Dana Okafor runs talent enablement at Meridian Health Systems, a 2,500-person company with 22 recruiters spread across four regions. Her organization rolled out an AI screening and outreach platform last spring, and the launch training went well: every recruiter sat through a two-hour workshop, passed a short certification, and left with a one-page cheat sheet. Three months later, Dana pulled the usage logs and found the pattern she feared. Four recruiters were using the tools fluently and inventing techniques nobody had taught them. The other eighteen had quietly drifted back to manual work. The training had not failed. What had failed was everything that was supposed to happen after it. Nobody owned the slow, unglamorous work of turning eighteen isolated learners into a network that taught itself. That work is what a community of practice does, and building one is the difference between a training event and a durable capability.

Why Training Decays Without a Network

Formal training delivers a fixed body of knowledge at a single point in time. The AI tools your recruiters use, and the tactics that work with them, change faster than any training cycle can keep up. The recruiter who discovers that adding a candidate's specific project to an outreach prompt lifts reply rates has learned something valuable, but if that insight stays in her head, the organization learns nothing. Multiply that across 22 recruiters and you have dozens of private discoveries that never compound. A community of practice exists to make that compounding happen: to turn individual learning into shared, retrievable institutional knowledge.

The framing is not new. Etienne Wenger, who coined the term, defined a community of practice through three elements: a domain, the shared area of interest that gives the group its identity; a community, the relationships and interactions through which members learn from one another; and a practice, the shared repertoire of tools, methods, and stories the group develops over time. Dana's situation maps cleanly onto this. The domain is using AI well in recruiting. The community is her 22 recruiters and a few hiring partners. The practice is the prompts, playbooks, and hard-won lessons they have not yet pooled. Her job is not to invent the domain, which already exists, but to build the community and capture the practice.

Start by Writing Down What the Community Is For

Purpose shapes everything downstream, and a community that has not stated one drifts toward whatever its loudest members find interesting. Ask the question plainly. Is this group here to share best practices with the screening tool? To troubleshoot problems as they arise? To surface and discuss fairness concerns? To celebrate wins and make good work visible? Each answer produces a different meeting. A best-practices community leans on demonstration and curation; a problem-solving community leans on live cases and working groups. A group trying to do all four without saying so ends up doing none of them well.

So be explicit and write it down, in one sentence a member could repeat accurately. Dana's read: this community shares knowledge, solves problems collaboratively, and celebrates wins with our AI tools. That sentence is the top line of the community charter, which is simply a one-page document recording the decisions this lesson works through: purpose, structure, membership, facilitation, and how knowledge gets captured. The charter is not bureaucracy. It is what lets a new joiner in month seven understand what they are walking into, and what lets the group notice when it has quietly become something else.

The Cadence That Sustains It

A community without a regular heartbeat dissolves. The single most reliable predictor of whether one survives is a predictable, low-friction cadence that people can plan around. Dana set a biweekly 45-minute session, every other Thursday at 11 a.m., booked six months out so it became a fixed point on calendars rather than a recurring negotiation. Forty-five minutes is deliberate: long enough to demonstrate something real, short enough that a busy recruiter does not resent the hold.

The format stayed consistent so nobody had to wonder what they were walking into. Ten minutes of "what worked since last time," in which two or three recruiters share a specific win. Twenty minutes on a single focused topic, often a live walkthrough. Ten minutes of open problem-solving, where anyone can bring a prompt that is misbehaving. Five minutes to capture decisions and assign whoever will document the session. That last five minutes is the part most groups skip and the part that determines whether anything survives the meeting, which is why it sits inside the agenda rather than being left to good intentions afterward.

Synchronous, Asynchronous, or Both

Structure is a separate decision from cadence, and there are only three real options. A synchronous community meets: engagement is high because people are present with each other, and the cost is that everyone must be free at the same time. An asynchronous community lives in a channel or forum: participation is broader because time zones and calendars stop mattering, and the cost is that discussion is less structured and harder to capture. A hybrid runs both, which suits most recruiting organizations because it lets people engage in the mode that fits their week.

If you choose hybrid, set the expectation in a single line so nobody has to guess which surface carries what: the session happens on its fixed cadence and attendance is optional, the channel is open anytime, and anything worth keeping ends up in the repository either way. Ambiguity here is what produces the community where half the members think the meeting is the real forum and the other half think the channel is, and neither group hears what the other decided.

