I was forwarded today a pretty underwhelming essay comparing AI to the surprimes crisis. The central argument was that millions of jobs will be lost when the AI “bubble” inevitably bursts, as unreasonable amounts have been poured into the sector over the last 18 months. That stupid sums of money were thrown at anything with an AI label on it is undeniable, but calling this a “bubble” is debatable. Investors may never recoup their money, but AI is already creating tangible value across industries, reshaping how we work and boosting productivity in unprecedented ways : layoffs in the Tech sector are not a bear signal ; they are a proof of concept.
The only valid parallel between subprimes and AI is that customers do not really understand what they buy, and it’s becoming increasingly challenging as AI solutions evolve from relatively simple models to highly complex systems. The asymmetry of information between providers and consumers has led to scandals in the past. And we can expect more.
However, this is where the similarity ends. AI has fundamentally changed how people work, particularly in software development and content creation. Independent developers, hobbyists, and content creators have embraced AI tools that give them super powers. Many report that they "can't remember how working was like before generative AI."
This impact isn't limited to individuals or small-scale businesses. Major corporations are also reaping the benefits of AI integration, as highlighted by Amazon's CEO in a recent post (see below).
The quick adoption of AI tools doesn't seem reversible. As the saying goes, "the toothpaste is out of the tube." We're now faced with the question of what comes next. In sectors where AI has already made significant inroads, we can expect adoption to continue accelerating. Moreover, there are entire markets and slower-moving industries that are virtually untapped.
You’d be right to point out that jobs will be lost, in particular in Tech, if AI contributes so much to productivity. But let’s not be luddites: today, the sector is plagued with parasites disguised as engineers and consultants. I should know because I was their target during my career in Finance. Getting rid of them would have a net-positive effect for society.
There will always be a demand for developers, maybe more than before, but the fact barriers to entry are materially lower will put the onus on customers (better UX, customization…) instead of shareholders (maximization of operational leverage, controversial Terms of Service…). All this would be positive. The success of AI code assistants, as tools created by developers for developers, evidences that this transition has already started. The model of "insider" tool development is likely to spread to other domains, potentially bringing product and development functions closer together. As AI assistants become more prevalent across various domains, we can expect significant improvements in overall user experience and product functionalities.
About a bubble
Last week, I had a chat with an MLOps consultant. He didn’t know, ahead of our conversation, that I was reasonably in the details when it comes to his field so he talked to me like he would talk to any outsider. Overall, I was surprised by his outdated information and factually incorrect statements.
Unlike my terrible experiences with McKinsey and Gartner consultants last year, I couldn’t really blame him for two reasons :
- he was really working on MLOps projects and the intense, time-consuming nature of these projects leaves little bandwidth to keep track of the latest innovations.
- the AI field is evolving so fast that solutions are outdated before they're even launched.
He told me about Llama-2 and Mistral-7B. He told me about RAG with CamemBERT. Even if he had talked about the state-of-the-art in March 2024, it would still be outdated.
If your solution is just an API call with a simple prompt, you might be able to switch easily. But any complex, custom AI solutions would require complete rebuilds to leverage new advancements. Even compared to three months ago, you’d think differently today about semantic search, data storage, or the unit economics of AI solutions.
This rapid turnover of technology explains why so much money will be burnt on AI, contributing to the perception of a "bubble": companies feel compelled to "invest" in AI because they can’t be perceived as old firms, anchored in the past, and refusing to ride the wave that will make all their competitors more efficient.
Yet everything they do today will have to be done again in 18 months. OpenAI’s latest model, o1, perfectly illustrates a paradigm shift that may require rebuilding everything from the ground up : inference-time scaling is compelling but hardly compatible with existing pipelines. The only missing piece now is a form of routing for simple queries to make it cheaper. Then everyone could transition.
When an organization “invests” in AI today, they're not acquiring static assets. Instead, they're investing in their workforce so they are better prepared for the next cycle. These investments may not translate directly into assets on a balance sheet, not even goodwill, hence the perception of a “bubble”. Yet they create significant intangible value.
As counterintuitive as it might be, the value of AI deployed today lies not in the technology itself, but in the people working with it. The most valuable employees in this new landscape are not those who simply apply AI tools as instructed, but those who try to understand why they do it, why it works and what’s missing when it doesn’t.
The rapid evolution of AI technology may give the appearance of a bubble, with constant reinvestment and quick obsolescence of solutions. However, this view misses the transformative nature of AI and the real value it creates. By investing in AI, companies are not just chasing a trend - they're building the capabilities, knowledge, and human capital necessary to thrive in the future. The true measure of success in this new era will be an organization's ability to learn, adapt, and innovate at a pace rarely seen before. Bureaucracies will have a hard time.