(FORTUNE Magazine) – If you kept a copy, you've forgotten where you put it: The document you signed on your first day at work, in which you promised not to reveal company secrets and agreed that the fruits of your labors--ideas, intellectual property, whatever--belong to your employer, not to you. It probably contained language like this:

"In consideration of my employment by Random Rightsizing Inc., I hereby assign to the company all inventions or innovations conceived or developed by me during my employment; I further agree not to use or disclose any confidential information I may receive as a result of my employment by the company..."

Swiping secrets is odious to both law and etiquette, and that's a legally enforceable document. It is also hornswoggle. First, you swap proprietary information with outsiders all the time; in fact, the company probably wouldn't prosper unless you did. Second, the real genesis and ownership of ideas and know-how aren't corporate. Nor personal, for that matter. They belong to something that is becoming known as a community of practice.

If the term "community of practice" wasn't invented at the Institute for Research on Learning, that's where it's most often bandied about. IRL, in Palo Alto, was founded in 1987 as a sort of eleemosynary spinoff of Xerox's Palo Alto Research Center. Its mission--to study how people learn--makes it a center for basic research for the Information Age. The fundamental finding in IRL's work is that learning is social: However romantic the image of the scholar bent over his desk in a pool of lamplight, learning happens in groups.

That's an insight with huge--and problematic--implications for managers. Not every group learns. You can't take a dozen people at random, give them a pot of coffee and a box of doughnuts, and expect them to learn something. Groups that learn, communities of practice, have special characteristics. They emerge of their own accord: Three, four, 20, maybe 30 people find themselves drawn to one another by a force that's both social and professional. They collaborate directly, use one another as sounding boards, teach each other. You can't create communities like this by fiat, and they are easy to destroy. They are among the most important structures of any organization where thinking matters, but they almost inevitably undermine its formal structures and strictures.

Communities of practice are the shop floor of human capital, the place where the stuff gets made. Brook Manville, director of knowledge management at McKinsey & Co., defines a community of practice thus: "a group of people who are informally bound to one another by exposure to a common class of problem." Most of us belong to more than one, and not just on the job: the management team; the engineers, some in your company and some not, trying to cram more circuits onto a wafer of silicon; the church choir. Trying to solve a problem for one community, his church choir (how could they mark the day's hymns without damaging the hymnals by dogearing or paper-clipping pages?), led tenor Arthur Fry to conceive the product that became Post-it notes, developed by another community, adhesives experts at 3M.

IRL's Etienne Wenger, who is completing a book about communities of practice, points to several traits that define them. First, they have history--they develop over time, indeed "you can define them in terms of the learning they do over time." Second, a community of practice has an enterprise but not an agenda; that is, it forms around a value-adding something-we're- all-doing. It could be a gang seeking to carve a place for itself on the streets, or a district sales office wanting to be the best damn office in the company; it could be people who don't work side by side but share a mission, like antitrust lawyers, Alcoholics Anonymous groups, or (an example several times described by Xerox PARC's chief, John Seely Brown) copier repairers who exchange tips at the water cooler. Third, the enterprise involves learning; as a result, over time communities of practice develop customs, culture--as Wenger puts it, "a way of dealing with the world they share."

Perhaps most intriguing, communities of practice are responsible only to themselves. No one owns them. There's no boss. They're like professional societies. People join and stay because they have something to learn and to contribute. The work they do is the joint and several property of the group--cosa nostra, "our thing."

These traits give communities of practice a distinct place in the taxonomy of the informal organization. Teams have a charter and report to a higher authority: Even if they have no box on an organization chart, they have an agenda, a deadline, accountability, a membership list. A community of practice is voluntary and longer lived, and has no specific "deliverable" like a report or a new gadget. Affinity groups and clubs--the salesmen who play poker every Friday night--are about fellowship rather than work. Grapevines and other networks may support your work but are not central to it.