Membership: Invite, Do Not Require

The membership question is who belongs: everyone in recruiting, only the power users, or whoever volunteers. Dana invited all 22 recruiters and required none of them, and the reasoning matters more than the choice. Voluntary participation is self-selecting, so the people in the room are the people who want to be there, and what they contribute is genuine rather than performed. Her first session drew nine. By the second quarter, regular attendance settled around fourteen, with others dropping in when a topic was relevant to them.

The alternative fails in a specific way. Mandating attendance produces a room containing people with conflicting commitments and people who do not yet see the value, and both contribute minimally. Discussion becomes strained, the few experienced members dominate because nobody else has anything they want to say, and the meeting reads as an obligation rather than a resource. Twelve engaged volunteers generate more real learning than 22 conscripts counting the minutes. Say the invitation in those terms: membership is open to every recruiter, and we hope you will join. Then make the sessions valuable enough that people choose to come back, which is a harder standard than attendance and the only one that means anything.

The Shared Repository as the Community's Memory

Meetings generate energy, but energy evaporates. The community's lasting value lives in what gets written down, and the centerpiece of Dana's effort was a shared repository. She started a wiki space, using Confluence, though Notion or any similar tool serves the same purpose; the tool matters far less than the discipline of maintaining it. At its core sat a vetted prompt library. The rule for the library was strict: a prompt entered it only after at least three recruiters had used it successfully on real requisitions. Everything else lived in a separate experimental area, clearly labeled so nobody mistook a draft for a standard.

The library opened with eight prompts that Dana and her two strongest recruiters had already proven. Over the next two quarters, fed by the biweekly sessions, it grew to 40 vetted prompts organized by recruiting task: screening, outreach, interview support, and research. Each entry carried the same lightweight documentation: a name, the use case in one sentence, the template with bracketed placeholders, an example of good output, and version notes recording what changed and why. The pairing is what made it work. The biweekly session surfaced new techniques; the repository preserved the ones that proved out. Neither alone would have sustained the other. A repository with no community goes stale because nobody feeds it; a community with no repository forgets everything it learns the moment the meeting ends.

What Gets Shared, and How It Gets Kept

The material a community trades in is more varied than a prompt library suggests. Tips and techniques for using the tool. Interesting cases that came up on a live requisition. Lessons learned from something that went wrong. Practices that have settled into a standard. What all of it has in common is that it is peer-generated, which is exactly why it carries weight: a technique demonstrated by someone doing the same job under the same constraints is more credible than the same technique in a training deck, and it is more practical because it has already survived contact with a real req.

Getting people to contribute is mostly a matter of asking in a way that does not demand expertise. Two questions do most of the work: what worked for you, and how did you solve that. Both invite a specific story rather than a claim of authority, which is why quieter members answer them. Then curate, because raw contribution is not knowledge yet. Periodically synthesize the scattered threads into a single durable entry, publish something like a top-ten list of techniques the community has proven, and share it beyond the community so the value is visible to people who have not joined.

Design the capture system explicitly, because the default is that nobody owns it. Decide the format, whether a wiki, a document set, or a periodic written summary. Decide who maintains it, whether the facilitator or a rotating responsibility. Decide how it is organized, by recruiting task, by tool, or by problem type, and pick the one your members would search by. Decide the update rhythm, whether after every session or as a quarterly synthesis. And decide the audience, whether the material stays inside the community, reaches all recruiters, or goes wider still. Five decisions, written once, and the difference between a knowledge base and a folder of meeting notes.

Show-and-Tell as the Engine

The most engaging part of Dana's sessions was the internal show-and-tell, the standing "what worked" segment. A recruiter would screen-share an actual prompt, the output it produced, and the result it drove on a live requisition. This did several things at once. It surfaced techniques no central trainer would have thought to teach. It gave credit publicly, which is its own fuel for participation. And it kept the community anchored to real work rather than abstract discussion. When a recruiter demonstrated an interview-summary prompt that cut her post-debrief writeup from roughly 25 minutes to under 10, that was not a claim from a slide; it was a peer the others trusted, showing the thing working.

