As I've been talking with engineers, more and more they are describing radically different ways of programming than they were a year ago. So I thought I'd put it to this community, what does your IDE, process or method look like post AI?
Built my own tui tool. It groups repos into projects and manages a task list for each repo. Creating a new task creates a new worktree, the task view contains two tmux sessions, one of which is a claude or codex session, the other a terminal. I have status dots for the current state of claude/codex and the current state of the pr if it exists. And i have a flow to review and annotate pr feedback and auto paste it back into my llm session with the instructions to address it and push. I have a task switcher which incorporates the status colors so i can tend to the red tasks, wait on the yellow ones, merge the green ones and close out the purple ones. Each task has a status that includes the last thing i typed to the llm. I also have hotkeys to launch cursor, embedded lazygjt, tower, or jump to the pr in the browser. I integrated linear and can create a task from the linear ticket and pull the ticket and all the comments into the context. People keep telling me i should add audio alerts, but i never have to step away. If everything is yellow i just start a new task.
vscode with claude and codex extensions. two running chats are fine for me, allows me to keep my eyes on the code rather than running 75 agents.