Distributed Culture

Distributed Culture

Controlling the Beyond-Control

Within a single computer, computation is principally limited by available memory and processor speed. In distributed computing, the time to complete an operation is dominated by laggy networks. It is crucial to design networks with both minimum latency (time to pass a message) and maximum efficiency (the most information in the fewest, shortest messages). Remote work has the same inherent lagginess. Every synchronous interaction–Zoom meetings, Slack chats, phone calls, or anything else that requires participation in the moment–takes more time than a quick question in the office9. Just as in distributed computing, effective distributed work must minimize latencies (wait times for responses) and maximize efficiencies (reducing the number of synchronous interactions).

Distributed cognition makes several observations about the impact of latencies and inefficiencies in collaborative work. In “Rebuilding Culture”, we will discuss the large body of research on overcoming latencies in remote education (hint: it’s all about asynchronous collaboration and autonomy). In this section, we’ll focus on a simple truth: our global experiment in remote work was ill-prepared and involuntary. Many sources of latency are largely outside the control of individual employees. As we’ve said repeatedly, people are different, and their circumstances are different as well. Unfortunately, research also shows that arbitrary factors, like these context differences, strongly influence our perception of others’ quality of work. Given the slightest opportunity, our brains love to make attributions where none exist10.

There have been many examples of companies successfully adopting limited remote work practice. Xerox, for example, had 11% of its employees in a virtual workforce program. Countries and companies that transition successfully to remote work tend to be large, resource-rich, and have robust digital platforms and payment infrastructure already in place. Additionally, their remote work programs have focused on high-skill professionals that were working in small teams.

What should be obvious is that the experience of 10% of the most highly-paid, highly-educated employees isn’t going to generalize well to everyone else. As the rest of the workforce joins them, laggy internet connections, closed schools, under-training, and the simple reality that not all work can be done from home will throw existing best practices into disarray.

Perhaps the most fundamental situational factor is access to high-quality and reliable IT, particularly broadband. Not all companies and countries have invested in adequate IT infrastructure. Poor neighborhoods, rural locations, and historically disadvantaged communities have been left behind in broadband connectivity. Employees working from these locations have poorer network connection when logging in for those crucial video conferences. Although the quality of that connection is both outside their control and not indicative of their quality of work, research suggests that coworkers will slowly but inevitably draw a connection between quality of work and quality of video.

Other IT inequalities exist in remote work that were never issues in the office. Employees might have wildly differing personal hardware, from computers to cameras, and sub-optimal wifi. More subtle but even more crucial, not everyone has a professional workspace available to them. Only the most fortunate will have a dedicated home office, but even the ability to work from the kitchen counter is impacted by crowding in a small household, family and flatmates competing for internet bandwidth, and neighborhood construction11.

Obviously none of these external factors have anything to do with the qualifications of an employee. In the office, everyone has access to the same infrastructure. At home, someone’s postal code can have an outsized influence on their capacity for work. The lagginess of that video connection or the inability of someone to run new software has little to do with an individual employee but negatively impacts our perception of their performance.

For managers, you can combat these biases by providing multiple channels of communication to allow employees to contribute in whatever way is most effective. For example, a meeting centered around Microsoft Teams or Zoom should also leverage simultaneous chat and asynchronous collaborative documents.

For large employers, step up and provide your employees with the infrastructure they need. Supply your workforce with high-quality equipment. Here are some good recommendations: “How to make video calls almost as good as face-to-face”. Treat an investment in their home office as an investment in your infrastructure.

In fact, for distributed work to be successful, you must consider community broadband a common good asset to your company, just like a reliable transportation network was for the office. This is not social justice12; it’s just good business.

In the context of Covid-19, many companies have responded with reduced hours, furloughs, and even layoffs. This only makes sense if you see employees as a fixed asset incapable of change13, but in fact, it’s not only possible for employees to change, that change is crucial for adapting to distributed work. Just as we’ve discussed upgrading IT infrastructure, how about upgrading your company’s human capacity infrastructure. Investing in human capacity is a common good asset, just like transportation and IT; job training pays off even when there is no guarantee that your firm will reap the increased performance benefits of the specific worker you trained. The biggest benefits will not come from “upskilling” for specific hard skills but from investing in developing the kinds of meta-learning–autonomy, communication, self-management–that are most related to success in distributed work.

