Programming languages in 2026 — what's growing, what runs fast, and what AI agents actually use well

Executive summary

The 2025–26 picture rearranges the top of the language stack in ways that look modest from inside a single index and dramatic across all of them. Python has overtaken JavaScript on every measure of new activity. TypeScript leapfrogged both on GitHub. JavaScript still has the largest installed base but the smallest growth rate of the major three. Among the established second-tier languages — Java, PHP, C++, Ruby — only C# is moving up: it is RedMonk's only top-10 mover this cycle (RedMonk, 2026) and TIOBE's "Programming Language of the Year" for 2025 (TIOBE, 2026).

On objective server-side performance, .NET sits much higher than its reputation suggests. ASP.NET Core leads the final TechEmpower round on the database-bound Fortunes test at roughly 610,000 requests per second — comfortably ahead of Go and Rust on that workload, and a clear order of magnitude past Node, with Python more than an order of magnitude behind that (TechEmpower, 2025).

On AI-agent coding fitness, .NET is the most uncomfortable story in this report. The standard agentic benchmarks (SWE-bench Verified, SWE-bench Multilingual, Aider Polyglot) all skip C# entirely. When Microsoft Research finally published a C#-native benchmark in November 2025, frontier models scored roughly 30 percentage points lower on C# than on equivalent Python tasks — a gap that has been roughly constant across model generations and is not explained by training-data scarcity (Yang et al., 2025).

In the labour market, a clear payroll signal has now appeared: workers aged 22–25 in the most AI-exposed occupations — software engineering prominently among them — show a roughly 16% relative employment decline since late 2022, while older developers are stable or growing (Brynjolfsson, Chandar & Chen, 2025). Microsoft's own 2025 layoffs concentrated in software engineering, and the CEO has stated publicly that around 30% of company code is now written by AI (The Register, 2025).

For a working .NET developer, the practical answer is stay, but broaden. C# is having a quiet renaissance and .NET 10 (Microsoft, 2025) is genuinely competitive on performance and AI tooling. The agent-coding gap and the entry-level squeeze make a single-language bet riskier than five years ago. Python is the highest-leverage second language; TypeScript is the most marketable web-side complement.

Background — what counts as "popular"?

There is no single number for how popular a programming language is. There are six widely-cited indices, each measuring something slightly different:

• Stack Overflow Developer Survey asks tens of thousands of developers what they "used in the past year." Self-report; large sample; biased toward English-speaking, Stack Overflow-using developers.
• GitHub Octoverse counts unique monthly contributors per language across public and private GitHub. Activity, not preference; biased toward open-source-shaped work and toward whatever language people use on GitHub.
• IEEE Spectrum weights a composite of seven sources (Google search, Stack Overflow, IEEE Xplore, IEEE Jobs, CareerBuilder, GitHub, Trinity College Dublin Library, Discord). Three views: "Spectrum" (general), "Jobs," "Trending."
• RedMonk plots GitHub pull-request activity against Stack Overflow tag discussions. Two-axis chart; slow-moving.
• TIOBE scrapes search-engine query share across roughly 25 sites. Old, well-known, methodologically criticised — over-weights legacy enterprise codebases and language-tutorial traffic.
• JetBrains State of Developer Ecosystem surveys ~24,000 of its own users. Strong on professional-developer detail; smaller sample than Stack Overflow.

A finding only becomes solid when several indices say the same thing through different methodologies. This report leans hardest on Octoverse (activity), Stack Overflow (preference among working developers), and the Microsoft Research C# benchmark for the agent-coding section.

One methodology caveat is becoming unavoidable. Stack Overflow's 2025 question volume was roughly 22% of its 2024 level — coding assistants now absorb most of the casual Q&A traffic (RedMonk, 2026; IEEE Spectrum, 2025). Indices that lean on Stack Overflow as a major input — RedMonk, IEEE Spectrum, TIOBE — are degrading as a measurement source faster than they are being replaced. It is the right time to weight activity-based signals (Octoverse) more heavily than search- or Q&A-based ones.

