Trying
Got acquainted with Claude from Anthropic — both as a chat model (claude.ai) and as a terminal agent (Claude Code CLI). After my experience with Cursor and agents wired into the IDE, this is a different kind of interaction.
What sets it apart from what I've used before:
- Claude Code in the terminal — the agent works with the file system and commands, not just autocompletes code in the editor. This changes how you frame tasks: you state a goal, not a specific spot to edit
- Long context and care with instructions — the model is noticeably better at holding big project rules (like
CLAUDE.md) and doesn't lose agreements over long sessions - Tool use / MCP — native work with external tools via [[mcp]] opens the door to your own integrations
In trial: already using it on work tasks, still mapping out the boundaries — where it's reliable and where it needs manual oversight.