Skip to content

SYSTEM Cited by 1 source

Zalando risk-based PR-approval bot

Definition

A Zalando bot, triggered at PR-creation stage, that classifies each PR's rollout risk as low / medium / high and auto-approves the low-risk ones — the author can then self-merge. Built to protect lead time to merge as AI coding drove up PR volume. Canonical wiki instance of risk-based PR approval and risk-based auto-approval. (Source: sources/2026-08-13-zalando-agentic-engineering-at-zalando-a-snapshot)

How it works

  • Ruleset derived from Zalando's own production incidents and typical outage drivers — "highly specific to our tech stack, deployment manifests, configuration files, etc."
  • Config-breaking typos = high risk — "would have saved us from the metadpata incident."
  • Backwards-compatibility breaks = medium risk — requires a human to double-check business rationale.
  • Documentation-only changes = low risk.

Impact

  • 33% of PRs are low-risk and auto-approved.
  • Auto-approved PRs saw 20–40% PR lead-time reduction vs all PRs.
  • Especially accelerates prototypers / internal-tooling work that would otherwise need a colleague to rubber-stamp.
  • Behavioral second-order effect: engineers restructure work to raise the probability of a low-risk PR — splitting fast-shippable backwards-compatible changes from lower-priority medium-risk ones (e.g. dropping unused fields). Previously such changes were mixed together, increasing time-to-market and rollout risk.
Last updated · 766 distilled / 2,225 read