OpenAI announced Project Mercury in late October 2025 (Bloomberg article). The idea is simple. ChatGPT can produce sophisticated financial models in Excel, but the models have errors and hallucinations. They also don’t understand formatting conventions. So OpenAI started recruiting ex-analysts from top investment banks and firms like Morgan Stanley, JPM, KKR and Goldman Sachs to build models on which to train their models.
The process is pure AI. Candidates are interviewed by a bot about their finance knowledge. Once accepted, they build a model a week. And they get paid $150 per hour. OpenAI has already recruited over 100 people including MBA candidates at MIT and HBS.
Who knows how quickly all this will be available, but it would seem like in a matter of months if not weeks, we will all be able to generate world class financial models for just about any deal or transaction. Certainly there will be some more time and effort required to get the right financial data into these models. It’s one thing to pull public financials, it’s something very different to understand and pull private company data out of internal spreadsheets and systems. But AI will solve that problem in short order.
So where does this take us?
I started my career as an investment banking analyst and the memories of days turning into nights and back into days as I labored over models is seared into me. I don’t miss that. But I also learned a huge amount in the process. I learned how to work. I learned how to be precise. I learned to get the details right. I learned that errors, even the most innocent and inconsequential, caused doubt and decreased trust. I marveled at the Managing Directors’ ability to get to answers within 10% of the correct answer in seconds off the top of their head while I furiously inputted updated assumptions into my models. I realized they understood the numbers at a deeper level from experience and knew intuitively which variables were the important ones. Maybe most importantly I learned how to take responsibility for my work.
More than anything, the investment banking system is an apprenticeship. You start as a grunt with little more to offer than a high functioning brain, some basic skills and a willingness to work until collapse. You are rewarded with experience and responsibility far beyond your peers. Plus a lot of cash. Twelve months ago you were standing around drinking beers. Now an entire conference room table has turned to look at you to know if the number is $620 million or $650 million.
So, when everyone can generate excellent models with a few prompts, I wonder what happens to that learning curve. I wonder how recent grads build up their experience and learn the intangibles of taking responsibility for their work, learning to work with and influence others.
I also wonder about how young people find their careers. For me, one of the best parts of my investment banking and PE experience was it not only showed me what I didn’t want to do with my life (investment banking and PE), it exposed me to many different industries, even if peripherally, and gave me hard and soft tools I could use in any new path.
I recently read an essay addressing this very topic and the author posited that AI will enable every college student to manifest their own unique destiny of vibe-coded entrepreneurialism. That’s certainly true for the young people already predisposed to being an entrepreneur. And just like there are college kids who always knew they wanted to be doctors, lawyers, scientists, engineers and investors, there are kids who know they want to do start ups. But in my experience talking with people of that age, they don’t know what the hell they want to do. They may say they do because they have been trained to have an answer to this. But talk to them six months later and the thinking has changed.
The big companies like the investment banks, consultancies and corporations provide a starting path. They provide that apprenticeship to teach some hard skills as well as the soft skills. So if AI takes away the need for learning the hard skills (model building), how many apprentices do you need? And how do they learn the intangible skills to start their careers? The current trend is that these apprentice programs i.e. hiring right out of college, is scaling back. The opportunities to figure it out in your early twenties while employed by someone else are going away.
For the young people who haven’t already chosen their careers, I see three paths.
One, the apprenticeships are available, but to far fewer people. Instead of hiring 1,000 new analysts a firm may hire 300. The nature of the apprenticeship dramatically changes. For example in investment banking, now that everyone has world class models in minutes, the analysts will be tasked with getting better data so that the banks can compete for deals on better, more unique insights.
Second, the nature of career building will change from industry specific to task specific plus industry knowledge. For instance, a young person may take their first job in the marketing department of a small food brand. In the past that would have tracked them into food and small companies. But now the young person will have to emphasize the tasks they know how to complete with AI and sell themselves on those capabilities to other companies. Our junior marketer may learn cohort analysis, Meta ad management and supporting sales to big box retailers. And its cohort analysis that becomes her speciality which she then sells to life sciences and tech companies.
Third, young people will be channeled into entrepreneurialism whether they want to or not. They will have to create big or small little cash streams and combine those together to generate income and build wealth.
We already see these things happening. So I think the advice to young people becomes intentionality. Which of the three paths do they see for themselves? Are they intensely competitive and like the structure of the established, but shrinking apprenticeships? Does the idea of working for themselves appeal? Or do they fall into that middle group? If so, their focus must be on building AI enabled skills they can deepen and apply broadly.