Keeping community human while scaling with agents
Source: Vercel Blog
The Community Guardian – How Vercel Scales Its Developer Community
At Vercel, our developer community is at the heart of everything we do. It’s how we stay closest to the people using what we build.
As the community grew, automation helped us scale, but we still faced:
- Lost questions
- Slow routing
- Context‑switching that pulled us away from work that truly required our expertise
Automation alone can’t replace the moments where you really connect with someone and help them. You can’t use AI to replicate the feeling of talking to a person who truly cares.
Introducing the Community Guardian
We built agents to take over the routing, triage, and follow‑ups that don’t need a human. The Guardian handles the logistics so our team can focus on the human side of support.
What the Guardian does
- Analyzes new posts – checks for duplicates.
- Assigns to the right team member – based on specialty and current bandwidth.
- Balances workload – each person handles up to 10 questions before new ones are routed elsewhere, keeping the load even across time zones.
- Ensures nothing is missed – if a question is unanswered for 48 hours, it is reassigned.
- Sends reminders – when we’re waiting for more information.
- Detects resolution – marks conversations as solved when appropriate.
Under the hood
- Model: Claude (via the AI Gateway).
- Execution: Runs on Vercel Workflows, which lets it check in every 10 minutes and sleep between cycles without consuming resources.
The Intelligence Layer – c0, the Research Assistant
While the Guardian manages logistics, c0 dives deep on research. It lives in Slack, where our team already works.
- Context packages – when a team member needs background on a thread, c0 searches our knowledge base, documentation, GitHub issues, and past discussions, then bundles the relevant information.
- Product‑team insights – c0 tracks community sentiment and recurring technical hurdles. Instead of manually auditing a week’s worth of posts, we can ask c0 for the “top product feedback” and bring real data to product conversations.
Impact in the first 23 days
- Helped 281 unique users.
- Every substantial answer still comes from a human; agents handle everything else (triage, tracking, reminders).
- Freed the team to focus on complex pair‑debugging, relationship building, content creation, and simply having fun with developers.
Building It Without Being a Developer
You don’t have to write production code every day to create something like this. All you need is an idea and a willingness to iterate.
My journey
- Initial demo – I showed how we were automating community workflows with traditional scripts and IF‑THEN‑ELSE rules. It worked but was brittle; every edge case required a new rule.
- Adding a “thinking layer” – I used a coding agent to move from static rules to a model that understands each post:
- “Read this post, understand what the person actually needs, then decide.”
- The layer can read between the lines, connect a user’s frustration to a GitHub issue from months ago, and decide whether to escalate or gather more context.
- Choosing the right tools – Instead of building everything from scratch, I described my needs to the coding agent and landed on:
- AI Gateway – for model access.
- AI SDK – a unified TypeScript SDK for building agents across Slack, Teams, Discord, etc.
- Vercel Workflows – for durable, suspend‑and‑resume execution.
Prompt evolution (simplified)
| Prompt | Result |
|---|---|
| “Build me an agent that helps with community operations like assigning posts and formatting. I don’t know which model will work best yet, but make it easy to switch without new API keys. Use the AI SDK.” | Scaffolded a basic agent with configurable model selection. |
| “Trigger every 10 minutes to check for the latest threads.” | Switched from cron jobs to Vercel Workflows for durable execution. |
| “Rotate assignments every 4 hours.” | Added rotation logic and state persistence. |
The key is describing the problem in plain English, getting working code back, testing against real threads, and iterating.
How You Can Build Your Own
- Pick a unified SDK – We used the Chat SDK (a TypeScript SDK) to build agents that work across multiple chat platforms.
- Use a durable execution platform – Vercel Workflows let your agent pause between checks and resume exactly where it left off.
- Select a model via an AI Gateway – Swap models without changing API keys.
Tip: You don’t need to know the right terminology or be a seasoned coder. Understand the problem, describe it clearly, and let the coding agent generate and refine the implementation.
Quick reference links
- AI Gateway – [link to AI Gateway docs]
- AI SDK / Chat SDK – [link to SDK docs]
- Vercel Workflows – [link to Workflows docs]
TL;DR
- Community Guardian handles routing, triage, reminders, and workload balancing.
- c0 provides deep research, context packages, and product‑team insights.
- The system runs on Claude via the AI Gateway, orchestrated by Vercel Workflows, and is built with the unified Chat SDK.
- You can replicate this without being a developer: start with an idea, describe it to a coding agent, and iterate.
Want to try it yourself?
Share what you build with us – we’re always happy to talk through what we’ve learned.
Read more about:
- The Community Guardian operations layer
- The intelligence layer: c0, the research assistant
- Reclaiming human focus
- Build your own
Built with heart
Metrics
| Metric | Outcome |
|---|---|
| Initial context gathering | 4,716 first responses triaging issues and gathering logs before a team member arrives |
| Thread revival | 1 in 8 “ghosted” threads brought back to life, resulting in 23 confirmed solutions |
| Operational scale | Over 1,400 agent runs in a recent two‑week period, from stale‑checks to auto‑solving |
| Duplicate detection | 4 duplicate threads detected via vector similarity, with 3 auto‑closed at 95 %+ confidence |
The prompts that built it
(Insert prompt list here)