Scaling and Standardization

At the most recent Lean Coffee Austin event we had an interesting discussion around the topic of scaling engineering teams. The introduction to the discussion went something like this:

I’ve done this several different times in different organizations. Each time I figured out some key insights and applied those approaches the next time. But it was very different each time. And it’s different when going from, say 3 people to 10, or 10 to 100, or 100 to 200, and so on. I don’t see the general pattern for how to do it.

Anyone who has scaled teams can relate to this. While my experience is similar to the intro, there are some general patterns that hold in scaling.

In the small start-up context you need generalists, people who can take on many different tasks, and deal with the priority flux that inevitably arises from early-stage discovery. The need for constant and often radical adjustment puts a premium on those able to cover multiple roles.

An organization grows sustainably when it can deliver customer value in repeatable fashion. Repetition invites automation. But to the extend that it is not subject to automation, it also invites standardization. So more mature organizations are heavier on specialists in these standardized areas. (Specialists with unique knowledge or capabilities are another matter.)

One apparent advantage of specialization is efficiency. A generalist is unlikely to have the same facility as a specialist does in his speciality. If the speciality is standardized rather than esoteric, that is, if it does not require rare capabilities, then it also supports another significant advantage: replaceability. Let’s call this a standardized role.

Teams of standardized roles are easier to manage than teams of generalists because of replaceability. A manager need not worry that losing a member of the team will stall operations. It may make sense to maintain some spare capacity in a standardized role to mitigate the issue.

An organization is also a network. As the number of people in an organization ($N$) grows, the organizational overhead grows with the number of connections between people, that is as $O(N^2)$. Therefore over time coordination costs come to dominate the costs of value delivery. The exact balance of coordination costs to value-deliver costs surely depends heavily on context, but the general phenomenon applies broadly. The need to focus on coordination costs leads to the management speciality. All of this drives organizations toward standardized roles and management as they scale.

In evolving toward a higher proportion of standardized roles and management, who can follow along? Early-stage employees are often unable or unwilling to learn the management skills needed to assume leadership roles. Maybe they can specialize. Or perhaps the organization still has room for their general capability. But they are no longer as fully aligned with the organization as before.

People pursue different strategies, some sticking with generalists and relying on automation instead of standardized roles, while others adopt frameworks that have an ecosystem of specialists around them. There’s no reason to expect a single winning strategy to emerge. But any growing organization will face these trade-offs. I contend that at some context-dependent size the burden of finding generalists will outweigh the benefits, and that the organization will begin to standardize roles.

These are very broad strokes. Organizations take various approaches to lower coordination costs: two-pizza teams, fractal-like hierarchies, ephemeral project teams, outsourcing, and so on. But the general tendency remains that continued growth leads inexorably to standardized roles and more management.

With AI tools swiftly improving in capability, how will that change these dynamics? I believe it will have two effects. A proliferation of smaller firms, and an expansion of larger firms. The difficulty will be in the middle ground. On the one hand, smaller groups will have enhanced capability to scale without increasing their headcount, relying instead on advanced tools, and perhaps scaling agents. On the other hand, these same tools can reduce the coordination costs for large teams in standardized roles. Each of these outcomes is in keeping with Ronald Coase’s observations in The Nature of the Firm.

Firms moving from small teams of generalists to large teams of standardized roles will have the same difficulties in transition as they have today. However, I suspect that more generalists will want to stay that way because the number of options they have will be higher, and many businesses will have to retool heavily to make way for standardization.

Some firms may need to move in the opposite direction, from large teams of standardized roles to smaller teams of generalists. This will be both rare and difficult in light of the institutional interests that grow around the status quo. I expect it to occur primarily in cases of existential crisis, when bankruptcy is the only viable option.

In sum, the need to limit coordination costs drives scaling organizations toward standardized roles and management. Many generalists struggle with this transition, or fail to make it altogether. Widespread adoption of AI will make both perpetually small firms and large standardized-role firms more efficient, and increase the difficulty of transition.

I am indebted to Arnold Kling for helping to clarify my thinking on these topics. His substack is an excellent daily read.