Findings

TypeScript is the standout grower; Python and C# are also up; the rest are not

The single sharpest 2025 finding is GitHub's: TypeScript took #1 on Octoverse 2025 with 2.64 million unique monthly contributors, ahead of Python (2.59M) and JavaScript (2.15M). TypeScript's contributor count grew 66.6% year-over-year — three times faster than JavaScript's (GitHub, 2025).

Stack Overflow's 2025 survey corroborates the trend at lower magnitude: TypeScript usage among all developers rose from 38.5% to 43.6% (+5.1 percentage points), and Python from 51% to 57.9% (+6.9 pp) (Stack Overflow, 2025). JavaScript in the same survey was statistically flat at 66%. Python now sits within eight percentage points of JavaScript in self-reported use, the closest the gap has ever been on Stack Overflow.

C# is the quiet outperformer. It moved from #5 to #4 in RedMonk's January 2026 ranking — the only top-10 movement that cycle (RedMonk, 2026). It gained 1.19 percentage points in TIOBE's index over the year, enough for TIOBE to name it Programming Language of the Year 2025 — its second time in three years (TIOBE, 2026). On Octoverse 2025 its monthly contributor count grew about 22% year-over-year. None of these are TypeScript-shaped numbers, but C# is the only language with a Java-or-older lineage that is moving up across multiple methodologies.

The decline list is short and stable. JetBrains's 2025 ecosystem report names PHP, Ruby, and Objective-C as in long-term decline; RedMonk's January 2026 ranking has Objective-C "continuing decline toward exit" from its top 20 (JetBrains, 2025). JavaScript's decline is more interesting and easier to misread: it lost share-of-new-activity on GitHub and slid from #3 to #6 in IEEE Spectrum's 2025 view, but it remains the most-used language on Stack Overflow at 66% and #1 in RedMonk's January 2026 ranking. The right reading is "JavaScript still wins the present, but is no longer winning the future."

.NET wins the objective performance comparison more often than its reputation admits

TechEmpower's Framework Benchmarks ran for thirteen years and twenty-three rounds. Round 23, published in March 2025, was the last; the project archived the following March (TechEmpower, 2025).

Round 23's "Fortunes" test — routing, ORM, HTML templating, and a database round-trip — is the test in the suite that most resembles a small real web application. The top frameworks per language separate by an order of magnitude or more.

Figure 1. Requests-per-second on the Fortunes test (HTTP routing + ORM + HTML template + database round-trip), top framework per language. Source: TechEmpower Framework Benchmarks Round 23, March 2025. Microsoft began sponsoring TechEmpower's hardware before this round — see methodology note below.

ASP.NET Core's 610,000 requests per second is roughly twice the next-best framework on this test (Go's Fiber at 338,000), nearly eight times Node's Express (78,000), and almost nineteen times Django (33,000). On the lighter "Plaintext" test the gap is wider still: ASP.NET Core peaks around 27 million requests per second, more than three times Fastify and over twenty times Express (TechEmpower, 2025).

Two methodology caveats. First, Microsoft began sponsoring TechEmpower's benchmarking infrastructure before Round 23 — that does not invalidate the numbers but means a #1 placement is not the same evidence it was in Round 21. Second, TechEmpower measures frameworks, not languages; ASP.NET Core is unusually well-optimised for the Fortunes-style workload, and a Python framework like FastAPI tested against a database-light workload narrows the gap considerably.

The corollary on the cold-start side is interesting for cloud workloads. .NET 10's NativeAOT in ASP.NET minimal APIs cuts memory from roughly 60–80 MB to 15–20 MB, and a published AWS Lambda case study reports cold-start latency dropping from 6.7 seconds to 940 milliseconds — about a sevenfold improvement (whitewaw / dev.to, 2025). Stephen Toub's annual performance post for .NET 10 catalogues hundreds of similar wins (Toub, 2025). Python and Node have also shipped real gains — Node 24's V8 13.6 raised JavaScript execution by ~17%, and Python 3.14's free-threaded mode delivers a 3.1× speedup on multi-thread CPU-bound work (Node.js, 2025; Grinberg, 2025) — but neither closes the order-of-magnitude gap on standard web workloads.

