pomarie

Super cool!

Founder of cubic here, thanks for the shoutout!

Thanks Denis!

Good q! So it depends.

Some teams are using Claude or similar models in GitHub Actions, which automatically review PRs. The rules are basically natural language encoded in a YAML file that's committed in the codebase. Pretty lightweight to get started.

Other teams upgrade to dedicated tools like cubic. We have a feature where you can encode your rules either in our UI, or we're releasing a feature where you can write them directly in your codebase. We'll check them on every PR and leave comments when something violates a constraint.

The in-codebase approach is nice because the rules live next to the code they're protecting, so they evolve naturally as your system changes.

We're building something at cubic that helps with this. You write your constraints in plain English, and AI enforces them semantically on every PR.

If you're curious, you can check it out here: https://cubic.dev

Happy to answer any questions about what we've seen working (or not working) across different teams.

One thing that actually works is getting AI to review the basic stuff first so you can focus on architecture and design decisions. The irony of using AI to review AI-generated code isn't lost on me, but it does help.

That said, even with automated review, a 9000 line PR is still a hard reject. The real issue is that the submitter probably doesn't understand the code either. Ask them to walk you through it or break it down into smaller pieces. If they can't, that tells you everything.

The asymmetry is brutal though. Takes an hour to generate 9000 lines, takes days to review it properly. We need better tooling to handle this imbalance.

(Biased take: I'm building cubic.dev to help with this exact problem. Teams like n8n and Resend use it to catch issues automatically so reviewers can focus on what matters. But the human review is still essential.)

That's a really fair point. Architecture-first is definitely the ideal, and teams that can invest that time upfront tend to avoid a lot of downstream pain.

Circle of life!

Pretty cool – when does it make sense to use this vs n8n?

Great idea! How does your tab model compare to other ones from Cursor/Windsurf..?

That's so weird – I can't repro at all! I'll keep digging.. If anyone else reading this is also experiencing this, please shout!

Definitely! Giving the AI the ability to fetch the context it needs was a big challenge (since larger codebases can't all fit in an LLM's context window) – it's not perfect yet, but the tools it has does give it a remarkable amount of insight into the overall codebase

Super useful, thanks for the feedback! We're definitely thinking of building something that would run the reviews in your IDE directly, before you push the code.

Thanks for the encouragement!

It's possible! What was the tweet?

Thanks for sharing that Alex! Definitely love having an AI be the strict reviewer so that the human doesn't have to

Yeah, that's a totally legit point!

The good news with mrge is that it works just like any other AI code reviewer out there (CodeRabbit, Copilot for PRs, etc.). All AI-generated review comments sync directly back to GitHub, and interacting with the platform itself is entirely optional. In fact, several people in this thread mentioned they switched from Copilot or CodeRabbit because they found mrge's reviews more accurate.

If you prefer, you never need to leave GitHub at all.

Quick update – we've merged a fix which should be live in ~15 mins! Thanks for reporting this :)

Definitely! As AIs write a lot more code, I think that the PR/review space is going to become way more important.

If you're interested in Stack PRs, you should definitely check them out on Mrge. By the way, we natively support them (in beta atm): https://docs.mrge.io/ai-review/overview

Great question!

We've heard from users who've tried both that our AI reviewer tends to catch more meaningful issues with less noise, that's really something you should try for yourself and find out! (The great thing is that it's really easy to start using)

Beyond the AI agent itself (which is somewhat similar to Copilot), our biggest differentiation comes from the human review experience we've built. Our goal was to create a Linear-like review workflow designed to help human reviewers understand and merge code faster.

Really happy to hear mrge is useful! :) Thanks for sharing

Definitely—larger teams do typically have more stringent security and privacy requirements, especially if they're already using self-hosted GitHub. Self-hosted or hybrid deployment is definitely on our radar, and as we grow, it's likely we'll offer a self-hosted version specifically to support those larger teams.

If that's something your team might need, I'd love to chat more and keep you posted as we explore this!

Great questions!

For applying code changes with one-click: we keep suggestions deliberately conservative (usually obvious one-line fixes like typos) precisely to minimize risks of breaking things. Of course, you should confirm suggestions first.

Regarding AI reviewers being "too nice" and missing obvious mistakes—yes, that's a common issue and not easy to solve! We've approached it partly via prompt-tuning, and partly by equipping the AI with additional tools to better spot genuine mistakes without nitpicking unnecessarily. Lastly, we've added functionality allowing human reviewers to give immediate feedback directly to the AI—so it can continuously learn to pay attention to what's important to your team.

Great question!

We've heard from users who've tried both that our AI reviewer tends to catch more meaningful issues with less noise, that's really something you should try for yourself and find out! (The great thing is that it's really easy to start using)

Beyond the AI agent itself (which is somewhat similar to CodeRabbit), our biggest differentiation comes from the human review experience we've built. Our goal was to create a Linear-like review workflow designed to help human reviewers understand and merge code faster.

Thanks for sharing that Jof! Glad it's helpful :)

Hey, thanks again—really appreciate the heads-up! Could you point me to the specific section where you're seeing the fade in on scroll? Also, what browser are you using?

I don't remember adding that feature so it might be a bug

Really appreciate the feedback, really happy it's helping you :)

These are all amazing ideas. We actually already see a lot of solo devs using mrge precisely because they want something to catch bugs before code goes live—they simply don't have another pair of eyes.

And I absolutely love your idea of having multiple AI models review PRs simultaneously. Benchmarking LLMs can be notoriously tricky, so a "wisdom of the crowds" approach across a large user base could genuinely help identify which models perform best for specific codebases or even languages. We could even imagine certain models emerging as specialists for particular types of issues.

Really appreciate these suggestions!

That's a very good point. We actually just kicked off our own SOC 2 certification process last week—I hadn't updated the website yet, but I'll go ahead and do that now. Thanks for raising this!

Appreciate the feedback around security as well; protecting against supply-chain attacks is definitely top of mind for us as we build this out.

Ah, totally get it—that’s frustrating. GitLab support is on our roadmap, so hopefully we can help you out soon.

In the meantime, good luck with that hairy review—hope it goes smoothly! If you're open to it, I'd love to reach out directly once GitLab support is ready.