Coatue has released another insightful chart, this time showcasing the usage of AI programming tools.
On the left, we see the NPM download numbers for Claude Code, while on the right, the installation figures for OpenAI Codex's VS Code plugin are displayed.
The shapes of the two curves are strikingly similar—gradual increases, then a sudden leap at a certain point.
Claude Code saw a staggering 70-fold increase in just a year. After the releases of Sonnet 4 and Opus 4 in May 2025, downloads began to climb. By November, when Sonnet 4.5 launched, the curve steepened dramatically, jumping from around 80 million to nearly 200 million.
Keep in mind, these aren't just user counts; it reflects cumulative NPM downloads—developers are actively using this tool in real-world coding environments.
On the other hand, Codex experienced a 7-fold increase in just six months. Before the release of GPT-5 Codex in September 2025, the VS Code plugin had around 600,000 to 700,000 installations. Post-launch, that number skyrocketed, exceeding 5.4 million by May 2026.
When we look at both charts together, one thing becomes clear: the growth of AI programming tools isn't linear—it's stair-step. There might be slow growth over time, but every time a foundational model is upgraded, we see a significant jump in usage.
This makes perfect sense. Developers are pragmatic; they quickly assess whether a tool works for them. Once a model's capabilities exceed a certain threshold, word of mouth spreads rapidly.
However, the disparity in scale between the two is noteworthy. Claude Code boasts 200 million downloads, while Codex has racked up 5.4 million installations—a nearly 40-fold difference.
Of course, these metrics aren't entirely equivalent—NPM downloads include automated downloads from CI/CD processes, while VS Code installations are purely manual. But even accounting for that, Claude Code's penetration within the developer community is clearly higher.
This aligns with the previous charts: Anthropic's revenue has grown 14-fold, largely driven by usage from developers.
In short, the growth strategy for AI companies has shifted. It's no longer about marketing to acquire users; it's about leveraging model capabilities to "unlock" new use cases. When a model becomes powerful enough, users will come on their own. If the model isn't strong enough, no amount of marketing will help.
This represents perhaps the healthiest growth model in the AI industry—product-driven rather than marketing-driven.
Each model upgrade serves as a signal for growth! 🌟
Top comments (0)