Three years ago, I was a Neely Fellow finishing an MBA, with a return offer from a technology consulting internship in hand and, academically at least, fully bought into the hype about a fourth industrial revolution driven by AI. I assumed my first years back in the workforce would be spent trying to steer something that was already moving quickly. But the pace turned out to be much slower than that.
From a classroom or a podcast feed, the future felt imminent. We thought DeepSeek would flip the industry’s dynamics on its head. Then Mythos and Project Glasswing put everyone on their toes. Last week it was declared that we (again) have AGI with Astra. Every week there’s a demo that appears to obsolete an entire function, and you start to feel late. I don’t mean to take away from the fact that these updates are often step changes in capabilities and an incredible testament to human ingenuity that we are even talking about such a technology.
However, inside large companies in the US, which account for ~60% the GDP, the story is quite different. Back when I was applying for the Neely fellowship in 2023, ChatGPT broke records as the fastest application to reach 100M users. But that was retail, users on free plans asking for poems and recipes. Even 3 years later, three of four businesses are not using AI.
I recently moved into the transformation group of my organization and through this lens see how clients measure and consider changes in periods of contract cycles and quarterly (or even annual) planning increments and not in response to current headlines. Active projects were greenlighted based on pre-AI pricing assumptions, leaders measured against SLAs don’t want to risk those with non-deterministic AI, legal governance denies permissions to use AI for core processes. “If it ain’t broke, don’t fix it”. These are structural and bureaucratic reasons for change being slow.
The other reason is that benefits realization lags adoption. No surprise, benefits are often from headcount reduction. The people whose work processes are being automated know the intricacies of those processes better than anyone else. Bringing them into the design process makes the product better. It also means that after launch there’s someone credible on the floor who vouches for the AI and educates their team, which matters more than any amount of change-management collateral. This can slow down benefits realization, but you can’t kick the can down the road forever.
That brings us to redeployment. Given a choice, leaders would like to retain headcount even if work is being automated since (if you like your glass half full) they empathize with their workforce and want to smoothen transitions. And if you like your glass half empty, headcount often serves as a proxy for power in corporate America. Redeployment, however, only works if someone starts on it early. Running the organization smoothly, while automating core processes with new technology can be challenging enough to get right. Identifying and operationalizing new opportunities that were not being done so far due to capacity limitations so people and finding the right people to redeploy to it is a whole other program.
Lastly – Upskilling. You could argue that the accounting industry is stronger than ever, and excel gave accountants more to do rather than automating their jobs in the 80s. I too used to imagine AI as an excel-like fundamental tool - Take a few weeks of online courses, bring a curious spirit and you’ll find your way. That isn’t it though. For most people, re-skilling can’t start until AI is actually adopted at their organization, encouraged by their leadership, and touching their own scope of work. Only then is there something specific to learn like the tools they’ll be using and the logic behind AI agents for their team. That sequence takes quarters or even years.
I wanted a dynamic career and I got one. My client experiences include the “innocent” projects and also projects that could invite debate across (1) industries including casino game producers and oil drillers and (2) project types including AI-driven rightsizing, offshoring and outsourcing.
In the MBA we often studied corporate decisions after the fact, and had the luxury to be critical about their externalities and long term consequences. But in corporate America the ethics of decisions isn’t really debated. Instead they’ll be hidden in plain sight under meetings about sequencing, capacity planning, rightsizing, or benefit realization. Good people may choose to pause and think about this, but they don’t have to. No one’s performance is tied to ethics. As we learnt in the MBA, we are all indirectly in service of the shareholder for whom a business is one of many assets in a portfolio that needs to stay competitive to retain capital.
The silver lining and irony is that the same scale that makes AI’s rise feel unstoppable from the outside is what makes it crawl from the inside - Big systems move the world fast and themselves slowly. This slow march of progress should let decisions simmer long enough for ethics to bubble into sight, teams to redeploy talent, and individuals to reskill.