This week, a Reddit post on the programmer community LocalLLaMA hit the front page with a question: why are more and more programmers abandoning traditional IDEs like Cursor and VS Code (the integrated development environment—the main interface where programmers write code), in favor of tools like OpenCode and Claude Code that let AI run commands directly in the terminal? What we care about is whether this shift in preference might be an early signal that AI Agents—AI that can autonomously execute tasks—are moving from chat-box assistants to directly operating systems.
The original poster actually uses both: large projects use Cursor or Zed, switching to terminal tools like OpenCode only when server configurations need editing. In other words, even the person raising the question isn't asking you to choose one—they're acknowledging that both workflows will coexist. What's actually worth discussing is: once AI gets shell permissions (OS-level execution rights), will developers still be willing to keep it boxed inside the IDE?
What This Is
Tools like OpenCode and Claude Code are called coding harnesses by developers—they don't offer pretty interfaces, just give the AI Agent an environment in the terminal where it can read files, write files, and execute commands. The difference from IDE plugins like Cursor or Copilot: plugins let AI write code under your watchful eye; harnesses let AI run the entire workflow itself and report back.
The original poster made it clear in their follow-up: large projects still use IDE mode; only system-level operations like "tweaking server configuration" go through the terminal. In other words, the essence of this debate isn't "which is more usable," but "should AI be kept in a box."
Industry View
The reasons supporting the terminal camp are straightforward: if AI can understand code and execute commands, adding an IDE layer as a middleman is redundant. This usage treats AI as a real Agent, not just smarter autocomplete. One Reddit commenter said: "Once you get used to letting AI run commands directly, you can't go back to clicking menus manually in the IDE."
Opposition centers on security and controllability. Granting AI shell permissions means deleting the wrong file, exposing secrets, crashing production environments—these are risks that IDE plugins are unlikely to create. A security engineer wrote in the comments: "What the compliance team thinks of these tools will determine whether they can be used in enterprises." This, we believe, is the objection more worth discussing—it's not a technical issue, it's a governance issue.
We note that the underrated third party in this debate is IDE vendors like JetBrains and Cursor. As AI Agents move to the terminal, their moat—"developers can't live without our interface"—gets bypassed. This isn't a product fight; it's a battle over direction.
Impact on Regular People
- For enterprise IT: when employees start using terminal AI tools to connect directly to servers, the previous "standardized development environment" approach gets bypassed. Security strategy must shift from "managing software" to "managing permissions and auditing."
- For individual careers: the gap between "knows how to use AI" will further widen into "dare to hand over the entire working environment to AI." This is no longer a technical question; it's a trust question.
- For the consumer market: terminal AI tools currently target mostly developers, but "AI computer operator" products being built by OpenAI and Anthropic are already moving in this direction. The time when ordinary white-collar workers get to use "AI clicks the mouse and modifies files for me" may arrive faster than expected.