How AI is Changing Programming Language Popularity in 2026
The emergence of LLM coding assistants in 2022–2024 is the most significant exogenous shock to programming language popularity since the mobile revolution. It is changing how developers choose languages, how beginners learn, and how companies staff engineering teams. The LangPop data shows early signals of structural shifts — some confirming intuitions, some surprising.
The three forces AI applies to language popularity
AI coding assistants do not affect all languages equally. Three distinct mechanisms determine which languages benefit, which face pressure, and which are largely unaffected.
Force 1 — Learning barrier compression
The steepest adoption barriers for languages like Rust and TypeScript have always been cognitive: the borrow checker, the type system, the unfamiliar idioms. When a beginner hits a wall, they leave. AI assistants change this equation materially. A developer can now paste a confusing error message into an LLM and receive a plain-English explanation within seconds. Code that previously required weeks of documentation-reading to understand can be explained on demand.
The effect is asymmetric: languages whose adoption barriers were primarily cognitive (Rust, TypeScript, Haskell) benefit more from this force than languages whose barriers were ecosystem or tooling problems. A beginner learning Rust with Copilot today has a meaningfully different experience than a beginner learning Rust in 2020.
Force 2 — The training data flywheel
LLMs are trained on code. Languages with more publicly available code produce better LLM assistance for that language, which attracts more developers, which produces more code, which improves the next generation of models. This is a compounding advantage, and Python currently sits at the top of it.
The flywheel is real but not absolute — TypeScript and JavaScript have enormous training corpora too, and even Rust has enough public code on GitHub for modern models to write competent idiomatic Rust. The primary beneficiary is Python; the languages at the trailing edge of the flywheel are those with small or largely non-English codebases.
Force 3 — Stack Overflow displacement
Stack Overflow publicly reported a 78% decline in question volume in 2023. Developers who previously posted a question and waited for an answer now get an immediate response from an LLM. This has direct consequences for popularity indices: Stack Overflow question volume has been used as a proxy for developer engagement and language interest for over a decade. That signal is now severely distorted. Languages that were disproportionately represented on Stack Overflow — Java, C++, and legacy enterprise languages with large, experienced communities who documented everything there — lose measured weight relative to their actual use. Python, which has extensive documentation spread across its own docs site, RealPython, Jupyter notebooks, and research papers, is less exposed to this measurement shift.
How LangPop responds to this: We reduced Stack Overflow's weight from 20% to 15% in our composite formula. Stack Overflow remains a valid signal of developer engagement — just not the signal it once was. TIOBE, which derives its index from search engine query volumes, and PyPL, which tracks tutorial searches, face their own measurement distortions as developers substitute LLM queries for search queries. No index is clean; transparency about the distortion is the best available response.
Python: the clear winner
No language benefits from the AI shift the way Python does. The reinforcing mechanisms are unusually strong:
Every AI SDK ships Python first
OpenAI, Anthropic, Google, Mistral, Cohere — every major LLM provider ships a Python SDK before anything else. The first working examples are Python. The tutorials are Python. The research papers are Python. If you are building on top of LLMs professionally, Python is the path of least resistance.
LLMs write Python better than any other language
The training data flywheel has run longest and deepest for Python. Models produce more idiomatic, correct, and well-documented Python than any other language. This is a practical advantage for developers using AI assistance day-to-day: Python code generated by Copilot or Claude is more reliable than equivalent output in less-represented languages.
Building AI products requires Python
The ML stack — PyTorch, TensorFlow, scikit-learn, LangChain, Hugging Face Transformers — is Python. Every team building a product on top of AI models has Python in their stack. This is not a trend; it is a structural dependency that will persist as long as the current ML framework ecosystem remains Python-centric.
AI assistance makes dynamic typing less risky
Python's dynamic typing has historically been cited as a maintenance liability in large codebases. AI coding assistants partially compensate for this: they can infer types from context, flag likely type mismatches, and generate type annotations. Python's adoption concern becomes less significant when you have an AI pair that can catch what the interpreter misses.
In the LangPop composite, Python ranks #1 with a score that is widening its lead over JavaScript (#2). The gap is not accidental — it reflects a genuine structural shift in where developer activity is growing fastest.
