Are AI coding agents actually getting better? Six months of diagnostics data say yes, with caveats. Kiro’s agents use LSP-based diagnostics to catch type errors, unresolved imports, and undefined symbols. Newer models aren’t just making fewer errors, they’re checking broader sets of files and self-correcting better. Full breakdown 👉 spr.ly/6018B1x1dG

Aug 26, 2026 · 6:59 PM UTC

1
8
62
6,072
Sort replies: Relevant Recent Liked
Replying to @kirodotdev
Broader checking matters more than raw error count. An agent can reduce obvious mistakes while changing the wrong surface area. A stronger evaluation would combine LSP diagnostics with diff size, test coverage, reverted changes and whether a human had to restate the task.
42