Automating 90% of support tickets: How we built Vercel's support agent
summary
Claims 91% of tickets resolved autonomously, 130,000 cases handled and 5,000 engineer hours saved monthly after a 3‑week build by two engineers. Describes the stack: AI SDK for the agent loop and UI, AI Gateway for multi‑provider model routing, Vercel MCP + Next.js to expose the exact internal tools the agent calls, and turnkey search via the Vercel marketplace. Shows concrete flows, tool call examples (Stripe invoice retrieval, approval UI), model swap that raised resolution by 5%.
AI-written summary · about
chapters (17)
- 0:00Support agent project intro
- 1:53Why build in-house agent
- 3:19Challenges with off-the-shelf
- 4:23Ideal support experience vision
- 5:25How agent processes tickets
- 6:50Agent loop with AI SDK
- 8:15Real-time chat interface
- 8:58Fast responses with AI Gateway
- 10:22Connecting to internal systems
- 11:46Simplifying MCP setup
- 12:49Keeping knowledge base current
- 13:51Search with Upstash Search
- 14:57Components on Vercel
- 15:38Agent demo
- 17:03Next steps for your agent
- 18:27Results and takeaways
- 19:08Call to action
related talks
18mAutomating 90% of support tickets: How we built Vercel's support agent
50mOpening Keynote
25mBuilding high performance, AI-native engineering teams
56mOpening Keynote
46mBecoming an AI engineering company
21mAgents at Work
33mAutomating Resolution of IT Support with Chatbots
18m1,200 analytics requests a day: Clay and Vercel's Bet on AI-Native Analytics