Claude Code

Updated 2026-04-04

CLI tool from Anthropic for AI-assisted programming directly in the terminal. Not an IDE plugin, not a web interface, but an agent that can read, understand, and modify code.

Why It Is Different

Claude Code differs from GitHub Copilot or Cursor in two major ways: it can access the full filesystem, not just the open file, and it can run commands on its own. That makes it less like autocomplete and more like a pair programmer that investigates, tests, and iterates.

The Rate-Limit Crisis (2026)

Claude Code became problematic for Anthropic for a simple reason: it was used too well.

On the $200 per month Max plan, users could realistically consume around $5,000 in compute, a 25x subsidy. That was not a bug, but a strategy to win users and create lock-in. The problem emerged when usage exploded and paying subscribers collided with enterprise demand and with internal researchers who also needed GPU capacity.

Anthropic’s response was peak-hour throttling. Between 5 a.m. and 11 a.m. Pacific Time, the main work window for enterprise customers, subscription limits are consumed faster than outside that window. Around 7% of users now hit limits they had not previously seen.

The criticism focused less on what changed than on how it was communicated: first via a tweet from one employee, after the change was already live, with no official dashboard, banner, or formal post.

OpenAI by Contrast

OpenAI has handled similar tension differently. Codex limits have often been reset around launches, bugs, and model updates. The public communication style feels almost opposite, which has led to very different community sentiment.

The Deeper Problem

Anthropic invested too late in GPU infrastructure. That is not only a technical issue, but a cultural one. The company was founded by researchers who primarily wanted to do research. Product and users were treated internally as financing necessities. That priority is now structurally visible in infrastructure decisions.

Connections