JavaScript and TypeScript: reinforced, not threatened
When AI coding tools became widely available, a reasonable concern was that they might reduce frontend demand — if AI generates UIs, you need fewer frontend developers. The data does not support this fear.
AI tools that generate UIs — v0 (Vercel), Bolt (StackBlitz), Lovable — generate TypeScript and React. More frontend gets built than before, not less. The demand for the output has increased even if the time-per-unit has decreased. Crucially, the stack that AI coding tools default to when generating a new web application is Next.js and TypeScript. AI is not replacing JavaScript; it is standardising on it.
TypeScript's type system provides an additional benefit in the AI-assisted development context: it creates a structure that AI can reason about and validate against. AI-generated TypeScript code can be checked by the compiler in a way that AI-generated JavaScript cannot. Developers who have adopted TypeScript report that AI assistance is more reliable in typed codebases — the model knows what types are expected and can self-correct when it produces incompatible output.
LangPop view: TypeScript's rising score in the index reflects this reinforcement. It now sits at #3 in the LangPop composite, tracked separately from JavaScript because its adoption pattern and job market are meaningfully distinct. The JS → TS migration that has been underway for five years is accelerating under AI, not pausing.
What the Stack Overflow decline means for measuring language popularity
Stack Overflow's 78% drop in question volume in 2023 is not a data quality problem — it is a measurement signal that the community's behaviour has changed. Developers still need answers; they are getting them elsewhere. The question volume that remains on Stack Overflow is increasingly the hard, nuanced, ecosystem-specific question that LLMs genuinely struggle with. The easy questions — which previously accounted for the bulk of volume — have migrated to LLMs.
This matters for any popularity index that uses Stack Overflow data, including LangPop. The languages that most benefited from high Stack Overflow volumes were the ones with long histories and active question-answering communities: Java, C#, C++, PHP. These communities were productive on Stack Overflow in part because their questions were complex enough to require experienced human answers — but now even complex questions are being asked of LLMs first.
TIOBE derives its index from the number of search engine results for “[language] programming” — a different form of measurement distortion as developers substitute LLM queries for search queries. PyPL tracks tutorial searches in Google. All three indices are being disrupted by the same underlying shift: developers are getting answers differently, which means the observable signals of language engagement are changing. There is no distortion-free index. Transparency about the distortion is the best available approach.
Languages that AI is making more accessible
Several languages have seen their adoption barriers reduced by AI assistance in ways that are already visible in the LangPop data.
Rust
The borrow checker has been the single largest barrier to Rust adoption since the language launched. It is not just unfamiliar — it requires a genuinely different mental model of memory ownership. Developers learning Rust in 2019 hit the borrow checker and often quit. AI assistants have made this barrier meaningfully lower.
GitHub Copilot and Claude Code can explain borrow checker errors in plain English, suggest the fix, and explain why the fix resolves the ownership conflict. This is not a replacement for understanding ownership — it is a tutoring layer that reduces the frustration peak that causes learners to abandon the language. Rust's improving score in the LangPop composite reflects this: GitHub activity is up, Reddit discussion is more optimistic, and the job postings that require Rust experience are increasing as more teams move systems-level work from C++ to Rust.
Go
Go's adoption barrier is lower than Rust's, but idiomatic Go has always had a learning curve — particularly around goroutines, channels, and error handling patterns that differ from Python or JavaScript conventions. AI writes idiomatic Go well, which means developers evaluating Go for a new backend service can prototype more quickly. The practical effect is that developers who would have defaulted to Python for a new backend microservice are more likely to start a Go project when AI assistance closes the initial unfamiliarity gap.
TypeScript
AI coding tools default to TypeScript when generating new web projects. A developer who has been writing plain JavaScript and has delayed adopting TypeScript now encounters TypeScript as the AI-generated starting point for any new Next.js project. The JS-to-TS migration that teams previously had to decide to undertake is increasingly happening as a default — accelerated by AI tooling that starts with types rather than adding them later.
Languages AI is pressuring downward
Not all pressure is negative — some is the natural consequence of better alternatives becoming accessible. But for some languages, AI is accelerating a decline that was already underway.
COBOL
AI-assisted translation from COBOL to modern languages is improving faster than most expected. Banks and government agencies that previously faced an enormous cost and risk barrier to COBOL modernisation are starting AI-assisted migration projects. This accelerates the replacement velocity for the language. The COBOL developer shortage — which previously preserved COBOL's relevance through scarcity — is becoming less relevant as AI reduces the human expertise required to understand and translate existing COBOL code.
