Google Antigravity CLI - Run coding agents directly from your terminal
by•
Antigravity CLI brings multi-step reasoning, multi-file editing, tool calling, and persistent history to the terminal.
Optimised for SSH sessions and keyboard-first workflows.
For developers who live in the command line.
Replies
personally it used to be something id prefer using when it was similar to a IDE not a full agentic workflow, since i could review and edit then and there only like opencode or copilot integrations but things are turning too much to let a agent handle everything and having to have another editor app open anyways to see if it is doing intended way with safety or not.
I know things keep getting better to require less intervention but with the recent vulnerabilities spikes in so many applications that are made fully by AI, it seems having IDE style editor with integration would been a better middle ground
The shift from "prompt one agent per task" to "define a task and let agents run" is the real change here, and CLI/SSH-first matches how a lot of senior engineers actually work. At ~120 engineers I watched the per-context-switch cost dominate junior productivity, and most of those switches were "wait, what was the agent doing again?" not real engineering. The interesting failure mode is when two background agents step on each other's commits across a long-running task. How does the CLI handle conflict resolution when agent A and agent B both touched the same module overnight?
Coordinating multiple things simultaneously while building a project is genuinely hard. You're making architecture decisions, thinking about testing, planning deployment all at once. The idea of orchestrating different AI agents handling these different aspects from one place could make solo development feel a lot less overwhelming. Does it support custom agent configurations, or does it work with predefined workflows?