Unreviewed code is immoral.

We still do code review for PRs to Cursor. The idea is to first get a working demo, test things yourself, and then ask others to review. You said "I would love it more than anyone if I didn't have to review code anymore". We agree. Think we can get there eventually but not yet

Code review is a chore and of course the fantasy is removing all chores :) I struggle philosophically though to imagine totally unreviewed code in the wild. It’s not like poorly reviewed code is anything new. But for it to be the industry standard, sounds like a strange new world.

@leerob If nothing else, it saves so much time to not have to review code if you see the demo is not what you wanted and can tell it to go back and change it before you review

@leerob good point

as professionals we can't assume responsibility without cognitive ownership, and that requires looking at and beyond the code. i get the feeling all of these cool tools are being driven by the marketing team.

cursor is just trying to win at all costs

they all are :)

it's totally fine for unreviewed vibe codes, like openclaw, since nobody really uses it to do serious stuff, they only use for X posts to show off; but a complicated engineering application running on thousands of k8s containers, a tiny typo could ruin to let all services down, we've been there, it didn't happen on normal cases, it only happens on peak time

yup 💯

Im glad my first job was at a startup where all I had to do was bug fix by looking at a shit ton of code across different languages. Ive always found code reviews tolerable since then

that makes one of you :)

HARD Agree. We are about to see a decade of very weak and broken software.

yeah..until we figure out what the new rules are

These companies are selling to the general population. So you’ll see alot of these ‘features’ which actually harm software development and the responsibilities that come with it.

this reminds me of when the music industry introduced synthetic sounds like. people complained (i believe still do) that it wasn't real music with real instruments. ima just leave this here for interpretation

I really enjoyed this videos, use AI is ok use it blindly is the issue, believe that will do everything is basically dumb and make no sense

glad you enjoyed it!

For now. But, I don't generally check compiler output. Eventually we will only check whether the output meets expectations and not the internal plumbing. The fact we are having this convo tells you the path we're on. The "morality" of reviewing code is a short term issue.

Nah, you are wrong. Not building the right backpressure and harnessing is the immoral part. cc @GeoffreyHuntley

@GeoffreyHuntley elaborate?

I’m just confused as to why a developer can’t run the code and test locally? Why is that not considered in their video? And looks like Cursor is showing us who their target market is through this video. Vibe coders, not devs who have a career and reputation on the line.

You're right but you are using a wrong example here. Most of the users who are not looking at the codebase is because 1. either they don't know whats right/wrong. 2. they don't care. 3. they are not building something important. They are most likely not building a crypto wallet with high security. from what I get in this demo is, they are targeting PMs, designers, etc. who wanna build visual things, clearly thats why there's a video. Reviewing the code would be passed on to an engineer who can actually review the code. Its their job.

Mo, it's like I think something, I can necessarily put it into words, and you make a great video about it

😄🤝

cherry picked take tbh. If you are relying on every dev in every team somehow has both the maturity level(s) and team dynamics strong enough to answer those questions with certainty that you claim they would have, maybe. Even then I'd still raise an eyebrow and be skeptical. The reality is most teams around the world aren't all ex-rockstar gun engineers who write secure code and aren't capable of screwing their own code bases up. I'd probably bet more on Ai finding security vulnerabilities in the working majority via multiple wash cycles than anything one human says "this is secure" I'd also assume that if you vibe code your cursor way of life and go into production thereafter, you'd ofc put it through various wash cycles in the build pipeline to check for security vulnerabilities. A lot of ai still relies on grep instead of semantics so that can also fail too. Swap out the vibe coded "it just works" story line and swap in an incompetent dev you've arrived at the same starting line for this argument. Hence my "eh..." response. I can see both sides of this being equally arguable. This is just a free kick for you is all :)

Thank you for being the voice of reason in the madness

🙏🤝

When you print a paper, do you review it. So that's the perfection in code we are reaching...

Agreed. It’s actually a bigger issue than it seems.

The only way code can go unreviewed by humans is if it's "reviewed" by something else - tests or agents or both. I can imagine the correctness percentage of systems like that trending upwards into 9-s, until they become better than human teams. But AFAIK we're not there yet.

I'd also love not to review code I don't understand.

agree with your claim. Even they both sound unprofessional individuals saying “or whatever” “Tested to make sure it works” They seem don’t know exactly what’s is going on and what are they talking about! Just using a provided feature

amazing content. very relevant. shortcutting reviews will lead to larger problems in the future. hackers are going to have a field day (the use LLM/agents too...)

thank you! 🙏

great video. So for hardcore code reviewers, how would you make the « human code review monitoring » part more reliable, less friction, much better overall experience than we have today, apart from using tools like Greptile or Code Rabbit? Curious to know how you think of this in your own -or other colleagues’- hardcore environments

If Boing finds about this, we should switch to boats

See, this is where I get worried. Someone is going to release one of these unreviewed coded agents and it is going to break the internet or something.

I am having a similar problem while reviewing any LLMs work with creative writing. What I discovered is that any LLM is good when you know exactly what you want (review time becomes almost minimal). With software engineering, it’s much more complicated.

it is a great tool for visual features, and it must be supported by other tools and separation of concerns and so on, so you just trying to hype, bro

I’ve been shouting this sentiment forever. Why are there so many people out there PROUD of not reading the agents code. Crazy times.

You are 100% correct. I love AI and think it's a fantastic productivity tool because I can read and review code quickly. That took many many years of writing and reviewing code though. If this is their direction, then they will be build an incredibly dangerous tool.

Every ai engineer answer will be let’s have agents to do the review.

Another thing that got completely forgotten is testing, I never saw any vibe coders mention testing, while tests are the perfect solution to stop LLMs breaking your apps, and it’s insane since sometimes u just spend 50% of working time on tests

I'm telling you, man, one of these days someone's going to "vibecode" code for a safety critical system and there are going to be people hurt or, in the worst case, dead.

I very much believe that in the nearest future, llms would be trained well enough to be able to review code and check for security vulnerabilities (I very much believe they already do this in fact)

Shipping unreviewed code would be insane, i hope to god folks dont do that. I think vibe coding and tools like what cursor built are incredible for prototyping, and showing proof of concept and testing with users. Then you throw it to engineers to actually build it for real.

Regarding how to know if the code is safe, I think there’s a chance that people rely on things like this It will never be perfect, but it might improve so much that it becomes better than what humans can cover by reviewing the code.

Why use w3c and w3schools, while you can vibe and surf🏄‍♂️? Surfing the internet that is 😂. #whylearn

This is something we can resolve by simply having you review code with your keen eye and then have the other top coding AI platforms review the same and see who performs higher reviewing code -- you or the top AI platforms. Let's setup the test. Primates vs. AI. #highnoon