Investing in training for individual employees is just a start. A number of existing business systems have proven to be incredibly fragile to economic shocks like Covid-19. A more ambitious agenda would conceptually align supply chains with distributed work. Where most of the thinking around remote work has concerned laptop jockeys attending meetings in their pajamas, distributed work like distributed computing is a more generalized idea of shifting away from laggy, fragile, synchronous systems. While we’ve been talking about employees, this could include entire factories that represent single points of failure in a production line. Even though manufacturing and warehousing require physical locations, these can be broken up into smaller, more dynamic facilities that are less vulnerable to cascading failures or outbreaks. They also have the advantage of being manageable by small teams with flat hierarchies, which as we will see, are drivers of success (and innovation) in distributed work. Distributed work practices themselves might support a shift to distributed manufacturing.

Broadband won’t help everyone because not everyone can work remotely. Only “37% of jobs in the United States can be performed entirely at home” and, as we’ve noted, these jobs mostly focus on professionals in knowledge and creative fields. Manufacturing, logistics and distribution, and a wide variety of other job verticals have traditionally required a physical workforce that can’t operate remotely in the same capacity. These industry differences impact certain cities and nations more heavily, as “lower-income economies have a lower share of jobs that can be done at home.” The economic disparities between employees that can easily shift to remote work and those that cannot are hard to ignore. Any strategic plan to create a distributed workforce needs to include everyone.

Most of the world has built a work culture around offices, factories, and other physical locations. We’re in a moment of profound transition that will expose a great many inequities in employees’ ability to work from home. They are still good employees even if we have failed to build an infrastructure designed for remote work. Saez & Zucman argued that we shouldn’t let otherwise viable small businesses fail during the lockdown and erode our long-term business capacity. The same applies for human capacity.


In previous sections we have thought about remote work largely in the context of individual employees and how managers and companies can support them. But a true distributed work strategy needs to think in terms of communities. The goal shouldn’t simply be policies that make selected individuals successful, but a culture that supports collaboration and innovation. Supporting a community culture is particularly challenging though, because just as in distributed computing, distributed work suffers from extremely narrow bandwidth that limits communication. Many organizations have tried to overcome this limitation with better technology. While improving our tools can help, the real solution is a culture that is native to distributed collaboration.

In many ways, the bandwidth of communication between two or more people has always been narrow12. However, in-person teams are always in implicit contact, whereas remote members are only ever in explicit contact (video meetings or social channels). If we force old habits through narrow pipelines, human communication can only be slowed by working remotely. Some might argue that cutting out unnecessary meetings and distracting socialization is a good thing; as we noted in “Business-as-Usual”, however, there is surprisingly little and conflicting evidence that remote work decreases distractions.

Overcoming narrow bandwidth requires increasing the richness of communication as much as possible. There are three prominent themes in the research on remote work culture. The first is the importance of intentionality in establishing that culture. The second is the crucial role of asynchronous communication and collaboration. And the last is the degradation of innovation and inclusion (and some hints on how to overcome it).

From a distributed cognition standpoint, culture is a set of shared tools that improve communication and problem solving. From this perspective, the number one rule of distributed work is don’t let chance, laziness, or bad habits define your work culture. For example, scaling effects of online networks cause many companies to form large, undifferentiated communities through their social channels, where smaller, more nimble groups would have existed in offices. I experienced this firsthand while briefing one of the world’s largest tech companies. They have many offices around the globe, but when work shifted wholly online, employees outside the headquarters’ timezone felt isolated, unappreciated, and excluded. Research has shown that large, centralized teams tend to develop ingroup-outgroup mentalities that promote conflict and degrade coordination. The scale effects of internal social networks can work against you by connecting everyone to everyone else. Organizations must actively establish smaller, more distributed teams.

Flat Hierarchies & Shared Leadership

Two crucial factors for the success of small teams are flat hierarchies and shared leadership. Narrow pipelines create communication bottlenecks that become a drag on the productivity of hierarchical teams. When top-down management cultures are constrained by limited communication channels, individual employees often wait for direction in response to unknown or uncertain situations. One reason relatively small, autonomous teams have been most successful in distributed work is that they are not limited by communication bottlenecks.

