Document View

While the recent focus on knowledge has undoubtedly benefited organizational studies, the literature still presents a sharply contrasting and even contradictory view of knowledge, which at times is described as "sticky" and at other times "leaky." This paper is written on the premise that there is more than a problem with metaphors at issue here, and more than accounts of different types of knowledge can readily explain. This paper suggests that often too much attention is paid to the idea of community, too little to the implications of practice. It is argued that analyses of systemic innovation should be extended to embrace all firms in a knowledge economy, not just the classically innovative.

Introduction

While the recent focus on knowledge has undoubtedly benefited organizational studies, the literature still presents sharply contrasting and at times even contradictory views of knowledge, which is variously described, openly or implicitly, as something "sticky" or "leaky." This paper is written on the premise that there is more than a problem with metaphors at issue here, and more than accounts of different types of knowledge (such as "tacit" and "explicit) can readily explain. Rather, these quite different descriptions of knowledge reflect two different and partial perspectives from which knowledge and organization are viewed. Finding a perspective that can account for both, the task this paper sets itself, thus seems a worthwhile theoretical endeavor.

To this end, we give examples of the contrasting views of knowledge and suggest why research often seems to evade this contrast rather than confront it. We then contend that the divergence is best resolved from the perspective of practice, which can embrace in a single framework the sticky and leaky manifestations of knowledge. We lay out this framework and then explore what it reveals about the external and internal relations of organizations. Externally, practice helps reveal the "embedding" conditions of organizations (Granovetter 1985, 1999). Internally, we argue, practice creates epistemic barriers among the different communities that make up a complex organization. The formal organization's advantage over the market lies in dynamically coordinating the knowledge produced by these communities despite such differences. Thus, we conclude, analyses of systemic innovation (Rosenberg 1994, Teece et al. 1994) should be extended to embrace all firms in a knowledge economy, not just the classically innovative. However, to extend the analysis this way calls for a transformation of conventional ideas of coordination and of the trade-off between exploration and exploitation (March 1991). Throughout this argument, our approach is deliberately informal. In this we hope to follow Winter's (1987, p. 180) intuition that, "without denying the possible usefulness of a formal modeling approach," it is sometimes useful to pursue "informal, looser, and more flexible" investigation.

Conflicting Views and Alternating Approaches

The literature overtly marks the conflict we hope to clarify in descriptions of knowledge as, on the one hand, "sticky" (von Hippel 1994, 1999; Szulanski 1996) and as, on the other, "leaky" (Liebeskind 1996, Wernerfelt 1984) or "mobile" (Hoopes and Postrel 1999). While some argue that there are two types of knowledge at issue here (arguments that we return to below), these contrary accounts of knowledge are nonetheless generally discussed apart. Sticky discussions focus primarily on the challenge of moving knowledge inside organizations. For example, von Hippel (1994) discusses the well-known path of resistance between a firm's research labs and engineering, while Szulanski (1996) examines the stickiness evident in attempts to move the knowledge inherent in "best practice" from one part of an organization to another. In general, then, such discussions focus on the movement of knowledge within the borders of the firm. Discussions of "leakiness" and the like, by contrast, generally focus on the external and (as the term suggests) undesirable flow of knowledge, in particular the loss of knowledge across the boundaries of the firm to competitors. Thus, Liebeskind (1996) sees a firm's competitive advantage as lying in its ability to prevent knowledge from leaking across potentially porous boundaries.

We might expect these distinct analyses of knowledge to converge in studies that try to follow knowledge both within a firm and across substantial borders, yet here too we detect a certain de facto separation. Research into knowledge in multinationals, joint ventures, strategic alliances, and foreign direct investment (for example Gupta and Govindarajan 2000, Hennart and Park 1993, Lyles and Salk 1996, Simonin 1999) often focuses primarily on the "sticky" problem of coordinating knowledge among parts or partners of such extended organizations, leaving aside for the most part "leaky" worries about the potential loss of proprietary knowledge as a result. These worries are taken up in more theoretical accounts of how the rents accruing from its knowledge can be kept within a firm's boundaries (Williamson 1981, 1991; Teece 1986). One perspective, then, helps to explain integration in terms of improving the reluctant flow of knowledge so that it reaches as far as the enterprise's boundaries (Kreiner and Schultz 1993, Kreiner and Lee 1999, Lee 1999); the other, in terms of creating boundaries-"protective governance structures" (Williamson 1981) or "regimes of appropriability" (Teece 1986)-to prevent knowledge from spontaneously overflowing. To the extent that they are thus balkanized, such discussions make it understandably hard to tackle the important task of explaining the boundaries of the firm in terms of knowledge (Kogut 2000).

Nevertheless, the literature does occasionally acknowledge that knowledge can appear to be both sticky and leaky. Two explanations generally follow. The first suggests that there are two kinds of knowledge; one kind is sticky and the other leaky. Thus, for example, Winter (1987) in his discussion of knowledge as an asset argues that "some [knowledge] assets are subject to major hazards of involuntary transfer; others may prove highly resistant to affirmative efforts to transfer them." This typology is often supported by the well-known distinction between performative (and generally nonmarketable, hence sticky) knowledge and declarative (and in principle marketable and potentially leaky) knowledge. The distinction often leads to distinctions of "know how" versus "know that" (for example, Garud 1997), for which the philosophical locus classicus is Gilbert Ryle's The Concept of Mind (1949) or "tacit" and "explicit" (for example, Nonaka and Takeuchi 1995), for which the locus classicus is Michael Polanyi's The Tacit Dimension (1966).

The difficulty for this argument is that, rather than sticking or leaking according to type, exactly the same knowledge can prove both sticky and leaky. Ideas, insights, inventions, and practices that are unable to travel within the organization can prove to be quite capable of travelling to competitors. Such sticky-yet-leaky knowledge helps explain what Zeigler (1985) calls "fissioning"-the process through which people who find their ideas blocked inside one organization form another to give the ideas a more sympathetic treatment. The successive generations that led from the transistor to the modern microchip and, in the process, from Bell Labs through Shockley and Fairchild to Intel and beyond are examples of knowledge (and knowers) stuck within one company fissioning out to another (Lecuyer 2000).

