Hi @Pawel Jozefiak , great question. The problem you mention is something that affect every aspect of interaction and let-it-do-the-work with AI Coding Agents.
The issue raised by the majority quite always is caused by “poor instructions and structure“ I mean, many thinking of AI Coding Agents can do things magically without any (human) interventions lands on very bad results, huge noise and time lost fixing and patching what AI did “wrong“.
To solve the issue and make/tune the AI to catch real issues you need to spend time describing/instructing/stating what means (good) quality/implementations/rules/practices that will help tremendously the AI to understand what is not and then catch real issues.
This is true regardless about reviews, tools, stages, whatever, this is true per any valuable usage of AI Coding Agents, and we use that level of detailing and describing in our projects so that Coding Agents acts and fixes strictly following the rules.
If this is a topic that interest you, the series of post we’re doing right now (Lead The Machine) will cover this issue too, or if you want we might create an issue only about this specific thing, let us know, we’ll be more than happy to help!
AI code review is one area where automation actually delivers value. Not perfect, but valuable.
I've been using AI coding agents (https://thoughts.jock.pl/p/claude-code-vs-codex-real-comparison-2026) and automated review catches things humans miss—patterns across files, security issues, style inconsistencies.
But it also flags false positives constantly. The signal-to-noise ratio needs work.
Your streamlining approach makes sense. The question is: how do you tune the AI to catch real issues without drowning teams in noise?
Have you found a good balance between thoroughness and usability?
Hi @Pawel Jozefiak , great question. The problem you mention is something that affect every aspect of interaction and let-it-do-the-work with AI Coding Agents.
The issue raised by the majority quite always is caused by “poor instructions and structure“ I mean, many thinking of AI Coding Agents can do things magically without any (human) interventions lands on very bad results, huge noise and time lost fixing and patching what AI did “wrong“.
To solve the issue and make/tune the AI to catch real issues you need to spend time describing/instructing/stating what means (good) quality/implementations/rules/practices that will help tremendously the AI to understand what is not and then catch real issues.
This is true regardless about reviews, tools, stages, whatever, this is true per any valuable usage of AI Coding Agents, and we use that level of detailing and describing in our projects so that Coding Agents acts and fixes strictly following the rules.
If this is a topic that interest you, the series of post we’re doing right now (Lead The Machine) will cover this issue too, or if you want we might create an issue only about this specific thing, let us know, we’ll be more than happy to help!