Takeaways from "You've been using AI the hard way"

• It's likely going to be better long-term to leverage CLIs as opposed to VS Code forks (i.e., Cursor IDE), web frontends (i.e., Claude), and even desktop apps like Copilot.
• Having all of your agent files locally and synchronized via things like Github means if better or different CLI tools come along, you can easily port your files around and make any necessary tweaks
• Some good tooling practices are going to become standards that should be applied, but since they're time-consuming they'll probably be left to the wayside:
  • Subagents for encapsulating certain tasks; running and validating builds, generating module templates, checking resources
  • Having agents document their "thoughts" and results to files to save context
• OpenCode might be the CLI that survives the test of time and industry because you can bring any LLM to it, including local LLMs