LLM-Driven Predictive Test Selection That Reads Your PR Diff And Picks Which Test Suites Should Be Blocking After CloudBees Bought Launchable
An HN top-thread asker wants 'an LLM tool that can sit on a CI pipeline to propose what tests should be blocking' by reading the diff, not just retry-pass patterns... and a way to estimate how many times to repeat new tests to prove they aren't flaky to begin with. Launchable was the obvious answer here, but CloudBees bought it and rolled it into 'CloudBees Smart Tests' enterprise tier, leaving smaller teams without an OSS or affordable SaaS path to LLM-based change-aware test selection.
Build it as a CI step that emits a JSON test-plan, not a hosted SaaS dashboard. Buildkite, GitHub Actions, GitLab and CircleCI users all want the same primitive... they don't want yet another login. Hard problem inside the LLM is staying cheap on monorepo-size diffs. Use embedding similarity to test files first, only escalate to a reasoning model when similarity is ambiguous.
landscape (3 existing solutions)
The 'change-aware test selection' category is now dominated by one acquired enterprise product (CloudBees Smart Tests) and a handful of retry-only flake detectors. There is no LLM-native, vendor-neutral, OSS or affordable-SaaS option.