MLOps maturity levels
Google Cloud’s MLOps architecture guide describes **three maturity levels** for how automated your ML delivery is—from fully manual (level 0) to automated training pipelines (level 1) to full CI/CD for those pipelines (level 2). Levels a...
What it is
Google Cloud’s MLOps architecture guide describes three maturity levels for how automated your ML delivery is—from fully manual (level 0) to automated training pipelines (level 1) to full CI/CD for those pipelines (level 2). Levels are a map, not a moral score.
<!-- IMAGE: Level 0 manual → Level 1 CT pipeline → Level 2 CI/CD + CT -->
Visual Spec & Architecture Diagram
Maturity levels 0→2 (or 0→3 if matching chapter): Level 0 manual scripts; Level 1 automated training pipeline; Level 2 CI/CD + automated retraining/monitoring. Icons escalate automation. Align labels to chapter's cited maturity model.
Why it matters
Teams often try to jump to “full MLOps platform” before they can retrain safely. Levels show what to automate first: first stop shipping notebooks by hand, then automate retraining, then automate pipeline *changes*.
How it works (plain)
| Level | Plain meaning | You mainly deploy… |
|---|---|---|
| 0 | Manual, script/notebook-driven | A trained model as a prediction service |
| 1 | Automated ML pipeline (continuous training) | A whole training pipeline that can emit new models |
| 2 | CI/CD for pipeline code + CT | New pipeline *implementations* automatically tested/deployed |
Climb levels when release frequency, data change rate, or failure cost demand it—not for fashion.
Everyday example
Level 0: home cook. Level 1: meal-prep Sundays on a checklist. Level 2: a test kitchen that can change the checklist itself safely every week.
Try it
Score your team 0/1/2 on: (a) how you deploy models, (b) whether retraining is automatic, (c) whether pipeline code changes go through CI. Pick the *single* next automation that would cut the most pain.
Myths
- ⚠️ Myth: Everyone must reach level 2.
- ✓ Reality: Level 1 may be enough if models change rarely but data changes often.
- ⚠️ Myth: Level 0 means “bad engineers.”
- ✓ Reality: Level 0 is common when starting; the risk is staying there while the world drifts.
Sources
- Google Cloud — MLOps levels: https://docs.cloud.google.com/architecture/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning ↗
- Microsoft Learn MLOps path (pipelines + GitHub Actions): https://learn.microsoft.com/en-us/training/paths/build-first-machine-operations-workflow/ ↗
- AWS MLOps planning: https://docs.aws.amazon.com/prescriptive-guidance/latest/ml-operations-planning/introduction.html ↗