It’s a neat idea, i think it’ll really take off if you change “before pushing to prod” to “before requesting another human’s time or attention”. If something like this was gatekeeping the “submit PR” and “update PR” buttons it would prevent a lot of waste.
Using AI to skip the humanity check is a lot more explicitly anti-social than using ai to write the code; if they are going to be axiomatically lazy I would at least have them do it indefensibly.
Sadly already happens. We have some rules for our devs that use AI and despite requiring a human to put a screenshot into the pr and hand write the pr they still have an ai do it all for them and will argue about how it’s too time consuming to write a pull request.
I’d love to try this project out but I really don’t think it’ll stop anything, and they’d probably just ask their agent to remove the hook anyways like they do with our husky hook.
That’s the problem, they’re still doing the work and delivering tasks but just not following the guidelines. That’s a discussion we haven’t really had internally yet since it’s all been moving so fast.
I certainly feel like certain AI coding practices I have leave me with nothing learned in a frustrating way. Sometimes the most effective way to getting what I want is just describing a goal really clearly, and then bashing the AI until it gets there. That bashing process can take a lot of energy and yet I don’t gain useful skills or a “better way” that I would if I was still coding. I generate more training data for BigAI corp, but that’s a slow return cycle.
Not sure if this is the right solution, but I would like to learn/understand/grow more as I pursue reckless efficiency.
Reason #1 to keep writing code by hand is that you don't ever lose that connection between you and the code.
There was an excellent post here on HN that unfortunately went unnoticed. I highly recommend reading it as it exactly points out this problem: understanding.
Learning is the bottleneck for shipping code. Building the model in your head, feeling out the edge cases and system interactions, and understanding the intricacies. Typing out the code is a small part of the process.
So, if you want to speed things up with LLMs, the bulk of the speed comes from reducing learning.
Imo if you’re reprompting you’re losing. If the agent oneshots the task its good but if it doesn’t manage it on the first try you will get there faster by doing it yourself.
Lol I built this and immediately got frustrated with it and stopped using it. I think I don't like being questioned by someone who I expect to work like magic.
I actually made this as a skill last week. It is helpful. Especially when reviewing a PR with 1500 LOC from a non-dev. It keeps me focused an helps me do a structured review. But it does not catch poor decisions, like calling the wring endpoint a such.
After noticing a drop in my understanding of the code I ship, I built and started using a similar skill ("Quiz me on the PR or the current branch"). I didn't tie it to a specific commit, though - to me it makes more sense to look at the PR as a whole.
So far I was positively surprised with the questions it could come up with.
I can kinda understand the motivation here: you push a bunch of slop (or is it deep, genius code?) but you have a kind of nagging suspicion that you should get to know it more familiarly, like back in the good old days a few months back when you sort of understood what was going on. I... don't... think this is going to work...? Remember Confirmation Fatigue? Soon we will have agents that interact with Sphinx for us.
I considered doing this with an open source app I have that let's you get a visual summary on a PR + get quizzed on it, but ultimately decided it's better to keep it an optional thing for people that want to understand AI PRs rather than force it as a gate, the volume is slowly becoming too high for even that. I agree with the sentiment behind this though as I felt this was needed as well.
Maybe the useful distinction is understanding what a PR does versus knowing how to tell if it’s wrong. A quiz can check the first, but the second is often where review matters most.
I’d be interested in a version that helps the reviewer identify one assumption the change depends on and a concrete way to test it. That could keep the exercise small while making it useful, especially if the reviewer spots something the AI-generated summary missed.
Keeping it optional also lets people reach for it when they’re uncertain, instead of turning it into another checkbox as PR volume grows.
All of this to just avoid knowing what you're doing.
Folks, this is the wrong direction. Use AI, but ffs take the time to properly learn what you're doing. We don't need to speed run civilizational collapse because we can't understand the concept of patience leading to better outcomes.
You can tell people are increasingly full of shit because they're doing this under the pretense that the problem is creating software faster, when really, a lot of people need to quit optimizing process and develop some skill and rigor. Most of the things that slow down software reaching production are human-created problems like "meetings," counting lines of code, or sending everyone to yet-another-conference that teaches them nothing (but boy does it look good on social media). It's all time waste and theater. And for what? So you feel included?
Pathetic. Learn to do this properly or get out of this industry.