A similar process between established firms can be seen in what might be thought of as "second mover" advantage, where ideas that failed to find a suitable reception in the first mover spread to another company that then makes them profitable. In a discussion of this problem, Teece (1986) offers the example of EMI and the CAT Scanner. More famous, if more anecdotal, is the example of the "Graphical User Interface" (GUI) developed at Xerox PARC, but profitable primarily for Apple (with the Macintosh interface) and later Microsoft (with Windows). Here again, ideas that stuck within the originating corporation leaked out and onto competitors' production lines with apparent ease (Hiltzik 1998, Smith and Alexander 1988). So, in all, those who argue that there are two types of knowledge, while addressing some sticky/leaky issues succinctly but separately, are less helpful in accounting for knowledge which manifests both stickiness and leakiness simultaneously and according to circumstance.

As that last phrase suggests, instead of addressing the inertia of knowledge in terms of inherent properties of knowledge itself, it seems more fruitful to look to the context or environment in which knowledge sticks or leaks. This is the second approach to the sticky/leaky problem, and it generally leads to "social" or "cultural" explanations of why and where knowledge sticks or flows. In such explanations, the firm usually emerges as an especially propitious place. Accounts describe it as a "social" (Kogut and Zander 1996) or "discursive" (Tsoukas 1996) community or a single "community of practice" (Constant 1987, Kreiner and Lee 1999). Similarly, others have suggested it is a unitary "knowledge" (Grant 1996) or "interpretive" (Daft and Weick 1984) system. Allowing questions of collective meaning (Wenger 1998), sensemaking (Weick 1995), trust (Powell et al. 1996), motivation (Kogut and Zander 1992), and related issues that clearly bear on knowledge to enter the discussion, these sociocultural accounts offer much richer explanations than those that treat stickiness and leakiness as properties of knowledge itself. Therefore, they help to explain why the same knowledge may appear sticky and leaky. Nonetheless, sociocultural accounts face difficulties of their own. First, they can suggest a culturalist equivalent of the old economistic "black box" model of organizations, where the inside is somehow free of all the forces at work on the outside. Second, while helping to account for (perhaps without explaining) internal flow, they leave internal stickiness unexplained. Yet as the sticky literature would suggest, it is exactly within the firm that evidence of stickiness seems most perplexing. And third, despite culturalist assurances of cultural discontinuity, knowledge does leak between firms, as the leaky literature confirms. Given these failings, we suggest there is a need not only to look beyond explanations that take knowledge as a well-defined substance, but also to look beyond explanations that take the cultural unity of the firm for granted.

To do this, we propose looking at knowledge and organization through the prism of practice-the way in which work gets done and, we would argue, knowledge is created. Such an approach, we must acknowledge in advance, raises problems of its own. In particular, instead of the balkanization of research, the perspective we present here tends to emphasize (and even exaggerate) the balkanization of the internal units of the firm, for it is, we suggest, at divisions created in practice that knowledge sticks, for these divisions (which inevitably occur within organizations that form around the division of labor) are epistemic in character. In dealing with these divisions, organizations find themselves dealing simultaneously with internal stickiness and external leaks.

Divisions of Practice, Divisions of Knowledge

Sociocultural accounts of knowledge and the firm generally turn on the relationship between individual learning and social identity. Learning is inevitably implicated in the acquisition of knowledge, but it is also implicated in the acquisition of identity. People do not simply learn about; they also learn, as the psychologist Jerome Bruner (1996) suggests, to be. Learning, that is, doesn't just involve the acquisition of facts about the world, it also involves acquiring the ability to act in the world in socially recognized ways. This last qualification, "in socially recognized ways," acknowledges that it is not enough to claim to be a physicist or a carpenter; people, particularly other physicists or carpenters, have to recognize you as such. People become managers or engineers not only by modeling themselves on managers or engineers (Ibarra 1999), but also by gaining the acceptance and recognition of managers or engineers. Learning, in all, involves acquiring identities that reflect both how a learner sees the world and how the world sees the learner. Learning any but the most simple job, then, is a complex social process, one that cannot simply be captured in the notion that "all learning takes place inside individual human heads" (Simon 1991, p. 125). Rather, as Teece et al. (1994) suggest, "learning processes are intrinsically social and collective phenomena" (p. 15).

Such a claim does not deny the integrity of the individual. It accepts, however, that what individuals learn always and inevitably reflects the social context in which they learn it and in which they put it into practice. When learning a job is at issue, this context usually includes the firm as a whole, immediate colleagues, and the relevant discipline or profession (as well as idiosyncratic external social forces bearing on each individual). To different degrees, these shape how individuals come to interpret the world (Daft and Weick 1984) and learn from it. Thus, "absorptive capacity" (Cohen and Levinthal 1990) is more than a function of the intelligence of the individual members of an organization. It is also reflects the way in which organizational context shapes the outlook of those individuals. "Absorptive" explanations of the flow and acquisition of knowledge both within and between firms need, then, to look beyond individuals to the milieu in which they work. Also, they must take into account the relationship between the two, a relationship in which neither one (the individual or the milieu) can simply be reduced to the other.

From this perspective, the critical question becomes at what level to analyze the milieu. In their exploration of the same question, Kogut and Zander (1992) contrast the sociological units of the great nineteenth-century trio Marx, Durkheim, and Weber. Marx gave primacy to class (Marx 1967), Durkheim to the "organic" community (Durkheim 1984), and Weber to the bureaucratic organization (Weber 1978). Kogut and Zander, in step with much organizational and business research, favor the last. Business organizations are undoubtedly a critical source of identity in modern society (Zunz 1990), but accounts can easily exaggerate and homogenize their contribution. As O'Connor (1999) shows, in the thirties, Chester Barnard (1938) saw the business organization as a mechanism for producing cultural homogeneity out of diversity (in short, the firm for Barnard was the engine of the national melting pot). By the fifties, Whyte (1956) and others were denouncing the homogenizing role of the firm, yet in doing so, they nonetheless accepted and even amplified the idea that firms are able to impose a uniform identity on their members. In the seventies, Arrow (1974) in his seminal analysis of the role of the firm in coordinating people and knowledge recapitulates Barnard (and a good deal of American sociology) by introducing the firm as a critical mediator between individuals and society. In the last couple of decades, as we have seen, research has reaffirmed the idea of organizations as cultural units.