LLM coding agents code Python much better than they code anything else — and C# worst of all the major languages

Every frontier model since GPT-4 has reported HumanEval pass-rates above 90%. The benchmark is saturated and no longer separates models. The current frontier benchmark is SWE-bench Verified — 500 hand-checked GitHub issues that an agent has to resolve end-to-end. The catch: SWE-bench Verified is Python-only.

Cross-language data is thinner. The original SWE-bench Multilingual paper measured Claude 3.7 Sonnet across eight languages: Rust 58%, Java 53%, PHP 49%, Ruby 43%, JavaScript/TypeScript 35%, Go 31%, C/C++ 29% (Yang et al., 2024). The same model on Python (Verified) scored 63%. Anthropic's Claude Opus 4.5 in November 2025 reported leading seven of those eight languages on the multilingual track and 80.9% on Python Verified (Anthropic, 2025). OpenAI's GPT-5 system card and Google's Gemini 2.5 / 3.1 model cards lead with SWE-bench Verified and do not prominently report multilingual numbers.

C# is absent from all of these. The first agentic benchmark for C# / .NET appeared in November 2025: Microsoft Research's SWE-Sharp-Bench, 150 issues drawn from 17 real C# repositories (Yang et al., 2025).

Figure 2. Frontier-model agentic coding pass rate on Python (SWE-bench Verified) and C# (SWE-Sharp-Bench), under matched harnesses. Sources: SWE-Sharp-Bench (Microsoft Research, arXiv 2511.02352, November 2025); SWE-bench Verified leaderboard.

The gap is roughly 30 percentage points, and it has been roughly 30 percentage points across every model generation from GPT-4o through GPT-5 and from Claude 3.5 Sonnet through Claude Sonnet 4. The benchmark's authors trace the cause not to training-data scarcity — The Stack v2 (the canonical open-source training corpus) contains 166 GB of C# against 191 GB of Python, an 87% ratio (BigCode, 2024) — but to patch complexity: real C# fixes more often involve multi-file changes, deeper inheritance hierarchies, larger dependency injection graphs, and projects that the model cannot reason about with the same shallow heuristics that work on a typical Python repo.

The compounding factor is the optimisation pressure that follows the published benchmark. Frontier labs all advertise SWE-bench Verified scores; their post-training data and reinforcement-learning environments are tilted toward Python agentic traces. Python improves; C# improves less. There is no obvious technical reason this asymmetry must persist, but as of mid-2026 it does.

The labour market for developers is bifurcating: junior demand is collapsing, senior demand is intact, AI roles are growing fast

The cleanest published evidence on the AI-coding shift in employment came in November 2025. The Stanford Digital Economy Lab analysis of ADP payroll records found that workers aged 22 to 25 in the most AI-exposed occupations — software developer prominently among them — saw a 16% relative employment decline versus the same age cohort in less-exposed occupations, since late 2022. Workers aged 35–49 in the same exposed occupations were stable or growing (Brynjolfsson, Chandar & Chen, 2025).

Microsoft's own 2025 layoffs are an unusually transparent corroboration. Of the company's roughly 15,000 cuts, more than 40% landed in software engineering. At the Redmond headquarters, software engineers were 817 of about 2,000 cut roles — more than the next four categories combined (The Register, 2025; GeekWire, 2025). The CEO has said publicly that around 30% of Microsoft's code is now AI-written. Indeed's January 2026 hiring report shows AI-mentioning postings up 130% above the pre-pandemic baseline while overall tech postings are 34% below it (Indeed Hiring Lab, 2026).

Inside that contraction, the per-language picture is more boring than headlines suggest. DevJobsScanner's most recent global posting analysis (about 12 million postings through September 2024) puts JavaScript / TypeScript at roughly 31% of openings, Python at 20%, Java at 12%, and C# at 12% — and notes that C#'s share has been flat across 2023–24 (DevJobsScanner, 2024). On ITJobsWatch's UK permanent-postings data for the six months ending 10 May 2026, the entire IT skills market lost rank as the total number of postings shrank, but the relative picture favours C#: its 4% salary decline year-over-year is smaller than Python's 8% or JavaScript's 9% drop, and its share of UK permanent postings (2.62%) is comparable to Java's (2.64%) and ahead of TypeScript's (1.71%) (ITJobsWatch, 2026).

