When you’re rolling out new technologies or launching new projects, a knowledge-based community of practice (CoP) is crucial for success. This is particularly true when it comes to a technology as complex and potentially transformative as AI. An interdisciplinary, cross-functional CoP helps break down knowledge silos and bring separate teams together around a single initiative, project, or issue. Stack Overflow’s view is that fostering a knowledgeable CoP around AI is essential to getting AI projects from initial enthusiasm to successful execution.

In this article, we’ll explain what a CoP is, unpack why they’re so valuable for AI projects, and offer some practical tips for building one.

What is a community of practice?

Communities of practice are self-directed groups organized around a common interest, whether that’s AI, civil engineering, or gardening. These communities are “groups of people who share a concern or a passion for something they do and learn how to do it better as they interact regularly,” as one formal definition goes.

CoPs allow colleagues from teams who might not interact on a regular basis to connect around a specific topic, share experiences and resources, and leverage their collective expertise to solve problems and refine best practices. With the community’s help, users can uncover important context and draw connections they might not have made on their own. By building on those connections, the community can generate new solutions and preserve reusable knowledge.

From a business perspective, CoPs serve several functions: breaking down silos and encouraging cross-functional collaboration, enabling more autonomy, building trust and confidence among employees, and accelerating innovation. Those benefits are all highly relevant for your AI journey.

Async communication tools aren’t a substitute for community

Platforms like Slack that keep colleagues connected asynchronously are central to how most of us work on a day-to-day basis, but they’re no substitute for a platform built to foster community knowledge. They’re also a chief source of distraction and productivity loss. Constant messages and notifications pull us out of our flow state, making complex and creative work more difficult. Apps like Slack and Microsoft Teams also do nothing to capture and preserve knowledge for the benefit of the whole organization, something that’s crucial to building a community of practice.

How do communities of practice help AI implementations succeed?

When it comes to a new and largely untested technology like AI, CoPs are especially critical. That’s largely because organizations are still learning how and where to leverage AI in their business practices, exploring the limits and possibilities of AI for customer-facing and internal use cases. In their 2023 report Become an AI-First Organization: 5 Critical AI Adoption Phases, Gartner stresses the importance of forming a CoP to enable knowledge-sharing and innovation around AI.

They enable knowledge-sharing and collaboration

Building an AI-centered community of practice at your organization brings people and resources together. People have one place to go to learn from one another, share experiences and suggestions, ask for help, and collaborate on new and better solutions.

CoPs also help organizations manage the complexity of AI implementations, which can be challenging and unpredictable. Far from following a set of neatly defined adoption phases familiar to any organization rolling out a new tool, AI adoptions aren’t orderly or unidirectional. Organizations frequently go in with sometimes wildly unrealistic expectations, from what the technology can help them do to how long it will take to deploy to how much it will cost. A dedicated community of practice helps level-set expectations from the start and prepare you for the challenges and complexities you’re likely to encounter.

They collect high-quality information to improve AI output

At Stack Overflow, we know that high-quality data is the bedrock of any successful AI project, whether you’re building a customer-facing chatbot or an internal tool to help your teams automate toil and solve problems faster.

As we’ve written, research from MIT and elsewhere shows that integrating a knowledge base into a large language model (LLM)’s performance is likely to improve the model’s output and reduce hallucinations. The maxim “garbage in, garbage out” certainly applies to any AI project you might be considering.

LLMs are entirely dependent on the training data you provide. When they’re trained on faulty, stale, or incomplete data, they’re vulnerable to hallucinations. From not-quite-right answers to totally off-base information about people or events, hallucinations are incorrect results. They can mislead users, contribute to a decline in code quality and maintainability, create security vulnerabilities, erode user trust in your platform, and a whole lot else that you want to avoid.

Creating knowledge that the AI can learn from and leverage to improve itself is another way a knowledge-focused CoP ultimately enables better output from both AI and human learners. CoPs enable cross-functional discussion and collaboration, allowing colleagues to work together to vet AI responses and refine prompt structure and other factors to improve output quality. This is a form of reinforcement learning: Humans apply their judgment to the AI’s output, thereby helping the model improve.

How to build your community of practice

Before you can reap the benefits of a community of practice, you need to build one. Here are some practical steps you can take to build a CoP around your AI project:

• Choose a community platform built to surface, capture, and preserve knowledge. Users need a place where they can learn from one another, compare notes, and work together on fresh solutions. Stack Overflow for Teams is the platform of choice for teams at Expensify, Instacart, and Dropbox.

• Bring everyone to the table. An AI-focused CoP needs to include a multitude of voices and perspectives: people across multiple teams, from different backgrounds and with divergent job titles. The people at your organization with the most expertise in or insight surrounding AI may not be the people you’d expect. Include folks from across the organization and encourage everyone to contribute to the CoP.
• Narrow your focus. AI is such a huge and evolving field that an “AI working group” could be an enormous and unwieldy area of study and input—whereas a group more narrowly focused on, for example, how to improve the search experience using AI will probably be more successful in understanding specific business challenges and opportunities.
• Set specific goals. Along the same lines, set specific goals for what you want your AI-focused community of practice to accomplish. “Eliminate some toil for developers by automating boring, repetitive tasks with AI” is a better goal to task your CoP with than “deliver more value to customers with AI.” To give you a place to start, we outlined several AI performance and business outcome metrics you could consider in your AI adoption journey.
• Provide plenty of support: New members and beginners should have easy access to resources via a searchable knowledge platform that helps them self-serve answers to their questions. CoPs are meant to help people learn and expand their skill sets, so making knowledge readily accessible to everyone in the community is one of their essential functions.