CMMI in the Age of AI-Assisted Delivery: What Appraisers Are Now Asking
Process maturity did not become irrelevant when engineers started pairing with models. It became harder to evidence — and more valuable to hold.
When a meaningful share of your code is model-generated, the appraisal question shifts from "is your process defined" to "does your process still hold when the work is produced faster than it can be reviewed".
The practice areas have not changed. What has changed is the evidence appraisers accept for peer review, configuration management and verification when generation is cheap and review is the bottleneck.
Three areas under new scrutiny
- Peer review. Reviewer attention per line of code has fallen. Show sampling strategy and defect-escape data, not just review completion rates.
- Configuration management. Prompts, model versions and tool configurations are now build inputs. If they are not under CM, your builds are not reproducible.
- Causal analysis. Defects with a generative origin cluster differently. Your CAR data should be able to see that.
Organisations pairing CMMI with ISO/IEC 42001 find the two reinforce each other: 42001 governs the AI you build and buy, CMMI governs the process that builds it. The evidence overlap is substantial and worth planning for rather than discovering during appraisal.