Median salaries diverge by ecosystem. In the UK, Go and Rust roles still pay roughly £90,000, TypeScript £75,000, Python £69,000, .NET £65,000, C# £60,000 — the "modern systems" languages carry a premium over the established enterprise stack, but every absolute number is slightly lower than a year ago. US per-language data is weaker; Levels.fyi reports an overall software-engineer median of $191,000 without language splits, and most "C# pays $X" claims trace to recruiter-blog estimates that should be treated as Tier-3 evidence (Itransition, 2026).

.NET 10 is a strong year for the platform, on its own merits and in context

Setting aside the labour market, the .NET platform is in unusually good shape. .NET 10 shipped on 11 November 2025 as a Long-Term Support release with support through November 2028. .NET 8 and .NET 9 both end support on 10 November 2026, forcing a unified migration deadline and meaning that anyone running anything .NET-flavoured will be on .NET 10 within a year (Microsoft, 2025).

The headline release features point at where Microsoft is investing. Aspire 13 — the cloud-native orchestration story — adds first-class support for polyglot Python and JavaScript components in a .NET-anchored solution, an explicit acknowledgement that real-world systems are mixed-language. The Microsoft Agent Framework, the Microsoft.Extensions.AI libraries, and the C# Model Context Protocol SDK push the platform's AI integration story past the OpenAI Chat Completions wrapper layer. NativeAOT continues to mature; the Blazor bundle dropped 76% in size; AVX10.2 / SVE vector intrinsics landed; garbage-collector pause times are 8–20% lower (Toub, 2025).

C# 14 itself is a smaller release than C# 12 was, but the language has finally reached a steady state where annual additions are quality-of-life rather than reshape-the-paradigm: field-backed and extension properties, the ?.= null-conditional assignment operator, modest pattern-matching extensions.

The ecosystem signal is consistent. NuGet is past 800 billion package downloads; Microsoft cites 7 million developers using the Visual Studio family monthly (a number that traces to a Microsoft press claim rather than an audit; treat it as an order-of-magnitude figure). Bing and the Xbox Gaming Copilot are running on .NET 10 in production; Unity's migration from Mono to .NET CoreCLR is proceeding on its own multi-year schedule.

Analysis — what these findings mean together

Three tensions are worth holding at once.

The "languages people use" picture and the "languages they get paid for" picture are the same story at different latencies. Python's growth in the surveys is almost certainly real; C#'s movement up the rankings is almost certainly real; TypeScript's leap on GitHub is almost certainly real. None of those is yet a paid-job signal at the magnitude of the activity signal. The job market is in absolute contraction (in the UK, every language lost rank because total postings shrank), so language-level moves have to be read as relative — and relatively, C# is doing better than Java, JavaScript, or PHP, while Python is doing best of all. A two-year lag between "what working developers are using" and "what employers are listing" is normal.

The performance picture and the agent-coding picture point opposite directions for .NET. A modern ASP.NET Core service runs an order of magnitude faster than a Django one and ships in container images several times smaller. By the metrics that matter to a senior architect choosing a stack — throughput, tail latency, cloud cost — .NET is among the three or four best choices in the industry. By the metric that matters to a junior developer trying to ride a coding agent through an unfamiliar codebase — pass rate per attempt — .NET is roughly thirty percentage points behind Python and stuck. A team that is building a new system has a different decision than a team that is teaching new developers how to ship; the "right" language depends on which dimension of leverage matters more.

The AI-driven labour-market shift is real, narrow, and will widen. The Stanford payroll evidence is the cleanest published causal-leaning signal on AI displacement we have. It is concentrated on workers aged 22 to 25 in AI-exposed occupations, and software engineering is in that exposed set. Older developers are stable. There is no public dataset breaking the junior-vs-senior gap by language, and there should not be much of one — coding agents are language-agnostic in their effect on hiring. What this means in practice is that the bottom rung of the developer career ladder is being sawed off, regardless of stack. The defensible response is to skip rungs upward (more architecture, more cross-system reasoning, more business-context judgement) rather than to switch stacks.

