For you doing Agentic Engineering with dbt, how do you make sure your agents aren't generating slop?
retained text
I find contra productive reading and manually testing all the code AI are generating, because I am turning myself into an important funnel and the company is paying for tokens and therefore wants fast and good delivery. I use Claude Code with dbt official skills, dbt mcp, lots of unity testing, some specific skills tailored for my projects, and the most important thing I think: deterministic checks. I run lots of python or bash script checks after specific hooks like pre commit or after calling a skill, or in the CI/CD, trying to force the implementation of our internal "rulebook". But I also don't think this is the optimum way to go, since it's also taking considerate amount of time. Also, do you let the AI query your db/dw? How do you guarantee fast and in a reliable way, your data diff is as expected? I am interested in listen from you about everything you've been doing nowadays trying to keep up this kinda insane market pace
document fetched 2026-10-03 00:24 · canonical https://reddit.com/r/dataengineering/comments/1wvso7m/for_you_doing_agentic_engineering_with_dbt_how_do