Impact estimates followed from those demonstrations rather than preceding them. Illustratively, the team came to estimate that the shared library was saving an active recruiter on the order of two to three hours a week once they adopted the vetted prompts, with the biggest gains in screening and feedback summarization. Treat a figure like that as the community's own working estimate, useful for showing members that participation changes results, and label it as an estimate when you repeat it upward. A community that oversells its impact to leadership sets a bar it will be measured against in a quarter when nobody has time to meet.

Facilitation, and Why It Rotates

Communities need light structure or they fade, and the lightest sufficient structure is a named facilitator. The role is smaller than it sounds: confirm the topic, schedule and remind, prompt the "what worked" sharers in advance, keep the discussion roughly on track, and make sure someone captures the session notes. A part-time commitment in the range of five to ten hours a month is enough to sustain a community of this size. What the facilitator does not do is control the agenda. Members propose topics and shape direction; the facilitator enables rather than directs, and a group whose topics are all chosen centrally has become a training series with a different name.

Two facilitation skills are worth naming because they do not come automatically. The first is drawing in quieter members, which is done by soliciting a specific contribution in advance rather than opening the floor, by asking directly in a one-to-one, and by keeping the asynchronous channel open for people who will write but not speak. The second is handling drift, whether an off-topic thread or a disagreement that has stopped being productive, which is usually a matter of naming it and parking it rather than suppressing it.

Then rotate the role. After the first two months, Dana published a schedule that assigned each session to a different recruiter, with herself as a backstop. Rotation distributed the load, developed informal leaders, and gave more people a stake in the community's success. It also protected against the most common failure mode, in which a single overextended owner burns out and the whole thing quietly dies with them. Rotation is the structural answer to the question of what happens when the founder changes roles, which is a question every community eventually faces.

Connecting Beyond the Team

A recruiting community does not have to be an island. Dana wired her group into two larger networks. Internally, she opened a Slack channel for the rare cross-functional moments when recruiting's AI use touched HR operations, IT, or legal, so a fairness question or a data-handling concern could reach the right people without a string of separate emails. The channel stayed scoped to that purpose rather than becoming a second, unstructured forum that competed with the repository.

Externally, she encouraged two recruiters to join an industry community of talent professionals working on the same problems, and to bring back one idea per quarter to the biweekly session. External communities counter the slow inward drift that small internal groups suffer, where the same dozen people keep reinforcing the same assumptions. A single outside technique, vetted against the team's own work and added to the library, was often worth more than a month of internal discussion. The discipline was the same as for everything else: import freely, but let nothing into the vetted library until it had proven out on real requisitions.

Sustaining Momentum and Designing Against Decay

Communities decay in predictable ways, and each has a structural fix. The first failure is drift into social chat: a channel fills with scattered questions and scattered answers, and a month later someone asks the same question because the answer was never captured. The fix is light curation, with the facilitator synthesizing recurring threads into a single repository entry so a question gets answered once and stays answered. The second is the absent owner, fixed by the rotating schedule that makes scheduling, reminders, and note capture somebody's named job rather than someone's spare time. The third is staleness, the sense that every meeting covers the same ground.

Staleness has more countermeasures than the other two because it arrives later and more slowly. Keep the cadence predictable and scheduled well in advance, so the meeting never has to be re-sold. Let topics evolve as the group matures, since the questions a community asks in month one are not the ones worth an hour in month nine. Rotate who leads. Recognize contributions publicly, naming the people whose techniques entered the library. And connect discussion to outcomes explicitly, so members can see that what happens in the session changes how the work goes, rather than filling a calendar slot. A community that cannot show its members that it changes anything will lose them to the work that visibly does.

Knowing Whether It Is Working

Attendance is the metric everyone reaches for and the weakest one available. Five readings together give a truer picture. Participation asks whether people attend and whether the same people attend consistently. Engagement asks whether they contribute and what the quality of those contributions is. Knowledge asks what is actually being learned and how much of it has been documented. Impact asks whether the community is solving problems, improving adoption, and reducing the support burden that used to land centrally. Sustainability asks whether it runs itself or is quietly burning out whoever facilitates.

