Agents don't understand what they do

If I deployed that to production without a code review, my database would be in outage and my API would have downtime.

Agents write a few lines of code that run queries against the database in a loop on a hot path because they don’t “see” and “understand” what they do.

Of course, it can be prevented with performance tests. But there is always one more thing to check to ensure that agents produce safe and performant code, and you might not notice it. You can’t foresee everything.