PHP
PHP's historical advantage was that it made server-side rendering accessible to developers who could not afford a separate backend team. Next.js, with AI assistance making it accessible to developers who previously needed PHP's simplicity, is absorbing some of this use case. This is a slow displacement, not a collapse — PHP powers enormous amounts of the web and has a healthy ecosystem. But developers evaluating a new web project are less likely to choose PHP when AI makes the Next.js/TypeScript path equally accessible.
Languages that depended on being "the learnable option"
Visual Basic, ColdFusion, and similar languages built adoption partly on being the accessible entry point for non-specialist developers. When AI makes any language approachable, the "simplicity" differentiation evaporates. The reason to learn a language must now be its job market, its performance characteristics, or its ecosystem — not that it is easier to start with.
What LangPop's data captures — and what it misses
Our seven data sources measure developer activity and market demand. They do not directly measure “what AI writes” — and this is an increasingly important distinction as AI-generated code becomes a meaningful share of GitHub commits.
| Source | AI signal quality |
|---|---|
| GitHub activity (25%) | Increasingly noisy — AI-generated commits are indistinguishable from human-authored commits. Python and TypeScript are inflated by AI coding assistant output. |
| Job postings (20%) | Clean and lagging — job postings reflect employer demand for human skills, not AI preferences. Lags the real market by 12–24 months as hiring plans adjust slowly. |
| Stack Overflow (15%) | Structurally disrupted — volume down 78% YoY. Signal has shifted from volume to quality: remaining questions are harder and more ecosystem-specific. |
| Google Trends (15%) | Distorted — developers increasingly ask LLMs instead of searching Google. Tutorial and "how to" query volumes are declining for all languages. |
| Package registries (10%) | Relatively clean — package downloads reflect actual usage in deployed applications. AI-generated code still runs packages; this signal is durable. |
| Reddit discussions (10%) | Leading indicator — community sentiment on Reddit is ahead of the market. Discussion of Rust, Go, and AI-adjacent topics tracks early adoption shifts before job postings move. |
| Tutorial platforms (5%) | Lagging and declining — new developers increasingly use LLMs rather than video courses. Platform enrolment is a weaker signal than it was in 2020. |
The honest position: LangPop measures human developer choices, with AI-generated code as growing noise in some signals. Job postings remain the cleanest signal of real-world language demand. Reddit discussions are the best leading indicator. GitHub activity and Google Trends require more interpretive caution than they did three years ago.
The 5-year forecast
Forecasting language popularity is a speculative exercise — ecosystems are sticky and predictions in this domain are routinely wrong. The following are considered positions based on current trajectories, not confident predictions.
Python's composite share will grow further
Basis: Every structural force discussed above continues to compound. AI SDK development, ML tooling, and training data flywheel effects all favour Python. There is no visible mechanism that reverses this within five years.
TypeScript will formally separate from JavaScript in most indices
Basis: It already has in LangPop. The job market increasingly distinguishes TypeScript-required from JavaScript-only roles. The languages have different adoption curves, different learning paths, and different employer signals. Tracking them together obscures more than it reveals.
Rust will crack the top 5 if AI keeps the borrow checker accessible
Basis: Rust's technical advantages — memory safety without a garbage collector — are increasingly relevant as security requirements tighten. If AI assistance maintains its ability to explain and resolve borrow checker errors, the adoption barrier stays low enough for Rust to continue climbing. If AI assistance plateaus, the barrier returns.
A new AI-native language is plausible but not imminent
Basis: Language designers are already thinking about languages optimised for LLM generation — more structured, more formally verifiable, more amenable to AI-assisted proof. Whether any such language achieves broad adoption is unknown. Developer ecosystems are sticky: Python has been dominant in ML for a decade despite technically superior alternatives. Momentum is hard to displace. The end of this decade is the earliest plausible window for a genuinely new language to register in popularity indices.
Current language rankings, source breakdowns, and composite scores are available on the full rankings page. The methodology behind LangPop's weighted composite — including how we adjusted Stack Overflow's weight — is explained in detail on the methodology page.
See how Python ranks across all data sources →
Python language page →