Idle thought of the day:
If you had internalised the idea that controlling variability is the foundation of management, then the results of early studies such as the 2024 DORA report (https://redmonk.com/rstephens/2024/11/26/dora2024/) or that NBER report on the effect of LLMs on the Danish labour market (https://www.nber.org/papers/w33777) aren't counter-intuitive at all but expected
From the Germ Theory of Management (https://beyondcommandandcontrol.com/wp-content/uploads/2015/06/the-germ-theory-of-management.pdf):
> Few people understand what uncontrolled variability does to the cost of doing business. Fewer still understand what can be done about it and what is the
management‟s role
And
> If you remove the sources of variability from any process, you make it more predictable and therefore more controllable. You can schedule activities closer together and eliminate waste and delay.
LLMs and generative models are massively volatile. The variability they inject into organisations and processes make those systems hard or even impossible to manage. What might be a minor productivity bump for an individual becomes process poison that throws off an entire organisation
@baldur
I enjoyed the essay. I'm not sure I can relate use of LLMs with the type of process described in it.
In my experience on the job, knowledge work can't really be subjected to this kind of systematic variability reduction; yes, you automate or otherwise systematize what can be automated, but once that is established, it is no longer knowledge work.
So I'm not convinced that LLMs are increasing variance in previously highly-deterministic activities.
I would expect most people simply haven't found a good use for them yet; in some cases they might never do so.
