Leadership as Layers of Indirection
In which we model organizational hierarchies in terms of indirection from the work and discover interesting properties of modeling leadership failures.
At my $day_job, I've been using a model to interpret and predict organizational changes, which I find very useful. This model is based on the concept of layers of operational indirection, and it works like this:
At Layer 0, someone does the work directly.
At Layer 1, we support someone else that does the work. We don't do the work ourselves, but we act through the actions of others.
At Layer 2, we support someone that supports someone that does the work.
At Layer 3, we support someone that supports someone that supports someone that does the work.
These layers of indirection emerge whenever the group of people required to do the work increases, and structures of support get created as a skeleton to support the increased weight of the organization.
What I find useful about this model is not how it maps to managerial roles and hierarchical expectations but how it is useful as a compass to understand whether our actions are aligned with our directional needs.
For example, suppose I have a team of people that I can influence, but I might have trouble influencing them, and it's easier for me to just do it myself. It's very common for senior engineers to struggle with this transition because all the muscles they built to be excellent at their job and recognized in their value do not translate directly into supporting others doing the job.
The job at Layer 1 is different than the job at Layer 0. It needs different muscles, and it has different failure modes.
Similarly, the job at Layer 2 is different than the job at Layer 1. It seems bizarre because we learned to support people in Layer 1. Why would it be any harder to support others, even if they are not the ones doing the work directly?
The problem is that the further away from the direct work we are, the less agency and control we have. It becomes increasingly harder to influence directly, and it becomes necessary to act indirectly. It becomes less about influence and more about reshaping incentive gradients.
In my head, I picture it like the problem of generating power. In the most direct layer, we use our hand to turn a turbine to generate electricity. If we want more power, we turn it faster, but eventually, we simply don't have enough power by ourselves.
So, we turn to other sources, like wind power. Now the problem is not how to turn the crank but how to convince air to turn the crank for us. All the muscles we built to turn cranks are now useless, and we need to find a place where there is wind and shape blades and make them big enough and place the rotor high enough from the ground to make them spin.
If we want more power, we need to make the blades bigger or change direction entirely, convincing, for example, water to turn the turbine instead. We can't influence each and every water molecule to push against our turbine to make it spin, but we can find a valley and move earth around and build dams. We can use our knowledge of how water molecules act to predict the fact that they will follow their incentive gradient (composed of condensation + rain + valley + gravity), they will fill up the lake behind the dam and eventually fall into a tube that pushes against the turbine and makes it spin, creating electricity, like we desire. We didn't have to convince a single molecule to do that.
Moving one layer of indirection further up would be equivalent to the problem of incentivizing the creation of renewable energy in a whole country. All the muscles we built finding valleys and building dams are now useless. We need to operate on different things like increasing taxation on non-renewable sources or incentivizing research into renewable energy sources. We don't even have to care about what actually ended up creating the power.
This example shows well how each layer of indirection often requires different skills because the constraints to obtain the desired outcomes change due to the increased indirection.
Another useful property of this model is the realization that we often operate at different levels of indirection even when performing the same role. I might need to write code (layer 0) to learn about something so that I can better understand the possibilities of a technology so that I can influence a discipline to adopt it (layer 2) so that it can reshape incentives in multiple teams (layer 1).
Lastly, this model is useful also for paying attention about when we “fall back” and operate at layers below we should be operating. This is a failure mode because it is a violation of the support system. The person at Layer N sees the person at Layer N+1 failing to support them and instead dive bomb and take empowerment away from them. This is often justified with things like “I didn’t have time to teach you” or “I tried many ways to help you and you weren’t doing it” but it’s often detrimental to the relationship between the supporter and those that are supported.
I found that paying attention to the center of gravity of my movement in relationship to layers of indirection has helped me be more effective and intentional about my activities and made it easier to recognize failure modes that are often hidden by the value of the resulting actions.