There is a dangerous leap in today’s AI narrative: because AI will transform the economy, companies associated with AI must therefore justify extraordinary valuations.
History tells us otherwise.
The internet genuinely transformed the world, reshaping commerce, communications, media and almost every major industry. But during the dot-com boom, investors confused the revolutionary nature of the technology with the value of individual companies.
The technology was real. The revolution was real. The prices were often fantasy. AI may be heading down a remarkably similar path.
Agency isn’t intelligence
Consider the fashionable word “agentic”. An LLM doesn’t possess agency on its own. Programmers create agency by surrounding the model with software: a goal, access to tools, some memory and a loop that effectively says, “keep going until you achieve the objective.”
Suddenly, a prediction engine looks like an autonomous worker. It can write code, run commands, inspect results and try again. That is enormously useful. But it also reveals something important about AI risk.
The danger isn’t necessarily inside the model. It is in the permissions we give it. A model that answers questions is one thing. A model that can modify files, execute code, access the internet and operate without approval is something else entirely.
The difference isn’t necessarily intelligence. It’s the length of the leash.
The real revolution is productivity
This is where the AI story gets genuinely exciting.
AI is increasingly moving from assisting humans to collaborating with them – and eventually performing entire tasks autonomously. That could be economically transformative.
Imagine a small business with AI-powered accounting, research, software development, marketing and customer service. Imagine one engineer delegating routine programming to multiple AI agents. Imagine scientists testing thousands of hypotheses simultaneously.
The most important question isn’t whether AI eliminates humans from the economy. It’s whether one human can accomplish what previously required ten.
That is potentially revolutionary. And it is exactly why we should separate the technology from market valuations.
We’ve seen this movie before
The internet provides the obvious comparison.
The dot-com crash didn’t prove the internet was a failure. It proved that technological importance and investment value are different things.
After the crash, the infrastructure remained. Technology improved. Costs fell. New business models emerged. Amazon survived. Google emerged. The internet eventually became embedded in virtually everything.
The same could happen with AI.
Today’s market may be overestimating the speed and profitability of the AI revolution while simultaneously underestimating its eventual importance. That sounds contradictory. It isn’t. A technology can be underestimated in the long term and overvalued in the short term.
The AI bubble could be good for AI
If AI valuations eventually fall sharply, that doesn’t mean AI has failed. It could mean the opposite.
Capital could shift from fashionable projects toward useful ones. Weak business models could disappear. Computing costs could fall. Competition could intensify. Businesses demonstrating genuine productivity gains could survive.
That’s how technological revolutions mature.
The end railway boom didn’t make railways irrelevant. The dot-com crash didn’t kill the internet. A correction in AI wouldn’t kill artificial intelligence either. It might finally force investors to distinguish between AI as a technology and AI as a trade.
The future is less magical – and more important
The ultimate victory for AI won’t be a spectacular demonstration of a machine apparently “thinking”. It will be when nobody talks about AI anymore. It will simply be part of the economy – inside software, factories, laboratories, offices, creative tools and scientific research.
AI will become infrastructure.
That is why it is possible to be deeply optimistic about AI’s long-term potential while remaining sceptical of the financial narrative surrounding it today. AI doesn’t need today’s valuations to be justified in order to change the world.
The hype may eventually collapse. The technology won’t.
And, just as after the dot-com bubble, the companies and applications that emerge from the wreckage may prove far more important than those that dominated the headlines during the boom.