I wrote a piece on Towards Data Science: “Range in depth – the generalist's value in your data group.” 1
My argument at the time was simple: While experts excel at solving complex, well-defined problems, lay experts are often more valuable because they define the problem in the first place, and then bring in experts where needed.
Because of the proliferation of AI in our daily lives, I was curious to see how much those thoughts still affected me, so I went back to reread that article. My intention was to rewrite, but to my surprise, I found myself coming to terms with everything slowly wrote to the youngest. Only one subtle but very important thing has changed.
Change: AI as the new technology
In the past five years, AI has advanced to the point where it can handle many of the tasks we used to rely on experts for. The kind of work that required deep expertise, clear, concise and well-defined instructions, is now where AI thrives. And unlike humans, it does this quickly and without fatigue.
So I decided to continue writing about it, but rather than rewriting, just reflecting on my previous thoughts, highlighting where some tweaks are needed.
1. We are still working in bad learning environments
We don't work with clean, closed systems. We are working on what David Epstein calls bad learning environments2-settings where the rules are unclear, feedback is delayed or misleading, and patterns do not repeat regularly. In these cases, you can do the “right” thing and get the wrong result, or the wrong thing and appear to be successful. That's what makes them dangerous.
The real challenge is not solving problems. It wasn't five years ago and it isn't today. The challenge is knowing which problems are worth solving, and whether the signals you use to guide you can be trusted.
AI does not remove this ambiguity. If anything, it amplifies it. The faster the answers come and the more convincing they seem, the greater the risk of confidently solving the wrong problem.
2. The need for hyper-specialization is decreasing (but not gone yet)
At the time, I argued that access to information reduced the need for deep technology. The abundance of Stacks, blogs, and articles meant that a skilled generalist could find things quickly to move forward.
Today, that dynamic has changed dramatically.
Information is no longer available. Selected, collated, compared, and presented… suddenly AI doesn't just help you find the answer. It gives you a performance feedback.
And that takes us further:
The need for hyper-specialisation is not disappearing, but being pushed back to the brink (some would say the abyss). Generalists are now given the ability to advance before requiring specialisation.
3. The integration effort is still a real killer
A generalist reduces the integration effort by removing unnecessary relationships, because they vary across all. They need to be empowered to make decisions and thus cut off the management of extra relationships.
This was one of my strong points then and it still holds true today. The cost of collaboration in organizations is often underestimated and that has not changed.
Jeff Bezos announced the “two pizza team”3 rule: groups must be small enough to be served with two pizzas. In today's world, you might argue that we are headed for it groups of one pizza. Not because the work is easy but because generalists are more skilled and AI fills more specialist gaps resulting in less handovers being needed.
4. The business problem has not changed
When you break everything down, the main questions remain exactly the same:
- How do we increase revenue?
- How do we retain customers?
- How do we work effectively?
The tool has changed (a lot). The methods have become more complex. But the underlying problems have not changed.
And five years on, businesses still don't care whether the solution involves a sophisticated agent model or a well-placed SQL query. They may say they do in Exec meetings, but they don't really look at how it was achieved, if only it was resolved.
So in summary, what has changed?
Not the importance of generalists. If anything, their number has increased.
The main changes are:
Generalists are no longer just connectors between specialists. They are the ones who wander in places where the problem is unclear, the signs are loud, and the way forward is not visible.
They don't just connect people, though abilities-deciding when to trust intuition, when to trust experience, and when to bring in the most needed expert, human or AI.
Their distance now has been increasedable to do the most profound work on its own. Not because the world became easier, but because they still work well with complexity, with AI as their special layer that is always available.
I look forward to my personal AI assistant doing some thinking in five years.
[1] Potgieter, C. (2021). Range in depth – the generalist value in your data group. Towards Data Science. https://towardsdatascience.com/range-over-depth-the-value-of-a-generalist-in-your-data-team-174d4650869d/
[2] Epstein, D. (2023). Good and Bad Learning Environments.
https://davidepstein.substack.com/p/kind-and-wicked-learning-environments
[3] Two Pizza Parties: The Science Behind Jeff Bezos' Law | Inside Nuclino. Blog.nuclino.com. Published in 2019.
