Ask HN: Will human code review still exist a year from now?
24 comments
Review bots are the autopilot. Humans are the pilots.
The autopilot will keep getting better, but you still want a human in the cockpit when something truly unexpected happens.
I'm 100% confident LLMs could replace any code review being done by people who think this is a good idea, since both parties understand the assignment to the same degree.
Haha nice one, took me a second there to understand this.
no, i just started doing copilot PR reviews on github and its pretty damn good.
Source? I'm sure you won't answer that one because tech bros never have proofs about what they do. Also don't bother if it's JS or TS.
it just asked me if i wanted to enable copilot code reviews. I think I have a paid plan.
I meant source as in source code, which most vibe coders don't want to show, and which is strange since they are so proud of their new skills.
That’s dumb. Your bosses will decide to pay copilot instead of paying you. We should use llms in a way that empowers us, not in a way that replaces us.
this is true but that's the reality. AI is taking a lot of jobs. and it will continue to do so.
lol. human code review is all we will do day-to-day in a year
(i'm the lead engineer on diamond)
(i'm the lead engineer on diamond)
Surprised that’s your take considering Diamond is one of the code review bots that are ahead of the pack. Curious why you think that’s the case? Is it a technical limitation? Or something about human code review that is just fundamentally hard to copy?
AI will certainly make human code review easier — the goal being to keep up with the velocity of changes to existing large systems.
"Code review" as defined as a human in the loop getting the final say on whether a change will be made to the system will be the absolute last thing to go. That process may look very different as both the inputs massively increase in scale and the methods get disrupted by AI.
"Code review" as defined as a human in the loop getting the final say on whether a change will be made to the system will be the absolute last thing to go. That process may look very different as both the inputs massively increase in scale and the methods get disrupted by AI.
>"Code review" as defined as a human in the loop getting the final say on whether a change will be made to the system will be the absolute last thing to go. That process may look very different as both the inputs massively increase in scale and the methods get disrupted by AI.
I know this is probably not something you can divulge but I look forward to how your team at Graphite plans to solve this! (Would also love your personal take!)
I know this is probably not something you can divulge but I look forward to how your team at Graphite plans to solve this! (Would also love your personal take!)
We've already soft launched the next step toward this to some existing Graphite users, so I'm fine sharing!
We're adding an agentic chat sidebar to the pull request that can help you gather context to understand a PR and make pre-merge changes without needing to leave the PR page. Open beta next week.
We're only going to spend more and more time looking at diffs as the amount of them increases, so the tools that we use here need to evolve as well. "AI code review" should be an interactive experience that directly builds on how code is reviewed today, not just a bot that comments on your PRs.
We're adding an agentic chat sidebar to the pull request that can help you gather context to understand a PR and make pre-merge changes without needing to leave the PR page. Open beta next week.
We're only going to spend more and more time looking at diffs as the amount of them increases, so the tools that we use here need to evolve as well. "AI code review" should be an interactive experience that directly builds on how code is reviewed today, not just a bot that comments on your PRs.
unlikely- as long as your code is fairly micro-serviced about, llm generated code is fairly superior in all ways.
> llm generated code is fairly superior in all ways.
Depends on how good your code is :-).
Depends on how good your code is :-).
Yes
I'll still be reviewing my own code
I have tried getting LLMs to review code that I have written and most of the feedback I get is useless. It's as if they can only spot the most trivial of issues, or even worse -- they find issues in places where they don't exist.
I guess that they are moderately useful for finding copy-paste errors.
I guess that they are moderately useful for finding copy-paste errors.
If i don't review the code anymore, don't expect me to be responsible for it.
That’s true haha, even if obsolete for catching bugs, etc. one reason for human code reviews might just be accountability.
Personally, I don't even think you'll have a code review bot. You'll just do it in the IDE.
If models keep improving, and we eventually inject these code review bots with more and more context to validate even things like business logic, tribal knowledge, etc. do you think we could eventually arrive very soon in a place where we no longer do traditional code reviews?
Just work with your coding agent to develop things, get it reviewed by a code review bot, fix any issues caught, and ship.