However, while accepting that firms may provide some degree of common culture for their members, it seems important to consider, for example, how much a CEO and a technician in a large Fortune 500 company really have in common. As accounts of social networks and occupational communities indicate, both are likely to have more in common with their peers in other organizations than with many of the other employees in their own. Focus on the organization as a cultural unit thus tends to overdetermine the contribution of often quite distant structures to groups of people with quite distinct practices. Certainly, Kogut and Zander (1996) like Tsoukas (1996), Martin (1992), Schoenberger (1997) and, indeed, Barnard (1938) all provide accounts rich enough to acknowledge other "outside" forces bearing on the identity of an organization's members, and in the process resisting and refracting for each individual the pressures towards uniformity coming from the organization as a cultural force. Yet collectively such qualifications still seem caught in the familiar bind laid out by Orton and Weick (1990): "When researchers define organizations as monolithic corporate actors, they overemphasize order and underemphasize elements; when they define them as mere aggregates, they overemphasize elements and underemphasize order" (p. 216). In this case, when they focus on the identity conferred by organizations, claims may overemphasize homogeneity; when they focus on external identity-forming influences, they risk overemphasizing the autonomous and unconnected identity formation of each of an organization's members.

The bind that Daft and Weick (1984) point to suggests a need to consider less distant social forces that, while not bearing on all members of large and diverse organizations equally, nonetheless bear on more than just individual members separately. The choice researchers face, we want to suggest, is not simply between "individual belief and an organizational code" (March 1991) or between individual alienation and organizational man. Indeed, the cultural forces most salient for the members of an organization are probably not those "determined by leaders" (Schein 1985, p. 2) nor even necessarily those espoused by their members. Rather, they are those arising through and at the point of an individual's engagement in the organization and its work. Setting aside, as we do throughout this paper, very small organizations built around a single practice, these forces obviously do not bear on all in an organization, but they do to some degree bear on all engaged in similar jobs, and in particular on those engaged over time and collectively on particular tasks. The identity and knowledge that people acquire when joining an organization, then, while they might appear to be those of an organization as a whole, are more likely to be those of the particular practice through which the individual joins the organization. The technicians Orr (1996) studied, this reminds us, become Xerox members by becoming technicians.

Work practice, then, seems critical to understanding the acquisition of identity and knowledge at work. Looking at learning and identity through the lens of practice shifts attention from production or delivery-from structural or organizational perspectives-to include consumption and the perspectives of participation.1 Turning things around this way helps to reveal that the identities developed through participation will be social if the work is social. They will be dynamic, changing as practice and the community changes. They will be cumulative, as over time the community develops a history of practice, and they will be at least as varied as the distinct practices within an organization. Taking such things into account, practice-focused analysis brings investigations of knowledge and identity in organizations closer to the point at which working life is lived, work done, and so working identities created, than analysis focused either on autonomous self-interest, on the one hand, or on what some cultural theorists recognized as the more abstract and distant sociological "slabs" such as class or organization (Johnson 1983) on the other. The identities visible in this way will reflect the organization's de facto division of labor more than its espoused culture or autonomous interests (and self-interest) of its members. In all, work identities, the perspective suggests, are less something mandated by structure or dictated through culture, and more something that participation actually helps to create.

In this regard, "communities of practice" as set out in Lave and Wenger (1991) and developed in an organizational context in Brown and Duguid (1991) offer a particularly helpful level of analysis for looking at work, learning, knowledge, and work identity formation. These groups of interdependent participants provide the work context within which members construct both shared identities and the social context that helps those identities to be shared. Members of such groups collectively develop an outlook on work and the world that may reflect the organization as a whole, but will most intensely reflect the local community. Within this, because of the shared perspective, knowledge can be readily shared. In proposing the community of practice, then, Lave and Wenger actually come closer to the sociology of Tonneis (1971) than to Marx, Durkheim, or Weber. Tonneis emphasized the importance of the local community (gemeinschaft) in relation to the larger abstractions of Marx, Durkheim, or Weber.2 At the same time, they suggest a solution to Winter's (1987) quest for a midlevel unit of analysis for looking at knowledge.

This unit of analysis helps to amplify underexplored hints within some organizational literature that culture and identity are indeed shared most densely not at the level of the organization (unless the organization is itself little more than a community of practice), but at the level of what Schein (1985) acknowledges are its "subcultures." Though Schein talks little about these subcultures and does not draw on the wide literature on the topic (see Lave et al. 1992), the groups he has in mind are, we suspect, generally more autonomous (and often more subversive) than the espoused cultures he discusses in greater detail. Similarly, the community of practice highlights Cohen and Levinthal's (1990) brief suggestion that absorptive capacity is not simply a function of an organization as a whole, but of its internal "units." The notion also intimates, as intuitively seems more reasonable, that cultural forces might well bind groups of technicians (Barley 1996) or of senior managers (Spender 1989, Lyles and Schwenk 1992, Schoenberger 1997) quite tightly, but probably not both technicians and senior managers together. To the extent that there is an overall culture embracing both, it is likely to be relatively ineffectual compared to the community of practice of each. The general ability to share participants' knowledge between these two practices is correspondingly limited. The perspective of practice thus embraces recognizable divisions of identity within organizations: not only such things as "R&D culture," but also the distinct culture of, for example, technicians (Orr 1996, Barley 1996), engineers (Constant 1987, Almeida and Kogut 1999), or claims processors (Wenger 1998). It also helps to explain the markedly different outlooks of people within an organization with apparently quite similar jobs, such as sales and marketing, accounting and budget forecasting (Arrow 1974), or nurses and doctors (Leonard and Sensiper 1998). Here distinct practices make the communities, and so identities and knowledge, distinct. Therefore, communication between the two can be more problematic than might at first appear.

