Cheaper rule writing can make checks worthwhile
Theo argues that some useful custom checks used to cost too much to justify writing them. His example is a rule requiring around 400 lines to check project-specific oddities left to manual review. He says cheaper rule-writing and cheaper tests change that calculation, especially when agents encounter those issues more often.
Reassess the cost of rejected check ideas
Revisit checks you previously dismissed because writing the rule and its tests would cost too much. Assess that combined effort again, including the tests that verify the rule's behavior on your project. When that work is now affordable, have the agent write the rule and accompanying tests for review.
Revisit manual checks once automation becomes affordable
This fits a backlog of project-specific issues that reviewers repeatedly catch by eye during code review. The opportunity is strongest when those checks stayed manual because their implementation effort was hard to justify. If the rule and its verifying tests remain too costly, this economic argument does not establish that you should build it.
Custom checks still need maintenance and review
Theo presents an economic argument, without measuring the savings from a custom rule and its tests. Custom rules still require maintenance, and an incorrect rule can silently block otherwise good code. Reconsider a previously unaffordable check when its full effort makes sense, while reviewing the rule like any other code.