The most terminal-native developers on the internet are publicly noticing entire workdays without opening a shell. That's not a lifestyle anecdote — it's a requirements document for the next class of dev tools.
There's a genre of confession spreading through dev circles this summer: the lifelong terminal person — tmux, SSH, stock git, the whole liturgy — who looks up one afternoon and realizes they haven't opened a shell all day. The loudest version came from Theo of t3.gg, who has argued publicly that for agentic work specifically, the command line has become a kind of skeuomorphism: a comfort object from the previous era, the way early phone apps dressed up as leather-bound calendars.
Strip the provocation and there's a precise engineering claim underneath. The terminal breaks down at three exact points, and all three appear the moment you stop running one agent and start running several:
Read as requirements rather than complaints, that list doesn't describe a better editor. It describes a control plane: persistent sessions that survive your laptop closing, a surface that renders whatever the agent produces, and first-class fan-out across many concurrent workers.
Which explains the current wave. Theo shipped an open-source, model-agnostic GUI for running agents across machines. Cursor pushes cloud agents you monitor rather than drive. OpenAI's Codex and Anthropic's Claude Code have both grown desktop and managed-cloud surfaces around what began as CLI tools. Different entry points, same destination: the pane of glass above the work, where a human directs and supervises instead of typing.
Nobody has won this surface yet — most teams are still bending editors and terminals to a job neither was designed for. That gap is the single clearest product opportunity in dev tools right now, and it's a design problem more than a model problem. The terminal isn't dying; it's being demoted to plumbing. The interesting question is what gets promoted.