For a variety of reasons, then, communities of practice seem a useful organizational subset for examining organizational knowledge as well as identity. First, they are privileged sites for a tight, effective loop of insight, problem identification, learning, and knowledge production. As von Hippel (1999) suggests, useful knowledge in organizations is often best developed not by "specialists" detached from a problem, but by those "who directly benefit from a solution," who, we contend, are likely to be the members of a community of practice in which the problem to be solved arises. Second, communities of practice are also significant repositories for the development, maintenance, and reproduction of knowledge. A community's knowledge is not held equally by all, but shared differentially across the community as a whole, though it is made available to all. Joining such a community (in the process Lave and Wenger 1991 call "legitimate peripheral participation") gives access to that community's identity and through that its collective knowledge. Third, community knowledge is more than the sum of its parts. Community members provide for one another social "affordances" (Cook and Brown 1999) that scaffold knowledge creation in practice. Fourth, organizational adaptability (Stark 2000)-the ability to adapt continuously and respond proactively to environmental change-is to a significant degree determined by communities of practice. Members of a community of practice are often simultaneously members of that organization and members of a larger, dispersed occupational group (van Maanen and Barley 1984). Thus, as well as providing a source of locally produced knowledge, the community of practice, as Constant (1987) suggests, also creates a vital link between organizational strategy and changes emerging outside the organization-a point we shall expand on in a moment.3

The idea of communities of practice has been taken up with a remarkable amount of enthusiasm. It nonetheless (or perhaps, for this very reason) needs handling with care, much like the notion of "culture" over which we hesitated above. On the one hand, much of the enthusiasm turns on the appeal of the word community, which Williams (1976) suggests can be a deceptive but "warmly persuasive word." (It is worth contemplating how wide the notion would have spread had Lave and Wenger decided to talk about a cadre or commune of practice.) Communities of practice are, in fact, as likely to be cold as warm, may sometimes be coercive rather than persuasive, and are occasionally explosive. Nevertheless, mediating as they do between individuals and large formal and informal social structures, and between organizations and their environment, they are where a good deal of the work involved in knowledge creation and organizational learning gets done.

The appeal of community has tended to obscure the importance of practice, hence our desire to redirect attention here. This notion, however, brings its own problems. In the first place, the ambiguity of the term allows practice to signify both work itself (the practice of a medical practitioner, for example), and rote tasks or exercises designed to help learn to work (as in piano practice). Let us simply assert that by practice we mean, as most theorists of practice mean, undertaking or engaging fully in a task, job, or profession. Second, and more problematically for those trying to build an organizational account, practice theory, whether from the right (in the hands of, for example, Hegel, Hayek, or Oakeshott) or from the left (in the hands of Marx, Bourdieu, or de Certeau), tends to remain at best neutral and at worst hostile to institutional and organizational theories. Consequently, in theory, the community of practice sometimes appears indifferent to other forms of social alignment, resembling a sort of social monad-a fundamental building block whose articulation with other better-known structures is hard to see. Finally, while the idea of community may comfortingly suggest that organizations are significantly culturally homogeneous, practice, as we have noted, uncomfortingly suggests they are to a significant degree divided, riven by practice even as that practice provides participants with their particular kind of organizational identity. Indeed, if regarded as a community at all, any but the smallest organization should probably be regarded as a "community of communities of practice" (Brown and Duguid 1991, Tuomi 1999).

Practice and Knowledge

So far we have made several claims about practice, knowledge, and communities, but done little to yoke them together theoretically, and we must acknowledge that the three make uneasy companions. Lying in many accounts on opposite sides of the standard division between theory and practice or mental and manual labor, knowledge and practice make something of an odd couple. Similarly, knowledge and community seem to violate the conventional idea noted already that knowledge belongs just to individuals.

Let us begin, then, with the issue of knowledge and practice. The distinctions of Ryle (1949) and Polanyi (1966), being widely cited in the literature, offer a good place to start and allow us to again take up the issue of sticky and leaky knowledge as two types of knowledge.4 Polanyi's most quoted line is probably "We know more than we can tell" (Polanyi 1966, p. 4). This unspeakable knowing is what Polanyi deems tacit as distinct from explicit. His distinction is thus often used to justify the idea that there are two kinds of knowledge, one tacit and one explicit, hence one that sticks and one that leaks. The distinction also suggests that translating the inarticulate form into the explicit form, turning sticky knowledge fluid, would be a way to promote required fluidity (Nonaka and Takeuchi 1995).

Yet, while many discussions of Polanyi end with this point about knowing more than we can tell, for Polanyi it was only a beginning. "I shall consider human knowledge," he writes, "by starting from [this] fact" (Polanyi 1966, p. 4). He spends the rest of the first Terry lecture contemplating the relation of this inarticulate knowledge to what can be articulated, and he concludes that knowledge always has an inarticulate component. This component Polanyi calls the tacit dimension-a point sufficiently important that he made it the title of the published lectures. He is not, then, arguing for two types of knowledge, merely for two dimensions-two interdependent dimensions, it turns out, for the explicit dimension of knowledge relies on previously "interiorized" (in Polanyi's terms) implicit or tacit dimension. Attempts to shake off this tacit dimension, buried as it is in personal identity, are in Polanyi's view at best futile, at worst counterproductive.5 Though knowledge undoubtedly can be usefully articulated and explicated, in use the explicit nonetheless always possesses this other, implicit dimension.

