Developer ditches IDE autocomplete for CLI coding agents
A programmer describes a 12-month shift from JetBrains IDE autocomplete to command-line AI agents that read, modify, and test code directly—now using IDEs only for Git diffs and occasional debugging.
A developer tracking their workflow evolution over the past year reports a dramatic shift away from traditional IDE-based coding toward CLI-driven AI agents that handle most implementation tasks. The progression moved from in-editor autocomplete to context-aware chat prompts to standalone agents that modify source files on command.
The user describes three distinct phases. Initially, LLM-powered autocomplete in the IDE felt indispensable—start typing a function, let the model finish it based on open tabs and context. That gave way to a custom script that packaged selected source files into context for a chat interface, which the developer used for about six months. Now the workflow centers on a CLI coding agent: describe the task, use @ to point the agent at relevant files, and let it write or fix code without opening an IDE at all.
IDE use shrinks to 5-10 percent of work
The developer still maintains a JetBrains all-languages subscription covering Java, Scala, Python, TypeScript, and Rust, but only for three things: Git diff visualization, step-by-step debugging, and code navigation (clicking into function definitions). Those tasks now account for an estimated 5-10 percent of working time. Debugging itself is increasingly handled by giving an agent access to program logs; when information is insufficient, the agent is asked to add more logging, then re-diagnose the issue.
Test-driven workflows have also moved into the agent layer. The developer asks the LLM to write tests, run them, read failure logs, and iterate on fixes. When the agent loops without progress, the instruction becomes explicit: "Cover this with tests and read the logs to understand how everything works." For complex tasks, the developer now requests a plan.md file first—a structured work plan the agent writes and then executes step by step. The shift raises a practical question: as agent-driven workflows mature, the value proposition of traditional IDEs narrows to a handful of edge cases.