Spotify Honk part 3: verification loops before the PR exists
December 2025 post on deterministic verifiers + LLM-as-judge gates that block bad agent PRs from opening.
Source: Background Coding Agents: Predictable Results Through Strong Feedback Loops (Honk, Part 3)
The money post for review nerds: Spotify runs format/build/test verifiers and an LLM judge that can veto out-of-scope changes before a PR is even opened, with stop hooks on Claude Code.
Why it matters for code review
Best “AI code review” sometimes never leaves a comment—it refuses to open the PR. Inner-loop verification is complementary to outer-loop PR bots; serious agent factories need both.