Microsoft CEO told me to build this: Coding with Agents: Building an AI Inbox to Handle Your Work

If you’re a developer looking to learn how to build fullstack AI-powered applications with real-time collaboration, multi-agent orchestration, and dynamic task automation—this is the deep dive you’ve been waiting for. Inspired by Satya Nadella’s vision of an “agentic inbox,” this video walks you through how Agent Inbox was designed and built—from its backend schema to its frontend user flow—using Convex.dev as the realtime database and serverless backend. You’ll see how agents collaborate, schedule future actions, trigger GPT-4 tasks, and even send emails… all while being powered by well-structured TypeScript and a clever real-time architecture You’ll learn: • How to build a multi-agent AI platform from scratch • Real-world challenges with prompt design, AI evals, and function orchestration • Convex folder structure best practices for scalable apps • How to debug and optimize agentic interactions • How to use OpenAI’s GPT-4 via the Vercel AI SDK • Building and caching UI state in a type-safe Vite + React app 🛠 Resources Try Convex (the backend used here): https://convex.link/agentinbox Source code available here: https://github.com/mikecann/agent-inbox Interview with Satya Nadella: https://www.youtube.com/watch?t=1755&v=4GLSzuYXh6w&feature=youtu.be) ⏱ Timestamps 00:00 – Why Satya Nadella inspired this build 03:13 – Agent conversations & prompt orchestration 06:43 – Frontend architecture with Vite + React 09:55 – Backend architecture with Convex 14:45 – Fixing agent message bugs + generating text with GPT-4 17:19 – Using Vel SDK + tools like noOutput and scheduleTask 21:14 – How agent roles and message replay work 24:23 – The final demo: multi-agent collaboration + movie script generation #AIagents #FullstackDevelopment #ConvexDev #GPT4 #TypeScript #React #Serverless #AIAutomation #DevTools #OpenAI #AgenticApps #DeveloperTools