TypeScript became the most-used language on GitHub in August 2025. The 66% year-over-year contributor growth — the biggest language ranking shift in more than a decade — is presented as a signal of developer preference. It's also a vote that a lot of developers didn't consciously cast.
The numbers are specific: roughly 2.6 million monthly contributors as of August 2025, surpassing both Python and JavaScript simultaneously. 40% of developers now write exclusively in TypeScript, up from 34% in 2024. Luau, Roblox's gradually typed language, grew 194% in the same period. Typst, a strongly typed document language, jumped 108%. Shell scripting in AI-generated projects surged 206%.
The pattern across those last four numbers is obvious if you're looking for it: every standout gainer is either strongly typed or AI-generated. That's not a coincidence.
The 94% Problem
A 2025 academic study found that 94% of compilation errors produced by LLMs are type-check failures. The models got the logic approximately right, but they produced something the type checker rejected.
This is not a defect in the models. It's how the underlying mechanism works. When you're generating code for a dynamically typed language — Python, JavaScript — the model has to infer what types values will be from context. It can get the runtime behavior wrong in ways that only surface during execution, often in production, often inconsistently. When you're generating TypeScript, the type annotations are explicit, machine-readable constraints. The model generates code that satisfies those constraints, and the compiler validates it immediately.
TypeScript, from the AI's perspective, is a language with more legibility. More signal to work from, more immediate feedback when the output is wrong. The 94% figure is evidence of this: 94% of LLM errors are exactly the kind that a type system catches. A type system is, among other things, an AI mistake detector.
What this means in practice: agentic sessions in TypeScript have fewer correction loops. Generate, type-check, done — or generate, type error, one revision, done. The same session in Python or JavaScript can involve runtime testing, debugging, multiple reprompts, discovering errors that only appear in edge cases after you thought you were finished. The subjective experience of AI-assisted coding is better in TypeScript, even when the underlying problem is identical.
The Loop
This is the feedback loop GitHub's researcher Ori Gazit named in the Octoverse data: "If an AI tool is going to generate code for me, I want a fast way to know whether that code is correct. Explicit types give me that safety net."
The mechanism is self-reinforcing. TypeScript's type system makes AI-generated code more reliable → developers who use AI tools prefer TypeScript → more TypeScript on GitHub → more TypeScript in training data → even better AI TypeScript generation → more developer preference. Every iteration strengthens the effect.
This loop has been running for roughly two years. It's not the only reason TypeScript is #1. Next.js defaults to TypeScript. Astro shifted there. The general industry trend toward type safety predates AI coding tools. But those factors don't explain 66% growth in a single year.
InfoQ's reporting on the Octoverse data in March 2026 called this the "convenience loop": when a technology feels particularly frictionless with AI assistance, developers gravitate toward it, which generates more training data, which makes the AI better at that technology, which attracts more developers. The convenience loop is running on TypeScript right now. It's starting to run on Rust and Go.
The Vote You Didn't Take
80% of new developers on GitHub now use Copilot within their first week. These developers are forming language preferences while the feedback loop is already running. Their "native" language — the one that feels natural, the one they reach for by default — is being shaped partly by which language their AI assistant handles best.
This doesn't mean those preferences are wrong. TypeScript has genuine technical merits beyond AI compatibility. But the preference is being formed through a mechanism that most developers don't notice, because it operates through the texture of daily work rather than through explicit reasoning.
Here's the version of the decision that might actually be happening: a developer tries TypeScript on a new project because the starter template defaulted to it. They start using Copilot. The experience feels smoother than their Python sessions. They attribute this to TypeScript being a better language for this kind of work. Over months, the preference solidifies. They call themselves a TypeScript developer.
Some part of what they're calling TypeScript's technical merit is actually AI tool quality. Those aren't the same thing.
What the Pattern Predicts
Shell scripting grew 206% in AI-generated projects specifically because AI tools generate reliable shell scripts, and developers increasingly use agents for automation tasks they wouldn't have tackled manually. The pattern is the same: AI makes a language more accessible → adoption rises → training data improves → accessibility increases further.
The losers in this dynamic are languages where AI generates with lower reliability — where the correction loop is longer and the daily experience is more frustrating. That doesn't mean those languages disappear. Python remains the dominant language for ML pipelines and data science, where the ecosystem is irreplaceable regardless of AI tool quality. But in general-purpose application development, where TypeScript and Python genuinely compete, the contest is increasingly being decided by AI tool performance, not by language properties.
What to Track
If you're using AI coding assistants heavily and you track your coding time, look at your language distribution over the past year. If TypeScript has been quietly gaining share, it's worth asking how much of that is a technical choice and how much is the convenience loop.
The question isn't whether TypeScript is the right choice — it might be. The question is whether you're choosing it for reasons that will hold if AI tool quality across languages converges, which it will gradually as training data normalizes. A stack choice made because "my AI assistant is better at it" is a different technical bet than a stack choice made because "this language fits the problem."
Most developers who've shifted to TypeScript in the past two years have a story about why. The story is usually about type safety, editor support, and scale. What's less in the story is the AI feedback loop that made TypeScript feel demonstrably better every time they opened their coding assistant.
TypeScript deserves to be #1. It also got there partly by being the language that gave AI tools the most information to work from. Knowing that changes how you evaluate the next time someone argues TypeScript is the obvious choice — and gives you a template for thinking about which technology is next in line for the same treatment.