Polanyi's tacit/explicit distinction echoes Ryle's famous contrast between know how and know that. Like Polanyi, Ryle insists that these are not independent types of knowledge. They are interdependent and cannot be reduced to one another. Knowing how Ryle insists "cannot be defined in terms of knowing that" (Ryle 1949, p. 32). Acquiring know that does not lead to being able to use it. Knowing the rules of chess, in Ryle's running example, does not tell you how to play chess. Know that may be both explicit and free flowing, but from Ryle's perspective it is neither actionable nor useful on its own. To make know that useful requires appropriate know how, something thus very similar to Polanyi's tacit dimension. Know how, moreover, is not acquired like know that, which may circulate as precepts and rules. It is, Ryle insists, quite different. "We learn how," he argues significantly, "by practice"-by, in Ryle's example, playing chess. Polanyi makes a similar claim when he argues that "comprehension," which is for Polanyi the acquisition of knowledge from another, is "both intellectual and practical" (Polanyi 1966, p. 48). In both of these well-known arguments, then, knowledge is two-dimensional and practice underpins its successful circulation.

Examples of butchers and midwives (Lave and Wenger 1991), photocopier repair technicians (Brown and Duguid 1991), or flute-makers (Cook and Yannow 1993) may make it seem that these strictures apply only to a special kind of knowledge-to "practical knowledge," knowledge that is distinct and separable from the more refined, cerebral knowledge, and so to manual labor but not to mental, to use a conventional workplace distinction. Ryle categorically refutes the notion. "Theorizing," he argues, is simply "one practice amongst others" (Ryle 1949, p. 26). Schon's (1983) work on professional practice also suggests that a practice perspective undermines conventional assumptions about the mental-manual divide, while Barley (1996) also rejects such a divide.6

So, Polanyi's and Ryle's arguments militate against the idea that knowledge circulation involves explicit knowledge alone. From the idea that tacit knowledge is "nontradable" and needs to be converted into explicit form to circulate, we come instead to the idea not only that conversion (if it involves uprooting knowledge from the tacit) is problematic, but also that tacit knowledge is required to make explicit knowledge usefully tradable or mobile. Only by first spreading the practice in relation to which the explicit makes sense is the circulation of explicit knowledge worthwhile (Cook and Brown 1999). Knowledge, in short, runs on rails laid by practice. This is not such a fragmentary conclusion as it may seem. Simply by being human, we all engage in a great deal of similar practice, and hence share a great deal of tacit understanding. People do share knowledge and insight by virtue of their membership in those overarching sociocultural "slabs." Nonetheless, much of the practice that forms identity and gets work done is more local and more dynamic.

The need to share some practice to be able to share new ideas reveals the challenge of workplace communication and coordination. Anthony Giddens (1990) addresses this point in terms of "disembedding" and "reembedding" knowledge. He suggests that as technologies increasingly allow people to communicate across space and time, as knowledge is disembedded in one place to be reembedded in another, the critical question concerns the degree to which the embedding conditions at both ends of the communication are similar. To the extent that they are, then communication and coordination are likely to be successful. To the extent that they differ, communication and coordination are likely to break down, and we are suggesting that the same challenge presents itself in attempts to communicate not across distance, but across different communities developed around different practices. Distinct practices create distinct embedding circumstances. Therefore, to understand where knowledge flows and where it sticks we need to ask where and why practices (and so embedding circumstances) are common, and where and why they are not.

Practice, Community, and Networks

First, however we need to address the question, if, as Polanyi and Ryle argue, there is a practice component in all knowledge, is there necessarily a social component? Certainly, Polanyi and Ryle establish no explicitly social connections. Spender (1996), however, uses Vygotsky to reveal the social entailments of Polanyi's argument. Moreover, it seems reasonable to argue that if people share a practice, then they will share know how, or tacit knowledge. So, as communities of practice are defined by their communal practice, they are likely to have communal know how developed from that practice. If shared know how or tacit knowledge make it possible to share know that or explicit knowledge effectively, then such communities, sharing common embedding circumstances, will also be effective at circulating explicit knowledge. Empirical studies (Barley 1996, Hutchins 1991, Orr 1996) suggest that communities of practice do indeed work this way, holding and sharing knowledge collectively because of their shared practice. Orr's example is particularly interesting because the technicians he discusses are presumed to work alone. In fact, they rely heavily on the collective knowledge of the group. While this knowledge is distributed unevenly across the group, members can share it because of their common base of tacit knowledge, or know how? The practice-based, tacit dimension of knowledge, then, is clearly implicated in the stickiness and leakiness of knowledge, for shared practice demarcates the extent to which knowledge can spread. Its role is not, we should emphasize again, confined to less cerebral kinds of knowledge. It would seem both to underpin and (where absent) to undermine the circulation of apparently self-sufficient and replicable scientific knowledge in scientific communities (Ziman 1967).

Work on such communities also helps to add an extra dimension to understanding the social reach of knowledge. Strauss's (1978, 1982, 1984) sociology of academic practice indicates that practice does not only bind small, tight communities together. It also allows extensive academic disciplines, most of whose members will never know one another, to form and communicate. Strauss refers to these as "social worlds," but their relation to knowledge is perhaps better understood through Knorr Cetina's (1999) term "epistemic cultures." Knorr Cetina's ethnography-based analysis is particularly valuable because it explores the extensive flow of knowledge across such communities that "create and warrant knowledge." Knorr Cetina's two examples, microbiologists and high-energy physicists, reveal how, where practice is common, communication can be global. Scientists within each of these disciplines can communicate and collaborate with colleagues around the world, for as a result of the practice common to all members of the group, the core of their embedding circumstances is widely shared. Of course, communication is not uniform across these global cultures. Small subsets of disciplines, working closely together and so within what Knorr Cetina calls the "hum" of "continuous reporting among participants"-in short, communities of practice-develop distinct ways of working that may simultaneously create conditions for effective local communication but barriers to global communication. Consequently, new practices and the knowledge emerging from them will not spread as readily or as widely as insights built on settled knowledge. In his analysis of communities of physicists, Collins (1975) notes the sorts of barriers new practice can make to the spread of apparently replicable scientific knowledge:

It's very difficult to make a carbon copy [of someone else's gravity wave detector]. You can make a near one, but if it turns out that what's critical is the way he glued the transducers, and he forgets to tell you that the technician always puts a copy of Physical Review on top of them for a weight, well, it could make all the difference. (Collins, 1975, quoted in von Hippel 1994 p. 431).