Unfortunately, in the absence of a strong, established culture, granting autonomy to every employee simply produces chaos. In fact, remote work can even exacerbate these problems because distributed teams over attribute communication failures to incompetence compared with colocated teams. Synergists flourish with autonomy because they possess a diverse set of meta-learning skills–self-assessment, emotional stability, conscientiousness, and resilience. These socioemotional factors provide an internal structure that keeps them aligned with their teams. For balancers, culture is an external structure that plays the same role. It provides them with the explicit boundaries that they need in order to work autonomously. For culture to effectively scaffold all team members, goals, roles, and communication norms must be formalized from the very beginning.

Additional research on remote work reveals that people’s relationships with their colleagues suffer compared to in-office work (though anecdotal evidence suggests a decreased number of downward comparisons, i.e. people worry less about others slacking off). A 2013 Gallup poll found that remote workers log an extra four hours per week on average; these increased hours result in additional work stress as noted in “People Are Different” and “Controlling the Beyond-Control”. Much like the misattribution of communication failures, this additional stress can also be misattributed to remote colleagues and lead to the degradation of coworker relationships. This will be particularly prominent for balancers in teams that lack explicit boundary norms.

Research on successful teams, both remote and in-office, support the importance of all the findings above. For example, Google’s well known internal research indicates that psychological safety is a principal predictor of team success. This meant that team members had a set of shared norms and trusted one another without need for constant communication. In fact, members of Google’s most successful teams had fewer synchronous communications internally and spent more time working autonomously. A study by the company Dropbox on collaborations between successful academic teams revealed that the most successful teams were small, autonomous, and maintained flat hierarchies. These qualities agree with the research findings noted previously, but their analysis also suggests the importance of role-modeling by senior members. Other research indicates that sharing stories company-wide can play the role of role-modeling in distributed work. Stories of successful problem solving by teams are the principal mechanism of establishing cultural norms. All of this research suggests the importance of transparent role-modeling in establishing culture within organizations, particularly in distributed work where potential role-models would otherwise be hidden.

From my own research on over 60,000 companies, it’s clear that culture is not the slogans a company writes on its walls, but rather what its employees actually do. Inspired by an analysis of personality characteristics conducted by Facebook, I collected data on tens of millions of employees at these companies and analyzed their own self-descriptions as a way to understand the “personality” of a company. This machine-learning analysis revealed 186 dimensions of cultural variability across companies. Some of the most prominent were, for example, management- vs. employee-driven cultures15 and research- vs. process-driven cultures. Engineers at management-driven organizations bragged about how they would return beautiful products when given explicit instructions; engineers at employee-driven companies bragged about themselves and how they would solve problems. Unsurprisingly, innovation correlated strongly with employee-driven cultures, while management-driven was related to better risk management16. As for process-driven cultures, let’s just say they were very common in the DC Beltway.

One of the most important takeaways from my analysis of company cultures is that they are the product of the employees, not the mandate of management. Because culture in my model was derived from employee behavior, I was able to experiment with predicting cultural fit simply by looking at how well a potential employee fit the model of a given company. One of the most important takeaways from this insight is that establishing (or re-establishing) a culture doesn’t come without effort. It requires shifting the norms of an entire population of employees. Driving such a shift requires at least two components: an intentional framework of cultural norms and role-modeling of those norms by leadership. The single biggest impact leaders have on their teams is not their memos or strategic plans, but their actual actions, the sacrifices they make to accomplish organizational goals. Those actions are often hidden in the best of times, and in remote work, entirely buried. It’s important these stories be shared transparently.

Another finding from my work is that the best teams are intentionally based on complementary diversity. This means that teams are designed so that members’ relative strengths and weaknesses are complementary to one another. These could be differences in expertise, experience, personality, identity, or many other dimensions. Composing teams based on complementary diversity allows peer role-modeling to play a powerful role in developing team capacity, but new teams by their very nature lack psychological safety. People that are different tend to speak a different language, at least metaphorically, and sometimes literally. Strong cultural norms are crucial for combating the stress and misattributions that tend to dominate new, diverse teams.

