Tony Robalik documents three concrete incidents on his Dependency Analysis Gradle Plugin where LLM-using contributors caused real damage: an AI-generated ClassCastException misdiagnosis that got nearly every claim wrong, a "fix" PR that was semantically incorrect and twice the size needed, and a path-sensitivity PR that would have caused stochastic build failures for every remote-cache user.

His response: a Strict No LLM / No AI Policy added to the plugin's Code of Conduct.

A kotlin { ... } block nested inside android { buildTypes { release { ... } } } looks like it scopes -Werror to release builds, but kotlin is a top-level extension, so Gradle quietly applies it to every Kotlin compile task in the project.

Aurimas Liutikas walks through this footgun and how Gradle resolves names by walking up nested blocks, parent projects, and layered gradle.properties until it finds a match.

Benedikt Ritter traces a year-old "Gradle 8.12 broke rich console on NixOS" report to two unrelated upstream design choices that silently compose: native-platform's .so cache keys only on NativeVersion.VERSION (so patched and bare upstream libs collide), and nixpkgs-built JDKs can't dlopen libraries from /lib64 even on FHS hosts like Fedora-with-Nix.

Workaround in the meantime: pin org.gradle.console=rich in ~/.gradle/gradle.properties to bypass auto-detection entirely.

Paweł Oczadły makes the case for treating org-wide Terraform/OpenTofu migrations as a structured refactoring problem instead of a sed-script problem: OpenRewrite's lossless semantic trees do the rewriting, Gradle is the execution engine, and responsibilities split cleanly into Java recipes (the how), YAML recipes (the what), and Gradle (orchestration).

Two Azure Verified Modules case studies show the pattern landing as clean, review-ready PRs rather than ad-hoc scripts. Develocity Build Scans get a brief nod as the natural observability layer if you already use it.

Gradle 9.5.1 is the first patch release for 9.5.0, fixing three regressions that hit some users on the original 9.5.0:

• OOM when upgrading from 9.4.1 to 9.5.0 (#37753)
• Tooling API can't be used with a relative project directory in 9.5.0 (#37797)
• Missing docs for CreateMD5.java (#37812)

If you've moved to 9.5.0, the upgrade is recommended:

./gradlew :wrapper --gradle-version=9.5.1 --gradle-distribution-sha256-sum=bafc141b619ad6350fd975fc903156dd5c151998cc8b058e8c1044ab5f7b031f && ./gradlew :wrapper

Alex Semin lays out Gradle's plan for closing the configuration-phase bottleneck on large projects, and why Configuration Cache is set to become the default in Gradle 10. The bottleneck has shifted from execution to configuration on large projects, and Configuration Cache, plus the experimental Isolated Projects feature, directly attack it:

• Configuration Cache treats configuration as a pure function and caches the resulting work graph. Block reportedly saves ~4 years of engineering time annually after adoption.
• Isolated Projects adds boundaries between subprojects so configuration can run in parallel, and per-project caching can kick in on cache misses. Benchmarks show 1.3x–2.5x speedups in Android Studio sync for 50–100-subproject builds.
• NowInAndroid is now Isolated Projects–compatible, a useful proof point that the migration is feasible for real Android codebases.

The post also embeds the original KotlinConf 2025 recording if you'd rather watch.