 I've become convinced that if you park somebody in front of a computer to play the game of life for long enough, they eventually start to contort themselves until they kind of look like a glider. I call it Conway's Law. Melvin Conway was one of those nerds who helped lay the foundations for modern computing. Like if you've ever heard of a co-routine, yeah, that's this guy. But he's more well known for his observations about how the shape of technology tends to echo the group dynamics that created it. When an organization designs some sort of system, a software suite, a transportation network, a mechanical pencil, agricultural legislation, whatever, the final product is usually composed of a number of subsystems, some of which will also be subdivided and subdivided again until each component can be fully understood in its totality by a person. Conway suggested that the relationships between those components and subsystems will mirror the relationships between divisions and members of the organization. Like say there's a team designing a car where one person is in charge of wheels and another person is responsible for designing axles. Because wheels and axles are being handled by different people, it's unlikely that the final design of the car will feature some sort of seamless hybrid of wheel slash axle, a waxle. The two designers will probably work together to negotiate a well-defined boundary between the components, an interface, allowing each component to be cleanly separated and worked on independently from the other. That point of separation would be apparent to anyone who looks at the construction of the car, a physical dividing line that marks where the wheel designer's responsibilities end and the axle designer's responsibilities begin. And that split perfectly mirrors the organizational split between the two individuals. This principle called Conway's law, or the mirroring hypothesis, is one of those things that once you know about it, you see everywhere, front end, interface, back end, legs, interface, bearing, interface, seat, battery, interface, battery, terminal, interface, you get the picture. It's sort of a corollary of the division of labor. By breaking big, unwieldy problems up into smaller, tractable sub-problems, we can claw our way toward a solution without anyone needing to understand all its details all at once. The folks working on it can develop knowledge and expertise about their own little corner of the answer without worrying about the other bits. It's hard to imagine building something as complex as a train if the person who's figuring out how to bury rail ties also has to have the know-how necessary to specify the metallurgical composition of the smokestack and everything in between. But in order to divide our labor, we have to have some sort of plan for breaking the problem up. And it makes sense that the plan we choose is going to map pretty well to the shape of our solution. However, Conway notes that there are some drawbacks to this approach. As systems develop from slapdash prototypes to more orderly and predictable machines, it usually becomes clear that a different architecture would probably work better. Maybe there's some redundancy that can be eliminated by combining subsystems. Maybe it makes more sense to put two densely connected components closer together. Unfortunately, if the organization can't easily reform itself along new lines according to these observations, its shape comes into conflict with the shape the system wants to be, leading to a bizarre patchwork of unnecessary compromises and epicycles to cram a round peg through a square hole. You may remember our discussion about Fred Brooks' book, The Mythical Man Month. Those ideas are both corollaries of Conway's law, and the original paper he submitted to Datamation highlights them in that context. If you add more people to a project, you have to specify more interfaces for both your team and the designed system, which, rather than accelerating the project, can absorb time and resources better spent elsewhere. If you don't have a single cohesive vision for what everyone on the team should be working towards, it's only a matter of time before one module will be designed at cross purposes with some other module as creators aren't thinking about, making the whole system clunky as it finds itself a house divided. Conway's law also has troubling implications for how organizations can react and adapt to new information. In a 2011 case study on manufacturers of jet engines, the three companies examined were, as expected, broken up along more or less the same lines as the systems they were designing. Each modular sub-assembly had a different group of people responsible for its development, ignoring knowledge and information with other teams, only through agreed-upon interfaces. When an innovation in the design of the turbine substantially improved its thrust, all three companies began to see a rash of gearbox failures. With the benefit of hindsight, it was a pretty straightforward problem. The turbine's thrust went up, so the gearbox was subjected to new forces at the junction where the two fastened together. But when the problem first came to their attention, the cause of the failures wasn't so easy to spot, and the structure of those