While intentionality and role-modeling are essential for establishing cultural norms, there are a number of well-grounded tactics that can help overcome narrow pipelines and improve communication efficiencies immediately. One of the most classic is to set a specific agenda with desired outcomes pre-identified for all meetings. This is as true in distributed work as it is in the office. For a more flexible approach to meetings, some have suggested “borrowing the idea of ‘office hours’ from academia”. Rather than having a fixed meeting, managers and employees more generally can post links to weekly video sessions where they can respond synchronously to any small questions that have popped up throughout the week. The idea is that this would cut down on the number of distracting interruptions by having a fixed and reliable time in which anyone can ask a question. (Having hosted a great many office hours myself, I can say that it can be an enjoyable distraction to simply have a free hour where you’ve dedicated yourself to a mix of answers and conversations with a random assortment of characters.)

Asynchronous

Another common recommendation is the use of message boards and chats as a solution for communicating while teleworking. Unfortunately, research hasn’t been as kind to this idea. As noted above, undifferentiated message boards and internal chat technologies can tend to become rather bloated and create in-group/out-group relationships in a company. They are also subject to network capture–undifferentiated social interactions that tend to have a winner-take-all quality–in which a small number of individuals dominate the conversation. While a thousand separate conversations within small groups of people may be a poor way to disseminate information, one massive conversation between five thousand people tends to erode innovation and (ironically) inclusion.

Meetings and chats are explicitly synchronous communication technologies. Email lives somewhere in between. Many people use it almost as a kind of chat, shooting off quick questions and expecting quick responses. For others, email is like letter writing in the 19th century, a labored, near-literary process17. A number of studies of email use have shown that more time spent on email increases stress and decreases perceived productivity. It turns out that most of this subjective productivity loss comes from people who allow emails to interrupt their workflow. Psychological literature suggests that this might be a form of learned helplessness, in which intrusive events outside an individual’s control decrease their long-term performance at a neural level on an unrelated task. Instead, I recommend treating email as an asynchronous technology. Rather than immediately responding to incoming emails, individuals need both the team norms and the self-regulation to address emails in batches when it fits with their workflow. To paraphrase the CEO of Automattic, I’ll get to it when it’s the right thing to do18.

Used in this way, email shifts towards asynchronous communication. In distributed computing, “asynchronous” is a challenge to be overcome. Most distributed systems operate on queues, where pieces of information and tasks are held in the order received. Individual computers have established rules of how to handle and order the queue, what to do when it overflows, and how to interact with other computers on their own schedules.

In distributed cognition, “asynchronous” is a superpower. It allows teams to make collective progress on projects even when they’re not in direct synchronous communication. Intelligent systems (like us (well...most of the time19)) have an advantage over the simple queuing algorithms of distributed computing in that we can build mental models of one another. We can infer the conditions and intentions of all of our teammates and adapt our actions dynamically, whereas hard-coded systems must cover all possible outcomes or break. As such, distributed work is much more robust than distributed computing when the three key ingredients are present: autonomy, psychological safety, and norms.

Many of those key communication norms are very grounded and tactical. For example, teams must enforce norms about using modes of communication that match the urgency of the message. Don’t send an email if you absolutely need an immediate response; don’t send a text message about a complex and emotional question. To make the most effective use of the asynchronous strength of email, we can actually follow the norms we identified above for meetings. Every email should have sufficient background, required outcomes, and an explicit due date. This gives the recipient the information they need to make their own decision about the contents of the email.

While email isn’t inherently asynchronous20, more and more digital tools are designed to be natively collaborative. In fact, it’s gotten to the point where I am disappointed in cloud software if two people can’t work on the same document at the same time. Unlike discussion forums and chats, tools such as Google Docs, git repos, Notion, and wikis aren’t just a log of every debate or offhand comment that led to a finished project. They are more than just a record of a process; their true value emerges from the evolving synthesis of all the collaborative learning that got you there.

Perhaps the most famous asynchronous tool of them all is Wikipedia, the world’s largest encyclopedia21. Each hyperlinked article might represent years of learning and negotiation between the editors collaborating on that subject. While the wiki tools log all of the individual edits, contributions, and comments, the live article represents a synthesis of learning, creation, and social communication. The wiki creation process doesn’t force Wikipedia editors to be in the same place at the same time, but it does transform their raw communication into shared knowledge. (Could you imagine if Wikipedia was a series of Slack channels detailing all of the arguments and debates over what constitutes the “truth” on a given subject?)

