded throughout their careers.
The same thing happened with open-source software. It happened with mobile computing. It happened with distributed systems. It happened with containers. It happened with microservices, although that journey included a healthy reminder that over-adoption can be just as dangerous as resistance. Every technological transition attracts people who underestimate the change and people who overestimate it. The people who consistently seem to do well are the ones who spend enough time understanding the technology to make informed decisions instead of ideological ones.
The events I have had to revisit over the past months have shaped how I think about AI.
I do not know whether current expectations are too high or too low. I do not know which companies will dominate five years from now. I do not know which products will become indispensable and which will disappear. I do not know whether the most important AI companies of the next decade have even been founded yet.
What I do know is that learning has historically been one of the lowest-regret responses to technological change.
Suppose a young professional spends the next two years learning how AI systems work, where they succeed, where they fail, and how they can be incorporated into existing workflows. If AI ultimately falls short of today’s expectations, that person still develops a deeper understanding of automation, workflow design, experimentation, systems thinking, and emerging technologies. Those skills remain useful even if the technology evolves differently than expected. In another case, if AI exceeds expectations, the benefits become even more obvious.
The asymmetry is difficult to ignore. The downside of learning appears relatively small while the potential upside can be substantial.
What makes this particularly interesting is that the most valuable skills may not be the ones many people expect. Every major technological shift changes the economics of something. When production becomes cheaper, distribution becomes more important. When information becomes abundant, attention becomes more valuable. When infrastructure becomes easier to provision, execution speed becomes a competitive advantage.
AI appears to be reducing the cost of generation across many forms of knowledge work. As generation becomes cheaper, other capabilities naturally increase in importance. Judgment becomes more important. Validation becomes more important. Context becomes more important. Decision-making becomes more important. Understanding whether an answer is correct often becomes more valuable than producing an answer in the first place.
For me, the deeper question is how professionals adapt when a capability becomes dramatically cheaper than it was before.
After looking at enough technological transitions, I keep returning to the same conclusion. Technology changes. Industries change. Tools change. Predictions change. Human behavior changes far less than we often imagine.
Every generation eventually encounters a technology that forces a choice between skepticism, resistance, evangelism, and learning.
History has not consistently rewarded the skeptics nor the evangelists. It has been remarkably consistent in rewarding the learners.
That is why learning remains my safest bet.