I'm trying claude code tomorrow

Today I'm going to try Claude Code for work. Here is the thing: I'm an avid user of 'Cursor'. This editor uses the vscode source. Now that's barely sufficient for serious use, but bear with me. We won't be doing old-school manual editing of files. I'll rant about it now because why not: barely sufficient for serious use, get used to crashes) and adds a agentic chat to it (and other features I do not find worth mentioning.) The agentic chat lets me connect MCP servers, index code bases, ask questions, run edit/exec commands, control automated browsers, run parallel bake-off agents in the cloud, run simple long running agentic chats on the web (which control a cloud based agent.) It's far more manual and less capable that Devin ai, but at a tiny fraction of the cost (and with a tiny fraction of the autonomous capability.) If you are interested in learning how to use LLMs in a deep way, try coding with a tool like Cursor exclusively for a project. I will give you my one killer tip right away. This is a rule I always force myself to follow now. It's been a game changer. Agentic coding agents are good at taking some existing code and enhancing it, or extending it, or adapting it. Start by finding 1-2 source repos (e.g. from github) that are similar to your project. Even if the only similarity is the architecture, or the product, or the language and tool: repos that have that information can help the agent immensly. Next, define your starting architecture enough that you have a list of dependencies. Kick off your project by installing all of those dependencies. Then checkout the relevant source code repos for each dependency, and checkout the correct tag too. Give your agent access to all of these repos. In Cursor I can do this by adding each one to my workspace. This causes Cursor to index all of those code files.