By the end of two quarters, Dana could read all five. The library had grown from eight prompts to 40. Adoption had spread from four fluent recruiters to fourteen regulars. And the support pattern had shifted, with recruiters increasingly answering one another's questions instead of routing everything to her. That last shift, peers teaching peers, is the point, and it shows up in the other organizational effects a healthy community produces: less support load on whoever owns enablement, more improvement proposed and tested by the people doing the work, a shared identity around using these tools responsibly, and stronger attachment among the members who are most engaged. The honest signal is not attendance. It is whether knowledge is compounding, which means the capability now outlives any single training event and any single owner, including the person who built it.

Anti-Patterns

A community with no named facilitator. The group is announced, enthusiasm is real, and the first two discussions happen organically. Then nobody tracks the action items, nobody schedules the next session, and within two months the community is dormant. It happens because the early energy makes structure feel unnecessary, and because naming a facilitator means asking someone for hours they do not obviously have. What goes wrong is that every recurring task, scheduling, reminders, agenda, and capture, is exactly the kind of work that is nobody's job by default. The counter is a named facilitator with a part-time commitment in the range of five to ten hours a month, and a written list of what the role owns.

Mandatory attendance. Every recruiter is required at the monthly session, so the room is full and the discussion is dead. It happens because attendance is the easiest thing to measure and the easiest thing to instruct, and because a voluntary group that draws nine out of 22 looks like a failure on a slide. What goes wrong is that people with genuine conflicts resent the hold, people who have not seen the value contribute nothing, and the few confident members dominate by default, which teaches everyone else that the session is not for them. The counter is to invite rather than require, and to accept a smaller engaged core as the better outcome.

The community that never becomes productive. The channel fills with "has anyone used this feature," a few scattered answers arrive, nothing is synthesized, and a month later the same question is asked because the answer is lost in the scroll. It happens because unstructured conversation is genuinely pleasant and curation is genuinely work. What goes wrong is that valuable knowledge is shared and immediately discarded, and problems are discussed without anyone owning a fix. The counter is light curation as a standing duty: synthesize recurring threads into a durable entry, publish periodic roundups, and when a real problem surfaces, form a small working group to close it rather than letting it dissolve back into chat.

A library with no vetting rule, or vetting so heavy nothing enters. The two failure modes are opposites with the same result. In the first, anything anyone posts becomes a library entry, so members stop trusting it and go back to asking colleagues directly. In the second, every submission needs sign-off from the enablement lead, so the queue grows, contributors lose interest, and the library ossifies around whatever was added at launch. The counter is a peer-scale rule stated in advance, such as Dana's requirement that three recruiters use a prompt successfully on real requisitions before it is vetted, plus a clearly labeled experimental area so drafts have somewhere to live that is not the standard.

Letting the founder remain the community. One person schedules, facilitates, curates, and answers, and the group is healthy right up until that person changes roles. It happens because the founder is usually the most capable facilitator and rotation feels like a downgrade in quality for the first few sessions. What goes wrong is that no one else develops the skill, members address every question to the founder rather than to each other, and the community's memory turns out to have been one person's memory. The counter is a published rotation schedule with the founder as backstop, started early enough that the first handover happens while the founder is still there to make it work.

Practice

• Write the one-page charter. State the purpose in a sentence a member could repeat accurately, then record the structure, the membership rule, who facilitates and what that role owns, and how learning gets documented. Keep it to one page, and date it so you can tell later when the group stopped matching its own description.
• Choose your structure deliberately. Decide between synchronous, asynchronous, and hybrid on the basis of how your team's weeks actually run, then write the expectation in a single line covering which surface carries what, and where anything worth keeping ends up regardless of where it was first said.
• Design the first three sessions and the launch. Decide how you will recruit initial members and how the invitation is worded so it invites rather than mandates. Choose a first topic likely to draw people who are curious rather than obligated. Name the quick win you expect in the first two months, and how you will make it visible.
• Write the facilitation guide. Specify the role and its time commitment, the cadence and agenda structure including a fixed capture slot, how quieter members are drawn in, how off-topic threads and disagreements are handled, and how the notes reach the repository. Then publish the rotation schedule with yourself as backstop.
• Build the capture system with five decisions. Format, maintainer, organizing scheme, update rhythm, and audience. Then write the vetting rule that separates a proven entry from an experiment, and state it in peer terms rather than as an approval from you.
• Define what health looks like before you launch. Write down what you will read for participation, engagement, knowledge, impact, and sustainability, including at least one signal that is not attendance. Add the specific observation that would tell you the community is becoming self-sustaining, such as members answering one another rather than routing questions to you.

