Airecmark Dossier · 14 min read
The Rise of AI Coding Agents: Autonomous Codebases vs IDE Copilots
Zero-human-edit merge rate in 2026 across CI suites, and what it means for the IDE-first toolchain.
Airecmark Research·2026-08-28
coding-agentsidebenchmarks
DOSSIER-2026-084. Autonomous coding agents crossed a threshold this cycle: multi-file refactoring patches now land without human loop-correction in a majority of CI suites we instrumented.
This dossier decomposes the composite Airecmark Score for the six tracked coding agents and maps where IDE-native copilots still hold structural advantages.
Live Appendix · Real Scores
The Tracked Field, Ranked
| # | Tool | Overall | Review |
|---|---|---|---|
| 1 | Cursor Anysphere Inc. |
89.2 | Inspect → |
| 2 | GitHub Copilot Microsoft / GitHub |
88.1 | Inspect → |
| 3 | Claude Code Anthropic PBC |
86.1 | Inspect → |
| 4 | Aider Paul Gauthier (OSS) |
86 | Inspect → |
| 5 | Windsurf Codeium |
83.9 | Inspect → |
| 6 | Replit Agent Replit, Inc. |
81.8 | Inspect → |
Computed live from the index — same five-dimension composition as every tool page.