What would change this picture? Three things.

If Anthropic, OpenAI, or Google start publishing C# benchmark numbers in their flagship system cards, that would be both evidence of progress and a forcing function for further progress (because optimisation pressure follows what is published). If an entry-level hiring rebound shows up for two consecutive quarters in Indeed and ADP data, that would invalidate the headline labour-market read. If Microsoft cuts an order of magnitude more software-engineering jobs in 2026 than it did in 2025, the displacement trajectory is steeper than current evidence suggests. None of those is the base case; all three are worth watching.

Conclusion

A developer choosing a language in 2026 should look first at what the language is for, second at what it pays, and only third at what is moving on a rankings page. Python has the broadest applicability and the best AI-tooling story. TypeScript is the safest bet on the web side. Rust and Go are real for systems-level work but command small absolute volumes of hiring. Java is a perfectly good legacy choice with no strong tailwind. PHP, Ruby, and Objective-C are in long-term decline.

C# is the most underappreciated choice on this list. It runs faster than its competitors on standard server workloads, it has a strong language-design moment behind it, it is the only legacy-lineage language that is moving up across multiple indices, and it is paid roughly equivalently to Java in most markets. Its weak spots are the agent-coding gap, which is real but might close, and the relative thinness of the early-career hiring market, which is structural and probably will not.

For a working .NET developer, the practical advice is unromantic: keep the .NET fluency that already pays the bills, treat .NET 10's LTS migration as the small-but-non-trivial event it is, and add Python as a second language for the AI-tooling work that will increasingly cross the desk. Picking up some TypeScript helps on the front end. Anything beyond that is taste.

The longer arc is harder to call. If coding agents continue to advance at their 2024–25 pace, the kind of work a working developer does will change — toward more design, more verification, less typing — faster than the language that work is done in. In that future, the language-ranking discussion in this report will look quaint. We are not in that future yet, and the language one chooses for the next two years still matters.