Reflection

• If the person who founded your community changed roles next month, what would still be running eight weeks later?
• When one of your recruiters discovers a technique that works, where does it go, and could a colleague in another region find it?
• Is your community's purpose written down anywhere, and would two members describe it the same way?
• What was the last question your team asked twice because the first answer was never captured?

Glossary

• Community of practice. A peer network built around a shared area of work, existing to turn individual learning into shared institutional knowledge. Distinct from a training program in that it continues indefinitely and its content comes from members rather than a curriculum.
• Domain, community, practice. Wenger's three elements. The domain is the shared area of interest that gives the group identity; the community is the relationships through which members learn from one another; the practice is the repertoire of tools, methods, and stories the group develops over time.
• Charter. The one-page document recording purpose, structure, membership, facilitation, and knowledge capture. It lets a new member understand what they are joining and lets the group notice when it has drifted into being something else.
• Cadence. The predictable heartbeat of the community: a fixed slot booked well in advance, short enough that a busy recruiter does not resent it, consistent enough that nobody has to wonder what the session will contain.
• Synchronous, asynchronous, hybrid. The three structural options. Synchronous buys engagement at the cost of simultaneity; asynchronous buys breadth at the cost of structure; hybrid runs both and requires an explicit statement of which surface carries what.
• Voluntary membership. An invitation rather than a requirement, on the reasoning that self-selection produces genuine contribution. A smaller engaged core is the better outcome, not a shortfall against full attendance.
• Vetted prompt library. The curated core of the repository, holding only entries proven on real work, with a labeled experimental area alongside it so drafts are never mistaken for standards.
• Vetting rule. The stated, peer-scale condition an entry must meet before it is treated as a standard, such as successful use by three recruiters on real requisitions. Stated in advance so the library is neither an unfiltered dump nor a queue awaiting one person's approval.
• Version notes. The record on each library entry of what changed and why, which lets a technique that starts producing worse output be traced to a specific edit rather than argued about.
• Show-and-tell. The standing segment in which a member demonstrates a real artifact and the result it produced on live work. It surfaces techniques no central trainer would invent and gives public credit, which is its own fuel for participation.
• Light curation. The facilitator's periodic synthesis of scattered threads into durable entries, and the countermeasure to a community that becomes pleasant conversation with no retained knowledge.
• Rotating facilitation. A published schedule assigning each session to a different member with the founder as backstop. Distributes load, develops leaders, and removes the single point of failure that kills most communities.
• Community health readings. The five signals that beat attendance: participation, engagement, documented knowledge, impact on problems and support load, and sustainability of the facilitation itself.

Related Lessons

• Hands-On Project: Design a Team Capability Building Program places the community inside the wider program of assessment, curriculum, coaching, and measurement that it sustains.
• Coaching and Support: Helping Individuals Build Confidence develops the one-to-one work that runs alongside the community, including how to reach the member who will not speak up in a group.
• Training and Capability Building: Ensuring Staff Can Use AI Responsibly covers the formal instruction whose decay this lesson is designed to prevent.
• Adult Learning Principles: How Recruiters Learn and Adopt New Tools explains why peer demonstration on real work outperforms curriculum coverage.
• Hands-On Practice: Build Your Prompt Library is where the repository at the center of this lesson gets built as a deliverable.
• Measuring Adoption: Tracking Usage, Proficiency, and Impact develops the health readings and the distinction between activity data and evidence of capability.

Closing

Dana's community was not an innovation. It was a fixed 45 minutes, a consistent agenda, an invitation rather than a summons, a library with a rule about what could enter it, and a rotation schedule that meant the group never depended on her being available. Everything durable about it came from decisions made once and written down, which is why it survived the quarter when she was pulled onto a different project.

The parts that will be tempting to skip are the ones doing the work. The five minutes of capture at the end of every session, because the conversation is more interesting than the note. The vetting rule, because it means telling a keen contributor that their prompt is not in the library yet. The rotation, because the founder is genuinely the best facilitator for the first few months. And the written purpose, because everyone thinks they already agree on it. Skip those four and you will have a recurring meeting, which is what most organizations mean when they say they have a community of practice.