In recognition of the difference between local and global scientific practice, the physicist Ziman (1967) regards making the emerging local knowledge of particular groups accessible to others within the broader epistemic culture a critical and challenging part of doing science. Ziman calls the knowledge thus produced "public knowledge" (as distinct from more local knowledge), but he is clear that this "public" involves the particular scientific community at issue, and not the public at large.

Ziman's public is essentially Knorr Cetina's epistemic culture. As Knorr Cetina shows, different scientific practices produce quite distinct epistemic cultures, and hence the sort of knowledge that might flow readily within one culture will not flow uninhibitedly between two. The distinctive practices that internally bind the epistemic cultures of microbiologists and the physicists externally divide them from one another simultaneously. The two together do not form an epistemically homogeneous "scientific community" any more than, as we noted earlier, technicians and senior management within an organization form a single culture. To return to Giddens's terms, the embedding circumstances of each community are quite distinct, making the process of disembedding and reembedding problematic.

There is no transparent language in social theory, but given the problematic ambiguity of Strauss's world, Knorr Cetina's culture, and Ziman's public, we prefer to think of these loose epistemic groups as "networks," while acknowledging that that term brings its own difficulties.8 Reflecting what binds these networks together and enables knowledge to flow within them, we call these extended epistemic networks "networks of practice." Practice creates the common substrate. With the term network, we also want to suggest that relations among network members are significantly looser than those within a community of practice. As Strauss indicates, and unlike in communities of practice, most of the people within such a network will never know, know of, or come across one another. And yet they are capable of sharing a great deal of knowledge.

Within the class of network of practice, we would include such things as the 24,000 technical representatives working for Xerox. It is hard to conceive of this as a single community of practice, though the much smaller subset which Orr (1996) studied clearly is one. To generalize, the people in a particular job category within a large organization do not necessarily make a single community of practice. The category may comprise several communities of practice which together form part of a larger network, which may extend not only across the company, but beyond it. The network of Xerox reps is itself only a subset of a larger network of photocopier repair technicians in general, many of whom work for other companies. Similarly, as we have indicated, we include under the rubric academic disciplines which, as Alpert (1985) and Strauss have argued, join departments from universities around the world to make up groups whose members, to the extent that they have common practices, are able to read and understand one another's work. Disciplinary networks of practice cut across heterogeneous organizations, including, for example, universities, think tanks, or research labs.9 Professions make up still other such networks of practice, where again similar practitioners, by virtue of their practice, are able to share professional knowledge through conferences, workshops, newsletters, listserves, Web pages, and the like. And as Spender (1989) suggests, managers form networks that extend well beyond their own organizations to include similar managers in other organizations. Despite being competitors, managers, as members of a network of practice, have extensive shared practice leading to extensive shared know how, which in turn allows extensive circulation of managerial knowledge. Using conventional spatial images, this argument suggests that different networks of practice cut horizontally across vertically integrated organizations and extend far beyond the boundaries of the latter. Along these networks, knowledge can flow.

Spender's notion of "industry recipes," ideas about management practices that networks of managers share and circulate, helps to emphasize the process of communication within these networks. As his analysis suggests, these "recipes" must first be disembedded from the local. This is not an easy task, as those who try to turn their local management practice into a general management guide know.10 Knowledge thus disembedded must also be reembedded in the local conditions of those hoping to use it, as everyone who uses a management guide wisely realizes. Whatever the recipe, as Spender's work suggests, you need first to be part of the practice-to extend Spender's metaphor, to be a cook-to understand it (as any novice who tries to use a cookbook quickly discovers).

Such networks then spread knowledge among practitioners, crossing the boundaries of particular organizations and following routes prepared by practice. Along these, knowledge will often leak to the extent that common practice has prepared the way (Powell et al. 1996, Kreiner and Schultz 1993, von Hippel 1999). To return to the example of the Xerox GUI, the knowledge developed at PARC leaked to Apple, where similar researchers were engaged in similar pursuits (Smith and Alexander 1988). As Walker et al. (1997) argue, the structure of these types of network is more than a description of the "flow of information" (as Casson 1997 would suggest), but rather "an expression of social knowledge."

Consequently, while these networks are, as Knorr Cetina argues, potentially global, they will tend to have a local character. While similar communities in different firms will take different approaches to common problems (Constant 1987), where such firms are close, the contacts among people in the network dense, and the movement of individuals (as bearers of practice, rather than as bearers of knowledge) from one firm to another (Almeida and Kogut 1999) frequent, practice is likely to be sufficiently shared to let specialized knowledge flow. Hence, areas of local network density help to explain "regional advantage" (Marshall 1890, Saxenian 1996) and the sort of institutional framework that supports such advantage (Lynn et al. 1996).

Stickiness, Leakiness, and Practice

From the perspective of practice, then, it is possible to understand the flow of new knowledge into and out of organizations. A community of practice and its members stand at the intersection of the organization and the network. As such, they lie within two dominant strands of social relations. One strand, bound together by organizational membership, extends along the value chain of the firm, but crosses its divisions of labor (and practice). The other, bound together by practice, orthogonally crosses the value chain but extends beyond the organization along the network of practice. We can enumerate several implications of this argument.

First, as knowledge will travel along the networks built by practice, it will quite understandably leak between the firms along these conduits. Second, as a result, a firm's boundaries are unlikely to encompass all its knowledge and, when informal links are taken into account, distinctions between "stocks" and "flows" of knowledge (Decarolis and Leeds 1999) may be hard to draw. Co-location of organizations from a particular industry gives rise to fine differentiation of practice (Porter 1998, Kenney and Von Burg 1999) creating complex "ecologies" of knowledge (Brown and Duguid 2000). Third, given the role of networks, trying to stop knowledge at organizational boundaries can be a very difficult task. Moreover, aggressive attempts to prevent leaking may be counterproductive. As a result of the two strands of identification-one in the organization, one in the network-members of a particular community are likely to have divided loyalties as Arrow (1974), Alpert (1985), Constant (1987), Knorr Cetina (1999), and Saxenian (1996) have all noticed in their different ways. Which strand communities favor or disfavor will have a significant effect on the direction in which knowledge flows. Aggressive organizational attempts to inhibit network relations are paradoxically likely to reinforce network identification, which may grow at the expense of loyalties to the organization. Fourth, moreover, trying to stop knowledge at these boundaries may be strategically counterproductive too. The lines that let knowledge leak out, also let it flow in. Firms trying to block leaks risk cutting themselves off from essential knowledge (Saxenian 1996, Mounier-Kuhn 1994). In all, strategic decisions about knowledge need to take into account the firm's degree of embeddedness (Granovetter 1985). In this regard, however, it is important to note that networks of practice may also inhibit the flow of knowledge. As Lynn et al. (1996) show, professional networks will occasionally work to resist the spread of ideas felt to be inimical to the interests of the network's members.

