Getting Started: Development with Cursor

AI agents can write code faster than ever, but on a real project speed alone isn’t enough: without clear requirements, decisions and standards, an agent will happily build the wrong thing very quickly. This course teaches you how to work with Cursor the way a professional team works, by building LivePoll, a real-time voting system with ASP.NET Core, Postgres, Ruvio, nginx, a YARP BFF, Server-Sent Events and a React frontend. We begin with the Agents Window, Ask and Plan modes, Rules and .cursorignore, then move to a Spec-Driven workflow where the AI asks the questions and we write Functional and Non-Functional Requirements, define what is Out of Scope, and capture what the agent learns with a custom /learn Skill. From there, we design the system in plan mode and face a problem every company has with its internal libraries: the AI has never seen Ruvio, so we teach it through llms.txt and Skills, and measure what it actually learned.

Once the foundation is in place, the project’s knowledge does the heavy lifting. We record decisions as ADRs with Multitask and Cloud Subagents, compare options on a Canvas, design an OpenAPI Contract, and enforce our standards with AGENTS.md, TDD, Hooks and a Custom Subagent. We split the work into waves of tasks on GitHub through MCP, set up Permissions and the Sandbox, and build the system with several agents in parallel, in the cloud and in local Worktrees, including a hands-on Best-of-N comparison. When every test is green but the running system still fails, we hunt down the Integration Gaps between parts built by different agents. Then we bring the whole stack up with Docker Compose, let the agent test the app in Cursor’s Browser, polish the UI with Design Mode, and prove our requirements with Replicas, Crash Tests, a 1,000-voter Load Test and a live dashboard. Finally, we keep quality high with Debug Mode, Local Code Review, scheduled Automations, and by measuring the Context Cost of our rules and skills. By the end, you will have a repeatable way of building real software with AI agents, one where the AI proposes, you decide, and every claim is backed by a test.