I wrote a chunk on In the direction of Knowledge Science: “Range over depth – the value of a generalist in your data team.” 1
My argument again then was easy: Whereas specialists excel at fixing advanced, well-defined issues, generalists are sometimes extra worthwhile as a result of they outline the issue within the first place, and solely then usher in specialists the place wanted.
Because of the surge in AI in our every day lives, I used to be curious to see how a lot these ideas nonetheless resonated with me, so I went again to re-read that article. My intent was to do a rewrite, however to my shock, I discovered myself agreeing with nearly the whole lot my barely youthful self wrote. Just one delicate however essential factor has modified.
The shift: AI as the brand new specialist
Within the final 5 years, AI has developed to the purpose the place it may well deal with most of the duties we historically relied on specialists for. The sort of work that required deep experience, a transparent transient and well-defined directions, is now precisely the place AI thrives. And in contrast to people, it does this quicker and with out fatigue.
So I made a decision to nonetheless write about it, however fairly than a rewrite, a mere reflection on my earlier ideas, highlighting the place some tweeks had been obligatory.
1. We nonetheless function in depraved studying environments
We don’t function in neat, closed methods. We function in what David Epstein calls depraved studying environments2—settings the place the foundations are unclear, suggestions is delayed or deceptive, and patterns don’t repeat constantly. In these environments, you are able to do the “proper” factor and nonetheless get the unsuitable final result, or the unsuitable factor and seem profitable. That’s what makes them harmful.
The true problem just isn’t fixing issues. It wasn’t 5 years in the past and it positively isn’t right this moment. The problem is realizing which issues are price fixing, and whether or not the alerts you’re utilizing to information you possibly can even be trusted.
AI doesn’t take away this ambiguity. If something, it amplifies it. When solutions come quicker and look extra convincing, the danger of confidently fixing the unsuitable downside solely will increase.
2. The necessity for hyper-specialisation is shrinking (however nonetheless not gone)
Again then, I argued that entry to data diminished the necessity for deep specialisation. Stack Overflow, blogs, and documentation meant {that a} succesful generalist may determine issues out fast sufficient to maneuver ahead.
At this time, that dynamic has modified considerably.
Data is now not simply obtainable. It’s curated, synthesised, in contrast, and introduced… immediately AI doesn’t simply assist you discover the reply. It offers you a working reply.
And that pushes us additional:
The necessity for hyper-specialisation isn’t disappearing, however it’s being pushed nearer to the sting (some would say the abyss). Generalists are actually empowered to go a lot additional earlier than needing specialist enter.
3. Coordination effort remains to be the actual killer
The generalist reduces the coordination effort by basically eliminating pointless relationships, as a result of they vary throughout them. They should be given the mandate to make choices and thus reduce out the administration of added relationships.
This was one in all my stronger factors again then and it holds much more right this moment. The price of coordination in organisations is commonly underestimated and that has not modified.
Jeff Bezos popularised the “two-pizza workforce”3 rule: groups needs to be sufficiently small to be fed with two pizzas. In right this moment’s world, you possibly can argue we’re heading towards one-pizza groups. Not as a result of the work is easier however as a result of generalists are extra succesful and AI fills many specialist gaps which leads to fewer handoffs being required.
4. The enterprise downside hasn’t modified
If you happen to strip the whole lot again, the core questions stay precisely the identical:
- How can we develop income?
- How can we retain clients?
- How can we function extra effectively?
The tooling has developed (considerably). The strategies have turn out to be much more subtle. However the underlying issues are unchanged.
And simply as 5 years in the past, companies nonetheless don’t care whether or not the answer entails a cutting-edge agentic mannequin or a well-placed SQL question. They may say they do in Exec conferences, however actually they don’t seem to be the way it was achieved, simply if it was solved.
So in abstract, what modified?
Not the significance of generalists. If something, their worth has elevated.
The important thing shift is that this:
Generalists are now not simply connectors between specialists. They’re those navigating environments the place the issue is unclear, the alerts are noisy, and the trail ahead isn’t apparent.
They join not solely folks, however capabilities—deciding when to belief instinct, when to depend on expertise, and when to herald an on-demand specialist, human or AI.
Their vary is now amplified, able to executing a lot deeper work themselves. Not as a result of the world grew to become less complicated, however as a result of they nonetheless function nicely in complexity, with AI as their ever-available specialist layer.
I’m trying ahead to my private AI assistant doing one other reflection in 5 years.
[1] Potgieter, C. (2021). Vary over depth – the worth of a generalist in your information workforce. In the direction of Knowledge Science.https://towardsdatascience.com/range-over-depth-the-value-of-a-generalist-in-your-data-team-174d4650869d/
[2] Epstein, D. (2023). Sort and Depraved Studying Environments.
https://davidepstein.substack.com/p/kind-and-wicked-learning-environments
[3] Two-Pizza Groups: The Science Behind Jeff Bezos’ Rule | Inside Nuclino. Weblog.nuclino.com. https://weblog.nuclino.com/two-pizza-teams-the-science-behind-jeff-bezos-rule. Printed 2019.
