Claude Code tutorial: connectors, subagents and skills

This tutorial introduces Claude Code as a coding assistant and a tool for broader research and automation tasks. The speaker compares Claude's chat, co-work and code interfaces, then discusses model selection and effort settings. He names Opus 4.8 and Sonnet 4.6 and recommends higher effort for his own work; these are the video's labels and preferences.

The prompting advice centers on stating an intended outcome and asking Claude to interview the user before starting. Connector examples include Gmail and Google Drive. A Firecrawl demonstration requests website brand information and a logo. The speaker attributes lower token use to extracting relevant information rather than full HTML, but provides no measured comparison.

Later sections explain parallel subagents for company research and describe an adversarial debate workflow the speaker calls Ultra Code, whose identity and availability are unverified. He describes Graphify as supplying relational codebase maps that help Claude locate files; its exact project identity is not established. Reusable skills package task instructions and can come from a repository or a completed workflow.

The final sections cover connector permissions and least access, followed by a promoted Claude Code operating system that reportedly tracks spending and suggests background work. Its setup is deferred to another video. This tutorial does not demonstrate how to run models locally.