The agent CLI space is getting pretty crowded.

Claude Code and Cursor are typically my daily drivers, but I played around with Sourcegraph’s amp and Google’s gemini-cli.

In particular, I noticed that gemini-cli has been getting a lot of attention.

If I had to guess, Amp seems to use Anthropic’s text editor tool which is quite nice, and what today’s best agents seem to consistently be built on top of. It seems unlikely gemini-cli would use this tool as it uses Google’s Gemini models.

These tools are all copying from each other, liberally. Like Claude Code, gemini-cli uses LLM-generated phrases while it runs its LLM calls in the background to do work for you.

Amp feels more minimal whereas Gemini CLI feels an attempt at an outright Claude Code clone. Having used Gemini 2.5 Pro for several weeks just before Sonnet and Opus 4 came out, this makes sense. Gemini is a great model for writing code and deserves a dedicated tool.

It also seems Google has gotten some very wide distribution with the launch.

It would be nice to see standardization on a file name for agent rules. agent.md seems like a fine solution for this. This agent rules format seems like yet another dimension on the MxN problem of everything in models these days — supporting everything eventually becomes an entire project or business (see OpenRouter, LiteLLM).

Hopefully, we see some consolidation here soon or we can just create a bunch of symlinks. Amp seems to try to be compatible will all existing agent files and import from them. I’ve not seem much commitment to this from other tools.

Notably, it seems only Cursor has deviated from the standard markdown format They use .mdc which has frontmatter to instruct the agent when to attach the rules.