companies didn't make it any easier. Two of the firms sub-contracted their gearboxes and turbines to different external suppliers. Manufacturers with no real reason to consider how their design decisions might influence each other beyond the horizon of what they were contracted to build. This made it a real struggle to realize that the problem actually existed at the interface between those two sub-assemblies. The third company made everything in-house, and was much quicker to diagnose and fix the issue. Most likely because there was less of an organizational gap between the two departments, and a greater incentive for them to figure out the problem together. The gearbox fiasco is a great example of how the simple act of dividing labor creates blind spots. If people take compartmentalization of work and knowledge seriously and stay in their lane, they'll often find themselves less capable of solving or even recognizing problems that don't follow the lines of the agreed flowchart. Unfortunately, managers are often pressured to break up projects into smaller parts before they have any real idea of what shape the solution should be. A practice researchers have termed premature modularization. Modularity is very attractive for project management. It's easier to optimize a simple subsystem with a well-defined function than to wrestle with a complex network of interdependent components under a vague mandate like make something that works well. Even if it would be more efficient overall to allow organizational ambiguity in the early stages to let everyone settle into a framework of sensible relationships on their own, that period of uncertainty can be stressful, difficult to measure, and hard to explain to higher ups. That pressure can result in an understandable impulse for managers to draw lines of modularization before the system, or the team working on it, is ready, with potentially disastrous consequences for both. Fortunately, Conway's law isn't a law of physics. The mirroring between organization and product structures can be disrupted, either accidentally or intentionally, by phenomena that actual researchers paid real money to write legitimate peer-reviewed papers about this stuff, have insisted on calling breaking or misting the mirror. Breaking the mirror happens when an organization looks at the architecture of whatever it's making, then defines its own structure in ways that will straddle boundaries. Maybe you have a team responsible for half of one set of somebody and half of another, or maybe teams are made deliberately multidisciplinary. You can probably imagine how having a welder in the room during an engineering meeting might result in different decisions being made. In contrast, misting is a blurring of the boundaries between sections of an org chart, a diffusion of knowledge and design considerations beyond any explicit division of labor that makes who's responsible for what fuzzier, a firm that encourages its members to share information, expertise, and tasks between departments, even departments that wouldn't normally have anything to do with each other, can sometimes reduce the threat of module-based myopia. If you've got to wire a few hundred circuit boards in a hurry, it's nice if you can drag someone from the R&D team over to help you out and maybe show you a few things about soldering. Several companies that actively pursue these anti-mirroring approaches have a decent track record, suggesting that it really is possible to design institutional safeguards against the issues we've noted. Even so, the mirroring hypothesis is alive and well, not because hard-line divisions of sub-assemblies, labor, and knowledge are the absolute best way to approach every single problem, but because it's how we're used to working in teams. When's the last time you applied for a job where the role you were intended to play at the company wasn't rigorously prescribed? It's clear that this strategy can result in massive gains. All sorts of technology benefits from competition and innovation within rigidly defined and well understood modules, from hard drives to shoelaces. But there's a price for creating those modules, one that can't be fully known when their boundaries are drawn. To get modularity, you sacrifice flexibility. To gain focus on a subsystem, you lose perspective and awareness of the bigger picture. To reduce the complexity of an unwieldy piece of the puzzle, you add more pieces and more interfaces to keep track of. Those trade-offs are often worth making, but only when we can be reasonably sure that we've more or less nailed the shape of the solution and organize ourselves to match it. Even then, it seems that there are advantages to breaking the symmetry between an organization and its product, to cultivating a more malleable and organic way of thinking about our chosen solutions and the roles that we might play in them. I guess what I'm trying to say is, it's not a crime to break Conway's law. Do you see evidence of the mirroring hypothesis in organizations you're part of? Can you think of any possible advantages to less rigid divisions of labor? Please, leave a comment below and let me know what you think. Thank you very much for watching. Don't forget to blah, blah, subscribe, blah, share. And don't stop thunking.