Claude Code Is Writing Itself: Anthropic CEO Reveals AI Is Now Its Own Developer
Claude Code holds over 50% market share, Anthropic's ARR has surpassed OpenAI at $44 billion — but the real headline is Dario's bombshell: Claude's newest features were almost entirely developed by AI itself. Programming Agents aren't just tools anymore — they're building themselves.
Claude Is Writing Its Own Code
On May 16, Anthropic CEO Dario Amodei confirmed in an interview: Claude's latest features were developed "almost entirely autonomously by AI."
This isn't a demo. This isn't a proof of concept. This is production workloads in the real world. Claude's team used Claude Code to write Claude's new features — AI iterates on itself.
This isn't an isolated case. More evidence shows programming Agents are moving from "human assistants" to "autonomous developers":
| Metric | Data |
|---|---|
| Claude Code market share | Over 50% |
| Anthropic ARR | $44 billion |
| Anthropic valuation | $900B, surpassing OpenAI |
| Claude Code user growth | "Unprecedented" |
Why Claude Code, Not Others?
The programming Agent space is already crowded — Cursor, Codex, Windsurf, Qoder... But Claude Code leads on three fronts:
1. Context Window Dominance
Claude 4.x's context window is effectively 2-3x competitors' in code tasks. For large codebases, this means less "forgetting" and higher accuracy.
2. Deep Code Understanding
Claude Code doesn't just "complete code" — it understands project architecture and proactively modifies. It reads your 100K-line codebase and edits the right place.
3. Anthropic's "Safety Brand"
An invisible factor driving enterprise adoption: Anthropic's "safety-first" narrative eases CIO concerns. Even if actual security differences are negligible, brand perception is the purchase decision.
Behind the Surge: Tech Breakthrough or Market-Buying?
In early May, Claude Code suddenly doubled its quotas: Max 20x from 900→1800, Max 5x from 225→450, removing peak-hour reductions.
Claude Code's API call volume surged 17x year-over-year — unprecedented growth for any SaaS product. But questions linger:
Optimists: Tech maturity justifies removing limits. 17x growth proves real demand. Skeptics: Burn cash for market share first, monetize after developers are locked in. Classic internet playbook.
Reality likely sits between: Claude Code does lead technically, but the timing of the surge (right after Codex mobile launched) is unmistakably competitive.
The Endgame Isn't "Who's Better"
The ultimate competition among programming Agents isn't "who writes better code." It's "who enters the developer's daily workflow."
- Cursor owns the IDE
- Codex owns the ChatGPT ecosystem
- Claude Code owns the terminal and command line
The likely future: three Agents for three scenarios, developers switching on demand. But the one with the largest context window and strongest code understanding has a natural edge.
In one sentence: Claude is writing itself — AI evolving from "helping humans write code" to "helping AI write code." The endgame isn't who's strongest, but who becomes the workflow developers can't leave.
The AI Agents column tracks the latest Agent product dynamics. Follow us to catch every product inflection point.