What is a community of practice?

A community of practice is a group of people who do similar work, often in different teams or sites, and who meet regularly to share what they are learning. It is voluntary and informal, and sets no rules. For teams working with AI agents, it is where reviewers and managers compare hard cases, agent quirks and fixes.

Why does it matter for agent oversight?

Because the same agent behaves the same way everywhere, but each team discovers its habits separately. A clinic manager in one town notices the agent holds every appointment request that mentions a child; a manager in another town noticed the same thing a month earlier and found out why. Without a way to share, each learns alone and slowly. A community of practice turns isolated discoveries into shared knowledge.

It also keeps judgement aligned across sites. Formal calibration happens within a team. A community of practice extends a looser version of it across teams, so a held action in one site is more likely to be decided the same way as in another.

How does one work?

Typically:

• A regular meeting, monthly or fortnightly, short and practical.
• Case sharing. Each member brings one interesting held action, refusal or correction from the record.
• A shared space for notes, questions and answers between meetings.
• A light coordinator who keeps it going, often a champion or senior reviewer.
• A link to the agent owner, so patterns that need a change to the agent reach someone who can make it.

It works best when members bring real cases. Discussions of principle tend to fade; a specific held item with the facts on screen keeps people engaged.

An example

Take a hypothetical group of twelve GP practices, each with a practice manager who reviews the appointments and patient-messaging agent's held items. The managers form a community of practice that meets online for forty minutes each month. In one session, a manager shares that the agent refused to rebook a patient whose phone number had changed, because the new number was declared, not verified. Two others had approved similar cases by calling the patient back; one had not known that was allowed. The group agrees a common approach, and the agent owner adds a clearer reason to the refusal. The group's work feeds into each practice's training; how to train the practice team to review AI agent decisions and the reviewer onboarding checklist for the practice team are where it lands.

How is it different from a centre of excellence?

A centre of excellence is a formal team with a mandate to set standards and build training. A community of practice has no mandate; it shares and learns. The two complement each other. The centre writes the standard; the community shows how it holds up in daily work and feeds back what needs changing. A community without a centre can drift into many local versions of good practice. A centre without a community can write standards that nobody follows.

How does it support training?

Communities are where people keep learning after formal training ends. They also prepare for rare events: members can run shared exercises, such as an AI agent incident drill for the practice team, and compare how each site responded. Over time, the community's cases become material for new starters. The guide to training teams for AI agents covers how formal and informal learning fit together.

Where the Academy fits

Pryme Intelligence Academy provides the shared baseline that makes community discussion productive: when every practice manager has completed the same Operator modules, they share a vocabulary for held, refused and withheld, depth and provenance. The community then adds what a course cannot, which is each site's real cases.