Key Takeaways

• Training is an event; a community of practice is the capability. A workshop transfers fixed knowledge once. AI tools and tactics change faster than training cycles, so without a network that keeps learning, adoption decays back toward the few self-starters while everyone else drifts to manual work.
• Map the work onto domain, community, and practice. Wenger's three elements clarify the job: the domain already exists, so a leader's real work is building the community and capturing the practice.
• Write the purpose down, then the charter. A group that has not stated whether it exists to share practices, solve problems, surface fairness concerns, or celebrate wins will drift toward whatever its loudest members enjoy. One page covering purpose, structure, membership, facilitation, and capture is enough.
• A predictable cadence is the heartbeat, and capture belongs inside the agenda. A short, fixed, voluntary session with a consistent format, booked months ahead, gives people something to plan around. The five minutes that assign documentation are what survive the meeting.
• Choose structure deliberately and say which surface carries what. Synchronous buys engagement, asynchronous buys breadth, hybrid suits most teams. Ambiguity produces two half-communities that never hear each other's decisions.
• Invite, do not require. Voluntary membership self-selects for people who contribute genuinely. Twelve engaged volunteers learn more than 22 conscripts, and mandated attendance produces strained discussion dominated by the few.
• The repository is the community's memory, and it needs a vetting rule. Meetings generate energy that evaporates; a shared library preserves it. State a peer-scale condition for entry, keep experiments clearly separate from standards, and record version notes on every change.
• Show-and-tell drives engagement; label impact estimates honestly. Peers demonstrating real artifacts on live work surface techniques no trainer would invent and give public credit. Report the community's own savings estimates as estimates when you take them upward.
• Facilitate lightly, and rotate the role. A named facilitator at a few hours a month owns scheduling, reminders, agenda, and capture, while members propose topics. Rotation spreads the load, develops leaders, and removes the single point of failure that quietly kills communities.
• Measure health, not attendance. Read participation, engagement, documented knowledge, impact on problems and support load, and sustainability of the facilitation. The signal that matters is knowledge compounding and peers answering peers.

Frequently Asked Questions

Our recruiting team is six people. Is a community of practice overkill? The instrument scales down further than most people expect, but two things change. Cadence can stretch, since six people generate less new material than 22, so monthly may fit better than biweekly. And the repository matters more, not less, because a six-person team has fewer chances for knowledge to survive by accident in someone else's memory. What you should not drop is the written purpose or the capture habit. A six-person group that meets without capturing anything is a conversation, which is fine, but it will not survive one person leaving, and on a team of six that is a much larger fraction of the institutional knowledge.

Attendance has been sliding for two months. Do I make it mandatory? Mandating attendance converts a symptom into a different symptom. Diagnose first, because the three decay modes have different fixes. If discussion has drifted into pleasant chat with nothing retained, the problem is curation and the fix is synthesizing threads into durable entries. If sessions are being cancelled or organized late, the problem is the absent owner and the fix is a named, rotating facilitator. If people come once and do not return, the problem is staleness, and the fixes are evolving topics, public recognition, and making the link between the session and measured results visible. Requiring attendance addresses none of the three and costs you the goodwill of the members who were still choosing to come.

How do I stop the same three people from doing all the talking? Structure the invitation rather than the room. Ask specific people for a specific contribution in advance, so they arrive with something prepared rather than having to compete for airtime. Keep the asynchronous channel genuinely open, because a proportion of your quieter members will write what they will not say. Use one-to-one conversations to find out what someone has been doing that the group has not heard, then invite them to show it. And rotate facilitation, since facilitating is a reliable way for a quieter member to acquire standing they would not claim by volunteering.

Leadership wants a number that justifies the time. What do I give them? Give them the health readings rather than a single headline, and be careful about which numbers you promise. Documented knowledge is the most defensible measure, because a library that grew from eight entries to 40 is a fact rather than an inference. Adoption spread is next, since the count of recruiters using the vetted material is countable. Support load is a strong third, because a shift toward peers answering peers is visible in whoever used to field the questions. Time-savings estimates that come out of member demonstrations are worth reporting, and worth labeling as the community's own estimate rather than a measured result, because the quarter you are asked to prove them is the quarter nobody has time to.