LanguagesPreview
Agent-marked pull requests by repository language: who leads where, and how fast each stack is adopting.
Coverage: tracks Claude Code, Codex, Copilot agent, Cursor, Devin, and Jules. Codex is measured through hosted task metadata, checked against sampled PRs. Gemini CLI and other terminal agents work through the developer's own identity, leaving nothing public we can detect yet, so they cannot be measured here. We are expanding coverage. Methodology
42.6%
Swift is most agentified
Share of all Swift PRs carrying an agent mark
16.3%
Vue is the frontier
93,685 monthly PRs, mostly unmarked
30.1%
Of all PRs are agent-marked
Across the 32 tracked languages this window
Movers
2026-10 vs 2026-08Month-over-month change in agent-marked PRs. Languages need 500 or more PRs in the prior month to qualify.
Language Heatmap
Toggle cells between each agent's share of the language's agent-marked PRs and its share of all PRs in the language (market dominance). The last column is how much of the language's total PR flow is agent-marked at all. Click any header to sort; language names link to detail pages. Copilot, Jules, and Devin counts are exact; others are estimates. Window: 2026-10-01 to 2026-10-31.
| Swift | 79% | 16% | 0.2% | 3.7% | 0.6% | 0.4% | 42.6% |
| R | 85% | 8.3% | 1.1% | 5.1% | 0% | 0.4% | 38.6% |
| TypeScript | 79% | 13% | 0.4% | 6% | 0.5% | 1.1% | 34% |
| Rust | 81% | 13% | 0.2% | 3.3% | 0.4% | 1.9% | 31.9% |
| Objective-C | 82% | 12% | 1% | 5.3% | 0.4% | 0% | 31.7% |
| Julia | 90% | 8.2% | 0.5% | 1.7% | 0% | 0% | 31.3% |
| JavaScript | 73% | 18% | 0.8% | 6.4% | 0.4% | 1.5% | 30.3% |
| Lua | 85% | 12% | 0.3% | 2% | 0.1% | 0.8% | 30.3% |
| HTML | 76% | 14% | 0.9% | 8.5% | 0.4% | 1.1% | 30.2% |
| Python | 76% | 17% | 0.5% | 4.6% | 0.7% | 1.3% | 29.7% |
| Dart | 77% | 15% | 0.4% | 5% | 0.4% | 2.4% | 29.6% |
| C++ | 79% | 16% | 1.4% | 2.4% | 0.3% | 0.8% | 28.7% |
| PowerShell | 71% | 21% | 2.3% | 2.1% | 0.1% | 3.6% | 28.6% |
| Elixir | 90% | 6.9% | 0.1% | 2.5% | 0.1% | 0.5% | 27.8% |
| Solidity | 79% | 8.1% | 0.3% | 7.4% | 4.7% | 0.3% | 27.8% |
| C | 77% | 16% | 0.8% | 3.1% | 0.5% | 2.3% | 27.4% |
| C# | 80% | 14% | 1.3% | 3.9% | 0.6% | 0.6% | 27.1% |
| Kotlin | 80% | 13% | 0.8% | 3.8% | 0.3% | 2.2% | 26.2% |
| Shell | 90% | 6.1% | 0.8% | 1.9% | 0.2% | 0.8% | 25% |
| Zig | 86% | 8.8% | 0.1% | 3.4% | 0.7% | 1.1% | 24.5% |
| Go | 82% | 13% | 0.5% | 2.5% | 0.9% | 0.8% | 24.2% |
| PHP | 80% | 12% | 0.9% | 5.8% | 0.2% | 0.9% | 24% |
| Haskell | 89% | 9.9% | 0.7% | 1% | 0% | 0% | 22.7% |
| Clojure | 61% | 35% | 1.7% | 1.9% | 0% | 0% | 22% |
| Groovy | 79% | 15% | 0.8% | 4.8% | 0% | 0.1% | 21.1% |
| Svelte | 79% | 9.5% | 1.1% | 3.2% | 5.4% | 1.3% | 19.5% |
| Java | 78% | 16% | 1.3% | 2.8% | 1.6% | 0.9% | 16.8% |
| Ruby | 82% | 9.7% | 0.6% | 6% | 0.5% | 1.3% | 16.7% |
| Scala | 87% | 7.1% | 1.2% | 4.2% | 0.4% | 0.2% | 16.5% |
| Perl | 88% | 7.1% | 2.2% | 0.9% | 0.2% | 1.4% | 16.4% |
| Vue | 79% | 12% | 1% | 6.6% | 0.4% | 1.4% | 16.3% |
| Erlang | 85% | 13% | 0.8% | 1% | 0.2% | 0.2% | 15.1% |
Showing share of agent-marked PRs, sorted by agent-marked share. Click a column header to re-sort. Hover a cell for both percentages and the raw count.
Who Is Adopting Coding Agents Most
Agent-marked share of all PRs across the 32 tracked languages. Window: 2026-10-01 to 2026-10-31.
What Each Agent Works On
Each agent's five biggest languages as a share of its own marked PRs. Window: 2026-10-01 to 2026-10-31.
Adoption vs. Market Size
Market size against agent adoption. Top right is the heartland (big and agentified). Bottom right is the frontier: big markets where agent-marked share is still low, which is where adoption runs next. Bubble size is agent-marked PRs. Window: 2026-10-01 to 2026-10-31.
Adoption Over Time
By month per language: agent-marked share of all PRs, or raw agent-marked PR volume. The slope is the story: which stacks are agentifying fastest, and where the volume is growing. Window: 2025-02 to 2026-10.
Method And Caveats
- language: filters by a repo's primary language, so a TypeScript monorepo's Python PRs count as TypeScript. Shares within a language are robust to this; absolute per-language totals less so.
- Codex is measured through hosted task metadata. Its language history before that coverage was validated is thinner than the others, and Codex CLI work remains uncounted.
- Signal families and exactness rules live on the methodology page.