Organizing Communities, Overcoming Stickiness

We have used the notion of networks of practice to explain leakiness. This is not, we have suggested, simply an inherent property of some kinds of knowledge. It does not result from making knowledge explicit and, so, tradable. It is rather a function of the common underlying practice, which creates social-epistemic bonds. Where practice doesn't prepare the ground, knowledge is unlikely to flow. As the CEO of Chapparal Steel told Leonard-Barton (1995), he can tour people through the plant, show them almost "everything and we will be giving away nothing because they can't take it home with them."

So stickiness, rather than being the property of a different kind of knowledge and subject to different explanations, can now be explained in the same terms. If knowledge leaks in the direction of shared practice, it sticks where practice is not shared. People with different practices have different assumptions, different outlooks, different interpretations of the world around them, and different ways of making sense of their encounters. Such differences challenge the fluidity of networks, where as we have shown newly emerging knowledge is hard to circulate until the practice underlying it becomes common beyond its originating community. Inevitably, these epistemic differences present an even greater challenge to organizations. As we have noted, while networks cmbrace people with a core of common practices, organizations, by contrast, deliberately embrace communities with fundamentally different practices, presiding as most do over a particular division of labor, and hence of practice and knowledge.

Ultimately, an organization's competitive edge, and hence survival, depends on its unique ability to coordinate knowledge across these divisions (Hoopes and Postrel 1999) better than both its competitors and the marketplace. However, the tendency to understand coordination principally in terms of individuals and organizations and not, as we are contending here, from a perspective that embraces the practice and communities, makes the challenge of coordinating knowledge more difficult. Ignoring practice, we suspect, is likely to exacerbate the internal stickiness that inevitably arises from the internal division of labor, to promote counterproductive outward leakiness, and above all to contribute to core rigidity (Leonard-Barton 1995).

The complementary character of the divisions in a firm has been well explained in the context of invention and systemic innovation (Schumpeter 1947, Rosenberg 1994, and Teece et al. 1994). Rosenberg and Teece et al. argue that innovation-making inventions marketable-often involves building coherent systems of complementary knowledges. In addition, they suggest that formal organization is often uniquely well adapted to this sort of dynamic coordination of complementary, inchoate knowledge. As outsourcing, technology, and the falling transaction costs that technology produces strip away those aspects of knowledge and routine that are better dealt with elsewhere, knowledge production and coordination will increasingly dominate the critical assets of firms. As a result, the charge of building and continuously developing such knowledge complementarity will fall to all firms, whether canonically "innovatory" or not.11 The knowledge arising from all of its different communities-as they interact with their part of the firm's environment, develop local solutions to their problems, and draw in knowledge from their network connections-will become increasingly important and firms will have to engage in a dynamic balancing act. Consequently, conventional ways of dealing with the potential autonomy inherent in each community of practice in a firm as it follows its particular practice will become increasingly inappropriate.

The conventional approach has been, in essence, to use the firm's hierarchical control to limit disruptive local change and so ensure coordination. Certain communities or areas within the firm receive the right to search for new knowledge (Nelson and Winter 1982). The others are strictly regimented and expected to follow routine. This strategic distinction (which in many ways follows the old mental/manual division we mentioned above) can be seen in Lecuyer's (2000) description of Fairchild. That company's founders ranged wide in their own intellectual exploration of the integrated circuit, but they specified in extraordinary detail the activities of all other employees and acted aggressively against anyone who did not follow specifications minutely. Routines inevitably shift workplace activity from practice as preeminently an adaptive, improvisational, knowledge-producing propensity of the sort we have been describing to practice as principally repetitive rote behavior of the sort to which the same term can also refer. However, as Stinchcombe (1990) suggests, there is always a little of the first kind of practice in routines, however regimented. Whatever insights or inventions such practice may generate will almost inevitably be lost to the organization. The organizational choice at issue here has been characterized by March (1991) as a trade-off between exploration and exploitation, by Demsetz (1988) as between specialization and coordination, and by Sorensen and Stuart (2000) who sum it up as innovative processes versus organizational learning. The former in each pair represents the search for new knowledge, the latter the increasing refinement of existing knowledge, with significant sunk costs (Arrow 1962) that make change costly. In effect, these arguments suggest, organizations conventionally seek a static balance between these antinomies by suppressing organizationwide ramifications of the sort of spontaneity important to knowledge work through most of the organization.

Of course, to survive, even hierarchical control must be adaptable and often is. Adler et al. (1999) give a description of organizational flexibility at Toyota' s NUUMI plant which shows how effective it can be. Effective hierarchical control is also evident in Microsoft's extraordinary achievement of turning the entire organization towards the Internet in the mid-1990s (Cusumano and Yoffie 1998). But, as the latter case reminds us, this radical adaptation decreed from above was necessary because noises from below, from those in the firm who "got" the Internet much earlier, could not be heard for what they were-signs of profound environmental change. As organizations shift away from routine coordination and towards innovative, continuously developing practice, such local insight into the environment will be too valuable for firms to ignore or to stifle. Instead, it must be able, at least potentially, to effect "from below" not just the practice of specific communities, but the organization's overall processes and strategies. As structuration theory (Giddens 1984) applied to the firm (Yates and Orlikowski 1992, Barley and Tolbert 1997) might suggest, communities of practice will become ubiquitous sources of knowledge driving organizational change. In such circumstances, there will no longer be a simple choice or static balance between exploration and exploitation, or spontaneity and structure, but a constant and pervasive need to dynamically balance and coordinate the two throughout the organization. Ubiquitous pressure for change presents a profound challenge to coordination and structure. Thus coordination is likely to be highly unstable, the structure always under construction and always under threat.

