When you give an agent the power to edit files, run commands, call MCP servers, create branches, and review changes, the hard part is no longer getting it to act. The hard part is keeping it pointed at the right target. The current core tools are Cursor and Claude Code.

The course baseline was refreshed on June 23, 2026 against official Cursor and Claude Code documentation and changelogs. Exact model menus, pricing, and version numbers will continue to change, so model-specific claims are dated or phrased as volatile.

The tools will change. The habits that keep you out of trouble are more durable:

• Give the agent the files, rules, and verification commands that define correctness.
• Use rules for always-on constraints and skills for repeatable workflows.
• Keep permissions explicit, especially for shell commands, MCP tools, cloud agents, and automation.
• Review diffs and run the quality gates yourself before shipping.
• Persist durable knowledge in the repository, not in a chat transcript.

You should finish this course with a working mental model for when to use inline edits, chat, local agents, cloud agents, subagents, hooks, automations, MCP, and continuous integration integrations.