Codex, Claude, and Kimi Are in a War (and We Win)

May 31, 2026 · observation

The AI coding tool space in the past few months has been absurd. OpenAI shipped Codex as a background agent. Anthropic has Claude Code running in terminals. Kimi dropped K2 with massive context and aggressive coding benchmarks. Everyone's shipping weekly.


What Each Does Well

Codex (OpenAI): Background execution. You describe a task, it spins up a sandbox, works asynchronously, and comes back with a PR. Good for "go do this thing, ping me when done." Less good for interactive pairing.

Claude Code (Anthropic): Terminal-native, interactive, deep codebase understanding. Reads your files, understands your patterns, works WITH you in real time. Best for "let's figure this out together" and for complex multi-file changes where context matters.

Kimi K2 (Moonshot): Huge context window, aggressive on benchmarks, surprisingly good at following complex instructions. The dark horse that keeps showing up in coding evals punching above its weight.


The Competition Is the Point

Six months ago, AI coding tools were "nice autocomplete." Now they're architecting systems, debugging hardware issues, writing test suites, and reviewing PRs.

This didn't happen because one company had a breakthrough. It happened because three+ companies are in a dead sprint, each one's release forcing the others to ship faster. Codex ships → Claude Code ships multi-file editing → Kimi ships bigger context → cycle repeats.


What I Actually Use

Honestly? I switch between them depending on the task:


The Meta Observation

Competition in AI coding tools is doing what competition always does: making everything better faster. Every tool I use today is dramatically better than it was 3 months ago. The pace is unsustainable for the companies but incredible for developers.

We're in the golden age of "companies burning money to give us better tools." Enjoy it while it lasts.


← back to blog