With the collaboration and transformation inherent to asynchronous tools, every evolving project document represents the culture underlying the work, or as Thomas Malone would term it, a “supermind”. The dynamics of asynchronous tools capture the intangible capital of a team and begin to break down the distinction between planning and working. They force an explicit embodiment of the collaboration–not just a list of action items but a shared understanding.

Asynchronous collaboration is where distributed work will shine, but it comes with unique challenges. I’ve led hundreds of projects across academia, industry, and the arts and learned, rather painfully, the value of asynchronous work. These days I use a combination of Google Docs and Notion (though I’m always on the hunt for new tools) to make the act of collaboration tangible. One of the most important rules I’ve learned is, “It must be in the doc.” Side notes and forks22 are absolutely not allowed; it is the collaborative resolution process that creates the value add of asynchronous tools, and so all work on a project must be subject to this group computation. For the same reason, it’s imperative that contributors aren’t passive, waiting to be told their role in some resulting list of action items; ideal asynchronous tools directly touch the finished product. Yet, these tools are also a form of communication. A useful framing is to consider whether a completely new team member could take up your role based only on the tools. They should be able to walk into the project and rapidly join that supermind.

This doesn't mean there’s no role for “meetings'' in distributed work–it can and should be a hybrid of asynchronous and synchronous. Research demonstrates that remote meetings are more effective when preceded by asynchronous discussions. The findings become even more persuasive when we look at the decades of research on remote learning in education.

Remote learning has been an active field of education research since the 80s and 90s. In recent years with the rise of platforms such as Khan Academy, Coursera, and edX, we’ve begun to understand how to create learning experiences that are native to remote technology, rather than simply being filmed lectures. Asynchronous learning is one of the core features of computer-supported collaborative learning.

Asynchronous collaborative learning involves a mix of tools: pre-recorded lectures for self-paced viewing, collaborative wikis, discussion forums, and online learning environments where students can collaborate on assigned problems such as virtual whiteboards and lab spaces. These technologies free students from being yoked to one another’s pace, but more importantly transform learning into a student-centered, active-learning experience. Instead of students passively listening to lectures together and completing synchronous in-class work, instructors use tools to monitor student engagement and progress on disparate problems, giving support for students’ individual experiences as needed. The main challenge is driving engagement with the students and preventing dropout. This is particularly important for students who fall to the periphery of social learning networks.

One study found that while traditional lectures yielded better student performance in fully synchronous classroom settings, when classes go online, asynchronous, problem-based learning produced the best results. The change in the medium from in-person to online qualitatively changed the process of learning. Another study showed, online students that participate in asynchronous discussions perform better than those that only follow class lectures and other traditional teaching practices, and still another found, “...asynchronous peer-to-peer discussion is more effective than traditional classroom lecture-discussion for undergraduate students.”

The research on successful asynchronous learning reveals a shift from a passive teacher-centric to an active student-centric model. Beyond the specific details of asynchronous tools, this dichotomy mirrors the management-driven vs. employee-driven (tightness/looseness) cultural dimension that I have previously identified. Remote learning flipped the hierarchy of the classroom, with teachers supporting individual students rather than leading the class. Remote work has turned traditional work hierarchies on their head as well, with leaders at the bottom acting as support for their employees rather than delegating from the top. It should stay that way.

All of this tells us what distributed work should look like. Distributed managers and instructors are the glue that keeps teams working towards a goal by establishing and role modeling norms through their own creative contributions and preventing dropout, burnout, and disengagement. It’s a shift from one person acting as a bottleneck of information and direction. Managing asynchronous teams also means knowing how to manage asynchronous technologies–it is an integral part of their role in the team. They are the ones that will set the norm for how an entire team uses tools. If they are lazy in their use of the technology, everyone else will follow suit.

Part of the beauty of asynchronous collaboration is its ability to bring balancers and synergists together on their own terms. The tools allow for flexibility in the norms of interaction (though these must still be explicitly agreed upon ahead of time). “I can only be contacted in these hours.” That’s okay if during that time you are engaged with your team’s asynchronous docs and wikis, so that I can contribute on my schedule. “You can contact me whenever, but I will get back to you when I have time, so don’t bug me,” is also transformed by adopting asynchronous tools. With few demands for synchronous interaction, individuals can engage in the way that best integrates with the rest of their lives.