Consequently, strategic coordination is likely to appear rather different from heretofore. In the first place, as we have stressed, the elements to be coordinated are not just individuals, but communities and the practices they foster. Second, as a result, focus on trust (Kramer and Tyler 1996) or individual motivation (Kogut and Zander 1992) is unlikely to achieve organizational coordination on its own. Both of these are undoubtedly important to organizational stability and success, but alone they suggest that the critical divisions in an organization lie between the individual and the organization and come down to matters of individual alienation or self-interest versus the collective culture and interests of the organization. Inevitably these contrasts, important though they are, gloss over the epistemic differences that we argue are inherent in the organizational division of labor, and present quite a different challenge. Certainly, trust and motivation might help overcome epistemic differences. However, as Sitkin and Stickel (1996) suggest in their analysis of the "quality movement," ignoring epistemic differences that distinguish different communities within a firm can actually serve to damage trust quite significantly.

On this interplay of epistemic differences and issues of trust and coordination, we would also cite an example from our own experience at Xerox. The Xerox research center developed a novel imaging technology and sought to move it out of the labs and into production. However, the technology found no sympathy with the engineers, who compared results of tests on key parts to the critical parameters for apparently similar parts on other print technologies and found the new device wanting. Negotiations between the two groups were all but abandoned as trust collapsed. The problem, it eventually emerged, was that the critical parameters for the new parts actually bore little relation to those for their equivalents on conventional print technologies. Though the result of tests on both might have looked much the same, simple comparison of a set of parameters did not produce a coherent set of results. Some measures essential to the engineers' understanding of the existing technology and manufacturing process just did not apply to the new technology. The example fits nicely with Constant's (1987) argument that communities of practice are keepers of a particular tradition and for engineers in particular a "tradition of testability sustains and defines, almost tautologically, the specific tradition" (p. 226). To generalize, while two communities may and often do acknowledge a single set of measures as a common boundary object (Star and Griesemer 1989), one, as this example suggests, may unilaterally change its own "tradition of testability" or comparable orientation to a boundary object in the process of developing new ideas, but in so doing may unwittingly disrupt communication, coordination, and even trust between itself and other communities hitherto managed through that boundary object.

In all, the example suggests to us that for organizations to take the evolution of knowledge and practice seriously, to gain the advantages that accrue from overcoming internal stickiness while simultaneously preventing disaffection from promoting external leakiness, it is important to address the epistemic-social issues of different kinds of knowledge as well as the psychosocial ones of different levels of trust and motivation. Of course, epistemic rifts within organizations are rarely Kuhnian gulfs with wholly incommensurable practices on either side. Usually, communities that work together share some common practices. Loss of synchronization may be common. Complete breakdown because of mutual incomprehension is rare. (As we noted at the outset, while bringing intercommunal differences helpfully to light, the practice perspective tends to exaggerate the internal balkanization of organizations.) For example, in almost all organizations, career changes of one sort or another will often take people out of one community and into another, leaving them with an understanding of both. Such people are then in a position to become translators or boundary spanners between the two. These sorts of connections often make the informal networks within a firm (like those without) much more important for spreading practice than the formal ones to which so much time and attention is given.

Moreover, between such communities there are usually many boundary objects. Such things as shared documents, tools, business processes, objectives, schedules, and the like not only serve to coordinate, but to record and signal changes in one community's practice. In so doing, they can make explicit the need for negotiation between the two over their diverging practices and prompt both sides to reconsider their practices and presuppositions. This way, coordinated communities can actually transform the notion of "single-" and "double-loop learning" (Argyris and Schon 1974) from "interpersonal behavior" (Schon 1987) to intercommunal behavior, with one community pushing another to reconsider its underlying assumptions. This intercommunal dynamic is, then, one of the advantages of the coordinated but adaptive division of labor.

The challenge of stickiness, hitherto primarily dealt with through the imposition of routines, changes in organizations whose primary asset is the ability to develop knowledge continuously. For such organizations, the most important relationship between quasi-autonomous communities within a dynamically structured firm must be one of negotiation-negotiation that allows change, on the one hand, to occur locally and, on the other, to affect global strategic change if necessary.12 The idea of intercommunal negotiation may seem like an anodyne or vague summation of such a lengthy analysis. Still, it is on the one hand profoundly different from conventional organizational coordination, and on the other capable of creating dynamism in the organizational structure adequately responsive to the dynamism within creative communities.

Conclusion

Starting with the paradox of sticky and leaky knowledge, we have attempted to find a perspective that allows the paradox to be resolved, and we have claimed that the perspective of practice does the task. This perspective supports a sociocultural view of learning and knowledge, but one that challenges conventional views of the internal homogeneity of the firm. In contrast, we have stressed the internal divisions and external connections highlighted by practice. These internal divisions, communal rather than individual in origin, help explain stickiness, while the external connections help explain leakiness. Together they suggest that a firm's knowledge base is not a property that falls within its boundaries, but one that in part draws on its embeddedness in broader structures. Consequently, knowledge may more easily flow out of a firm than move productively within it. Ultimately, the greatest benefits to knowledge accrue from coordinating its development across the division of labor, coordination often better achieved within the structure of the firm than in the marketplace. We conclude that coordinating a firm around knowledge and practice, a dynamically demanding strategic task, is quite different from coordinating it around more conventional routines, and helps to uncover the innovative potential and challenge in all firms.

Acknowledgments

The authors are grateful to Mario Bernassi, Martin Kenney, Bruce Kogut, Kirstina Lee, Sid Winter and two anonymous referees for help with drafts of this paper.