Ground
The code
Let the pipeline check the boring parts
When half the pull requests are drafted by an agent, reviewers get tired fast. The code is usually tidy, the naming is consistent, and the bugs hide in the parts that look most confident. An agent never sounds unsure, even when it should. Human attention is now the scarcest resource in the pipeline, so stop spending it on formatting, coverage and import order. Those are deterministic questions, and they deserve deterministic answers that run on every push before anyone is asked to look at the diff.
- Formatting and linting enforced in CI, not in review comments
- PMD rules for Apex that fail on new violations only
- Coverage thresholds per class, not an org-wide average
- Every changed class has a changed or new test
- Secret and dependency scanning on every push
What is left for people is judgement. Does this change belong in this class at all? Does it match the spec, or just the ticket title? Will the next engineer understand why it works? Does it handle the unhappy path, or only the one in the ticket? Is the agent solving the problem it was given, or a slightly different one that was easier to solve? Those questions take ten focused minutes when everything else has already been checked, and an hour of tired skimming when it has not.