Like Poe's purloined letter, communities of practice are so obvious that we don't usually notice them. When we do, however, the implications are anything but obvious. James Euchner, a vice president in Nynex's research and development department, began thinking about them when he was puzzling over why some groups at Nynex were quick to adopt new technologies, others not. For example, some groups needed, on average, 17 days to set up data services for customers. Euchner hired an anthropologist to learn why. She found that different departments involved in the process never communicated informally and, as a result, didn't understand one another's roles and needs or solve problems together. When she and Euchner put the workers together in the same room, they created an environment that allowed informal groups to form around various tasks, which soon grew into a full-fledged community of practice. Result: a mutual sense of purpose and a sharing of ideas that cut the time to provision data services to just three days.

Euchner found himself face to face with a challenge communities of practice pose: Organizational learning depends on these often invisible groups, but they're virtually immune to management in a conventional sense--indeed, managing them can kill them. A study by three academics--Ronald Purser of Loyola University in Chicago and William Pasmore and Ramkrishnan Tenkasi of Case Western Reserve in Cleveland--shows why. The professors followed two product-development projects in the same big American manufacturer. One, a major upgrade of a key technology, was rigorously managed and relied on big fortnightly meetings to keep everybody up to speed; the other, a radical innovation, was scarcely managed at all: The professors called it "self-organizing...informal...egalitarian." The former slogged; the latter soared; the main reason, Purser et al. found, was that the formal structure of the first group inhibited the informal exchanges learning depends on.

If you can't manage communities of practice, managers can still help them. How? First by recognizing them and their importance. They're fairly easy to spot in a department or business unit, like those copier repairers; harder to see are communities that cross lines. Look for jobs that exist in several functions, business units, or geographic areas, suggests George Por, head of a Santa Cruz, California, firm called Community Intelligence Labs, who has helped Intel and Dow Chemical support communities of practice. Plant or office managers, sales reps, metallurgists, tech weenies--from Abilene to Aberdeen, these have common enterprises and would benefit from closer contact.

Managers can do a lot for these groups. Communications systems are usually laid out along existing departmental lines. Communities of practice don't need many resources: Let them build an intranet, use your conference room, put a get-together on the expense account. There are big benefits from linking people who may unknowingly duplicate one another's efforts or walk away from projects too big to tackle single-handed. At National Semiconductor, communities of practice have been officially recognized by the company and given semi-official roles; one group, for instance, reviews microchip designs, adding a cross-functional perspective that business-unit leaders might lack. Says Skip Hovsmith, who directs National's mobile-computing research: "We're decentralized, but we had to share and collaborate, pressured by the convergence of markets and shorter product lifetimes. Communities of practice are the bridge."

Fertilize the soil, but stay out of the garden. Says Valdis Krebs, a Los Angeles consultant who helps solve organizational design problems by mapping networks to reveal how work really gets done: "Fund them too much, and you'll start to want deliverables. It won't work. You'll get what the community wants to deliver." That's because it is motivated by its enterprise--this thing we're all learning about. For its members, boundaries exist to be crossed, just as mountains exist to be climbed. Information wants to be free.

That's the subversive part. Says Stephen Barley, Stanford professor of industrial engineering: "As communities of practice proliferate, occupational principles begin to compete with administrative principles." A person's responsibilities to the communities of which he is a member sometimes conflict with each other, and with the rules and interests of the companies he works for.

It happens all the time. Watch a bunch of scientists at a convention: They swap secrets like street vendors opening their jackets to flash contraband Rolexes. In the late 1980s, Eric von Hippel, a professor at the Sloan School of Management at MIT, studied how manufacturing-process engineers in the steel mini-mill industry traded proprietary information even with direct competitors. They withheld big scoops and were unlikely to talk to rivals that they figured didn't have much to offer in return. But with so much to learn in their relatively young business, the steelmakers--from companies like Nucor and Chaparral Steel--evidently figured that sharing secrets was a fair price for progress. Says von Hippel: "It happens everywhere. Famous chefs trade recipes--'But not my mousse, I will never give away this secret.' The standard corporate view is you're giving away the store, but the fact is, if others are cooperating and you decide not to, you fall behind."

The community of practice belongs to itself. Competitive advantage comes to the companies best able to spread the human capital forged in those intermural and intramural places and act on it. Says Nynex's Jim Euchner: "Boundaries don't just keep information in. They keep it out too." REPORTER ASSOCIATE Victoria Brown

Join Tom Stewart in the FORTUNE FORUM ON COMPUSERVE (GO FFORUM) or by E-mail: