How to Scale with AI Agents While Keeping the Community's Human Touch

As the community grew in size, repetitive operational tasks such as question routing, duplicate checking, and follow-ups surged. Vercel built two AI agents to reduce this operational burden and let team members focus on complex debugging and relationship building.

Community Guardian handles operations and logistics. When a new post is created, it analyzes the post to check for duplicates and assigns it to the appropriate owner. It manages the operational process by reassigning questions that have been left unattended for 48 hours or sending reminders when information is lacking. This agent is built on Claude and operates efficiently in terms of resource usage through Vercel Workflows.

c0 is a research agent that assists team members with their work. Operating within Slack, it searches the knowledge base, documentation, GitHub issues, and past discussions needed for team members to answer questions, and provides a Context Package. It also tracks community sentiment and recurring technical issues to deliver data-driven feedback to the product team.

As a result of the rollout, the following outcomes were achieved in the first 23 days:

• 4,716 instances of initial context gathering and issue classification
• Recovery of 1 in 8 abandoned threads, leading to 23 solutions
• Over 1,400 agent runs over two weeks
• Detection and automatic closure of duplicate threads via vector similarity

This system goes beyond simple 'If-Then' rule-based automation, featuring a Thinking Layer that understands and judges user intent. This demonstrates that even non-developers can build complex workflows using the AI Gateway and AI SDK.