GitHub Octoverse 2025: TypeScript Overtakes Python as AI Reshapes Development

TL;DR

• TypeScript became the most-used language on GitHub in August 2025, overtaking Python and JavaScript for the first time in over a decade
• 180 million developers now build on GitHub—one new developer joined every second in 2025, with 80% using Copilot in their first week
• India overtook the US as the largest contributor base to open source projects and will account for one in three new developers by 2030
• AI projects exploded: 1.1M+ repositories now import LLM SDKs (+178% YoY), and coding agents created 1M+ pull requests between May-September 2025

The Big Picture

GitHub's 2025 Octoverse report isn't just another year-over-year stats dump. It captures the moment AI stopped being a novelty and became infrastructure. TypeScript dethroning Python and JavaScript marks the most significant language shift since GitHub started tracking this data—and it's not random.

The numbers tell a story about convergence. More than 36 million developers joined GitHub in a single year, the fastest absolute growth rate on record. Nearly 80% of them used GitHub Copilot within their first week. That's not adoption—that's expectation. AI-assisted coding is now the default entry point for new developers.

But the real shift isn't just speed. It's choice. Developers are gravitating toward typed languages that make agent-assisted code more reliable in production. TypeScript's rise reflects a pragmatic bet: if AI is going to write half your code, you want a type system catching errors before they ship. Python still dominates AI and data science workloads, but TypeScript's 66% YoY growth signals where greenfield development is headed.

This isn't hype. It's structural change. The data shows AI influencing not just how fast code gets written, but which languages and tools developers trust to take AI-generated code into production.

How TypeScript Won

TypeScript didn't just edge out Python—it added over 1 million contributors in 2025 alone, hitting 2.6 million monthly contributors by August. That's a 66% year-over-year jump. Python grew fast too (+850k contributors, +48% YoY), but TypeScript's momentum came from a different place: frameworks.

Nearly every major frontend framework now scaffolds with TypeScript by default. Next.js 15, Astro 3, SvelteKit 2, Qwik, SolidStart, Angular 18, Remix—all generate TypeScript codebases out of the box. Run npm create or pnpm dlx, and you're in a typed environment before you write a single line.

The technical reason matters: type systems reduce ambiguity. A 2025 academic study found 94% of LLM-generated compilation errors were type-check failures. When AI writes your code, types catch mistakes before they hit production. That's not theoretical—it's why teams are choosing TypeScript for new projects.

Python still leads AI and ML work. Jupyter Notebooks remain the go-to for experimentation (403k repositories, +17.8% YoY in AI-tagged projects). But when it's time to ship, TypeScript's typed contracts scale better across AI-assisted teams. The data backs this up: TypeScript led new repositories in 2025, with 5.4 million created in the past year (+78% YoY).

JavaScript isn't dead—it's still massive at 2.15 million contributors. But its growth slowed to 24.79% YoY as developers shifted toward TypeScript. The combined JavaScript/TypeScript ecosystem still dwarfs Python in total activity, but the trend is clear: typed languages are winning.

What This Changes For Developers

The shift to TypeScript isn't just about syntax. It's about workflow. AI tools like Copilot and coding agents work better with typed languages because they can infer context from type definitions. That means fewer hallucinations, faster autocompletions, and more reliable refactors.

This shows up in the data. Developers merged 43.2 million pull requests per month in 2025 (+23% YoY). Code pushes hit nearly 1 billion commits (+25.1% YoY), with a record 100 million in August alone. Activity spiked after GitHub previewed its coding agent in March and introduced code review in April. By July, developers closed 5.5 million issues—1.4 million more than the prior month.

The productivity gains are real, but they're not evenly distributed. Projects that configure Dependabot with auto-merge rules remediate vulnerabilities faster than those relying on manual review. Average fix times for critical severity vulnerabilities dropped 30% (from 37 to 26 days), and 26% fewer repositories received critical alerts. Automation is working—when teams actually use it.

Security is the other side of this coin. Broken Access Control overtook Injection as the most common CodeQL alert, flagged in 151k+ repositories (+172% YoY). Much of this stems from misconfigured permissions in CI/CD pipelines and AI-generated scaffolds that skip auth checks. Authentication and authorization remain hard for both developers and LLMs. Copilot Autofix helped—developers accepted AI-generated fixes for Broken Access Control in 6,000+ repositories per month by mid-2025.

