The cart, the horse, and what it means for you
The expression “the cart before the horse” provides a useful framework for examining the current state of artificial intelligence (AI) development. Currently, those developing frontier models are trying to build a tool that is both impactful (in terms of automation and economic stimulus) and profitable for their business. In my view, using the above idiom, the horse (or the first thing companies should be concerned with is impact). Followed by the cart, which I view as profitability. Unfortunately, it seems many of these companies are shifting gears away from developing better, more impactful models and are instead looking for a quick buck.
The horse
The AI community has been using the term “artificial jagged intelligence” to describe the current capability of these models. The jaggedness reflects that, although these tools are very powerful in some respects, they are not all-encompassing yet and cannot do everything as well as, or even at par with, what human beings can. One just has to look at these models struggling to count the number of “r”s in strawberry or see one too many hallucinations before it becomes apparent that these models are far from being super useful or disruptive.
Cal Newport recently released a video outlining essentially six things that generative models (e.g., ChatGPT, Gemini, Claude) can do very well. He mentioned that programming will not be considered — despite the models being very good for this task — because it is still a niche task that primarily affects computer programmers. Also, AI use cases that are not generative and have been in development for a long time are not considered, as they were developed before the large influx of capital into this sector. The six things generative models are really good at are:
Pattern recognition
Presentation creation (e.g., creating a PowerPoint)
Summarizing
Customer support agents
Writing boring texts that do not require creativity
Making stuff more digestible (e.g., explain this very complicated topic to me like I am five)
That’s it. That’s essentially all the use cases that are available right now from these generative AI models. This is far from what we’ve been promised; we have not been given the crazy productivity enhancements, and the labour market does not appear to be affected. Some companies have credited layoffs to AI adoption, but this is mainly posturing. They were already planning to cut these jobs (because of economic slowdowns or just to restructure), but it's way easier to blame AI.
Certainly, there’s potential that these tools become groundbreaking in the future — but as it stands, this is it. Now we’re seeing reports that companies (outside of AI development) are desperately trying to figure out how to turn all this AI adoption — and the costs that come with it — into profit.
The worst part for the companies developing frontier AI models is that progress appears to be slowing. Instead of the promised all-knowing chatbot that can produce anything and everything, we are seeing a slowing down of progress and a variety of different side quests. For example, OpenAI (creator of ChatGPT) has created a text-to-video platform (Sora) and is planning to enter the custom chips and consumer electronics business.
If you were really on the cusp of building an AI chatbot that would change the world, why would you distract your employees from this with a bunch of different side projects?
The cart
Keeping with OpenAI, mind you, many frontier model developers are in a similar boat. In 2026, they are trying to get up to $100 billion in new investment. Yes, it is one of the fastest-growing companies in history, but it is burning through cash (for chips and cloud computing) at an astronomical rate. And given that this is still a private company, it requires private investment (mainly from venture capital firms). Many investors are still willing to fund OpenAI — viewing the promised land in sight — yet some are raising concerns (such as its losses being as big as the deficits of entire countries).
That leaves this firm — and many like it — with a few options.
Sell advertising.
Reduce the inference cost by shortening answers.
Increase the cost of access to these models.
OpenAI has already announced it will test ads in the US. Not to mention, again, only in the US so far, Walmart and Etsy now allow people to shop through ChatGPT.
The last two approaches mentioned above have yet to be taken, but there is a very real possibility that they will be taken as the search for profitability continues.
What this means for you
Use a variety of models with continued — and likely increasing need for — skepticism of their responses. If ads are sold or responses are shortened, the likelihood of lower-quality AI output increases. That’s why I suggest using a variety of models. That way, you can see the differences in responses to make a better-informed decision on the quality of the output, and if a model were to cease to exist, you would be familiar with different products.
Continue to develop and enhance critical thinking. I give some actionable steps in this article. Human-machine teaming is likely to be the continued way of working. The jobs that will remain are those that require critical thinking and sound judgment.
Don't give up. I have seen an increase in reporting of young people ditching university and white-collar jobs for blue-collar professions. There is nothing wrong with that — if you want to work in these jobs. In fact, I would implore more people to pick up these jobs as we need these skilled workers. That said, if you don't want to, continue down the path you are on, while being mindful that the jobs of the future will likely require critical thinking, human-machine teaming, and creativity. If you want to be a writer, do everything in your power to be the best writer. Don't just accept defeat and do a job you hate for the rest of your life because people who do not know you claim that your job has the potential not to exist.
These are the exact three approaches I am taking. I use generative models, including Claude, ChatGPT, Perplexity, Gemini, NotebookLM, and Co-pilot, to name a few. I am trying out agentic models like Zapier and want to try N8N. But while I am doing this, I still write things the “old-fashioned way,” sometimes on paper and other times just typing. All the while, I continue to remind myself that all is not lost. That I should continue my studies and develop my skills — despite what some might claim.
I believe the worst thing someone can do is keel over and give up. Keep trying to navigate this world of AI. I will continue writing with the hopes that it helps you chart this journey.
Take care,
Emanuel