Sources

Primary — surveys, benchmarks, official disclosures

1. Anthropic. Claude Opus 4.5 System Card. November 2025. Accessed 2026-05-10. https://assets.anthropic.com/m/64823ba7485345a7/Claude-Opus-4-5-System-Card.pdf
2. BigCode (ServiceNow / HuggingFace). StarCoder 2 and the Stack v2. arXiv:2402.19173. February 2024. Accessed 2026-05-10. https://arxiv.org/abs/2402.19173
3. Brynjolfsson, Erik; Chandar, Bharat; Chen, Ruyu. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of AI. Stanford Digital Economy Lab. 13 November 2025. Accessed 2026-05-10. https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/
4. DevJobsScanner. Top 8 Most Demanded Programming Languages. 2024 edition (data through September 2024). Accessed 2026-05-10. https://www.devjobsscanner.com/blog/top-8-most-demanded-programming-languages/
5. GitHub. Octoverse 2025: a new developer joins GitHub every second as AI leads TypeScript to #1. 28 October 2025. Accessed 2026-05-10. https://github.blog/news-insights/octoverse/octoverse-a-new-developer-joins-github-every-second-as-ai-leads-typescript-to-1/
6. GitHub. Octoverse 2024. 29 October 2024. Accessed 2026-05-10. https://github.blog/news-insights/octoverse/octoverse-2024/
7. IEEE Spectrum. Top Programming Languages 2025. 23 September 2025. Accessed 2026-05-10. https://spectrum.ieee.org/top-programming-languages-2025
8. Indeed Hiring Lab. January Labor Market Update: Jobs Mentioning AI Are Growing Amid Broader Hiring Weakness. 22 January 2026. Accessed 2026-05-10. https://www.hiringlab.org/2026/01/22/january-labor-market-update-jobs-mentioning-ai-are-growing-amid-broader-hiring-weakness/
9. ITJobsWatch. C# / Python / JavaScript / TypeScript / Java / Go / Rust / .NET skill pages. Six-month rolling window to 10 May 2026. Accessed 2026-05-10. https://www.itjobswatch.co.uk/jobs/uk/csharp.do
10. JetBrains. State of Developer Ecosystem 2025. October 2025. Accessed 2026-05-10. https://devecosystem-2025.jetbrains.com/
11. Microsoft / .NET team. Announcing .NET 10. 11 November 2025. Accessed 2026-05-10. https://devblogs.microsoft.com/dotnet/announcing-dotnet-10/
12. RedMonk (Stephen O'Grady). Programming Language Rankings, January 2026. 14 April 2026. Accessed 2026-05-10. https://redmonk.com/sogrady/2026/04/14/language-rankings-1-26/
13. Stack Overflow. 2025 Developer Survey — Technology. July 2025. Accessed 2026-05-10. https://survey.stackoverflow.co/2025/technology
14. SWE-bench team. Multilingual leaderboard and methodology. Accessed 2026-05-10. https://www.swebench.com/multilingual-leaderboard.html
15. TechEmpower. Framework Benchmarks Round 23. 17 March 2025. Accessed 2026-05-10. https://www.techempower.com/blog/2025/03/17/framework-benchmarks-round-23/
16. TIOBE. TIOBE Index. May 2026 snapshot. Accessed 2026-05-10. https://www.tiobe.com/tiobe-index/
17. Toub, Stephen. Performance Improvements in .NET 10. Microsoft .NET Blog. 10 September 2025. Accessed 2026-05-10. https://devblogs.microsoft.com/dotnet/performance-improvements-in-net-10/
18. Yang, Jiacheng et al. SWE-Sharp-Bench: A Benchmark for Software Engineering Agents on Real C# Repositories. Microsoft Research. arXiv:2511.02352. November 2025. Accessed 2026-05-10. https://arxiv.org/html/2511.02352v3

Secondary — corroborating reporting and analysis

19. ByteDance Seed et al. Multi-SWE-bench. arXiv:2504.02605. NeurIPS 2025. Accessed 2026-05-10. https://arxiv.org/abs/2504.02605
20. GeekWire. Microsoft cuts another 4% of its workforce — about 9,000 jobs — in continued push for efficiency. July 2025. Accessed 2026-05-10. https://www.geekwire.com/2025/microsoft-cuts-another-4-of-its-workforce-about-9000-jobs-in-continued-push-for-efficiency/
21. Grinberg, Miguel. Python 3.14 Is Here. How Fast Is It?. 8 October 2025. Accessed 2026-05-10. https://blog.miguelgrinberg.com/post/python-3-14-is-here-how-fast-is-it
22. Itransition. Most In-Demand Programming Languages for 2026. January 2026. Accessed 2026-05-10. https://www.itransition.com/developers/in-demand-programming-languages
23. The Register. Microsoft's axe falls heaviest on software developers. 16 May 2025. Accessed 2026-05-10. https://www.theregister.com/2025/05/16/microsofts_axe_software_developers/
24. White, Wow. .NET NativeAOT on AWS Lambda: 7× faster cold starts, 73% lower costs. dev.to. 2025. Accessed 2026-05-10. https://dev.to/whitewaw/net-nativeaot-on-aws-lambda-7x-faster-cold-starts-73-lower-costs-np3

---

Methodology note. This report leans on activity-based indices (GitHub Octoverse) and primary survey data (Stack Overflow Developer Survey, JetBrains State of Developer Ecosystem) over search-engine-based ones (TIOBE) where they conflict, on the grounds that LLM-mediated coding has degraded the traffic signal at search engines and Q&A sites since 2024. Performance figures cite TechEmpower Round 23 — the final published round — and the Microsoft .NET performance blog, which is first-party but reproducible against BenchmarkDotNet. Agent-coding figures favour benchmark papers over vendor announcements where the two diverge. Salary and posting-share numbers are UK perm or DevJobsScanner global; US per-language compensation data of comparable rigour was not found.