The geographic shift matters too. India added 5.2 million developers in 2025 (14% of all new accounts) and overtook the US as the largest contributor base to open source projects. By 2030, India is projected to account for one in three new developers globally. Brazil, Indonesia, and Nigeria are also growing fast, driven by remote hiring, fintech booms, and mobile adoption.

The AI Infrastructure Boom

AI projects aren't just growing—they're exploding. More than 1.1 million public repositories now import an LLM SDK (+178% YoY), with 693k created in the past 12 months alone. Monthly contributors to generative AI projects climbed from 86k in January 2025 to a peak of 206k in May (+132% YoY), then held near 200k through summer.

The fastest-growing open source projects reflect this. Six of the top 10 by contributor growth were AI infrastructure: vllm (high-throughput LLM inference), ollama (local model runner), ragflow (RAG orchestration), llama.cpp (lightweight local inference), and Cline (LLM-native command-line shell). These aren't demos—they're production runtimes.

Coding agents entered the mainstream. GitHub's coding agent created 1 million+ pull requests between May and September 2025. Early data shows it's being used in established projects, not just throwaway experiments. Repositories with at least one coding agent-authored PR tend to have more stars, larger size, and greater age. Teams are testing agents on real work.

Python still anchors AI development (582k repositories, +50.7% YoY), but the shift toward production is visible. Jupyter Notebook growth flattened mid-2025 while Python-based code accelerated, signaling teams are packaging experiments into deployable stacks. TypeScript and JavaScript also grew sharply in AI projects (+77.9% and +24.8% YoY), reflecting the rise of demos, dashboards, and lightweight apps built around model endpoints.

Shell scripts saw the fastest rise (+324%), reflecting how teams codify eval harnesses, data prep, and deployment pipelines. C++ crossed 7,800 repos (+11%), a reminder of its role in performance-critical inference engines. Dockerfiles appeared in 1.9 million repositories (+120% YoY), likely fueled by the need to sandbox agents and LLMs securely.

Try It Yourself

If you want to see where this is headed, start with the tools developers are actually using. The fastest-growing projects by contributors tell the story:

• vllm-project/vllm — High-throughput LLM inference engine for production workloads
• ollama/ollama — Local model runner that makes it trivial to run Llama, Mistral, and other models on your machine
• cline/cline — LLM-native command-line shell that reasons over local context
• astral-sh/uv — Blazing-fast Python package installer and resolver written in Rust
• infiniflow/ragflow — End-to-end RAG orchestration framework

For a practical starting point, check out GitHub Agent HQ, which shows how multi-agent orchestration is being built directly into developer workflows. If you're curious about how GitHub evaluates AI tools internally, their MCP server testing methodology offers a deep dive into offline evaluation.

The Model Context Protocol (MCP) hit 37k stars in just eight months, showing how fast standards are emerging. If you're building anything that touches LLMs, understanding MCP is worth your time. It's becoming the interoperability layer for agent-assisted development.

The Bottom Line

Use TypeScript if you're starting a new project that will involve AI-assisted development or need type safety at scale. The tooling is mature, the ecosystem is massive, and every major framework defaults to it. Skip it if you're deep in Python ML work or maintaining legacy JavaScript—there's no reason to rewrite working code.

The real risk isn't picking the wrong language. It's ignoring the shift toward typed systems and AI-native workflows. Developers who treat AI as a curiosity rather than infrastructure will find themselves debugging untyped LLM output while their peers ship faster with typed contracts and automated fixes.

The opportunity is geographic. India, Brazil, Indonesia, and Nigeria are adding millions of developers who are learning AI-first workflows from day one. If you're hiring, building open source, or launching a developer tool, these markets aren't emerging—they're already here. One in three new developers by 2030 will come from India alone.

AI didn't replace developers. It brought 36 million more into the ecosystem in a single year. The question isn't whether AI will change how you code—it already has. The question is whether you're building for the workflows that are winning.

Source: GitHub Blog