How about to meet an aribrary deadline your boss set? Or how about a deadline needed that keeps the company afloat? Are you going to tell your boss: we need 6 more months because it's irresponsible to ship code we don't fully understand?
And yes, I have before and yes, I was right, and yes, they appreciated it in the long run.
Only idiots think "speed" is the sole determining factor in all of this. Every time I encounter someone who's over-focused on speed, I get them cut from my team. Those people are corrosive for the very reasons your hypothetical presents. It promotes a culture of unthinking that eventually leads to disasters, hands in pockets, and whistling. Loser behavior.
Get comfortable with becoming human Roundup. Weed the garden.
Mentalities like these are the reason "Software Engineering" is laughed at by actual physical Engineers. Do you want your discipline being taken seriously? Don't be a doormat, take responsibility for your actions!
It’s ironic that all negative comments on here are dismissive because “slop’s baaaaad”. Yes, slop is bad. Carelessness is bad. Lack of understanding is bad. It always has been. Code has been copied from stackoverflow. Built-in methods or std libraries have been used in a wrong way. Edge cases haven’t been covered. Rate-limiting and other hardening tasks haven’t been taken care of until the system broke. The code wasn’t DRY.
Now, people are building tools that not only hide the solutions to these problems in some obscure coding tool, markdown file, or agent harness, but they build gates that help developers to elevate themselves from the darkness of not knowing better. The number of developers roughly doubles every ~3 years, so at any given time, half of the people building software have less than 3 years of experience. Most of them didn’t even study CS before starting their careers. Tools like this will be the new warning voice that told people like us 5 to 10 years ago to never copy code from stackoverflow that you didn’t understand. It’s great.
AI writes unarchitectes code if you don't have a system for developing and maintaining a consistent architecture over time. In human coding, a lot of architecture decisions tend to only live in human brains, communicated through ephemeral means (text chats, voice calls, etc) and are not concretely documented. Since LLMs don't really have a way to manage long running architectural decisions (memory systems can only get so far, like it won't capture architectural decisions if you never talk about architectural decisions), you have to (have your agent) capture architecture design within the repository.
I personally like using openspec for this, because it makes the model iteratively ask you architectual decisions, and then manages creating and accessing the documents as needed.
> The problem with AI SE is that AI writes un-architecture-ed code
Not necessarily. I use SOTA models with established stacks like .NET which allow my agent to easily follow decades of best practiced patterns (DI, EF Core, ASP.NET Core...) so you still and up with proper architecture.
In fact I can easily still manually inspect everything bc I already know how it's supposed to be written.
From my experience, the latest models still struggle to create good architecture in the sense of good system design. You mention a couple of generally helpful patterns for the structure of the code. But they don't help much to create logical units, with intuitive boundaries, good cohesion etc. - at which even the latest models constantly make awkward decisions.
> But they don't help much to create logical units, with intuitive boundaries, good cohesion etc.
Completely, 100% agree on these exact points. I have access to every frontier model at work. None of them write what I would call clean code. At work the teams that have gone fully vibe-coded have repos full of broken abstractions that just accumulate over time. Eventually most of the effort of the team and their AI tokens seemingly goes to fixing issues that then break other things. Any attempt at 'refactor this mess, make no mistakes' turns into a _huge_ churn of changes ("shuffling the deck chairs") without ever really fixing things.
Honestly I love to see people working on this problem ! I actually built a similar project myself (https://github.com/bhurghundii/vibecheck.dev) except it prevents you from merging on the PR level. I’d love to get in touch to talk about it sometimes.
Edit: I see there are a lot of people questioning the value of this. The thing is that there are teams where people are just pumping tons of AI code without understanding their work (literally take the ticket and dump it) and shipping it. culture change is the right answer but v hard but this at least slows things down and is quick to implement
I also experienced pushback when I shared a similar skill with some of my colleagues (and it didn't mention "vibe" in the name). The main objection was "I am doing careful planning upfront, so I have the same level of understanding of the code as before".
Maybe they do. In my experience, the agent was going over the scenarios that I failed to consider during the planning stage, and it was net positive. I haven't tried it with a spec-driven approach.
I get what they say but what I see or have seen is totally different. I come from an engineering culture where we got AI token leaderboards and mandates to use AI for every single step of the development process. We started running into bad bugs recently due to people just copy pasting the Jira into the prompt so I feel there is space to put some intentionality in the areas which matter.
Not yet. But you’re like the 5th person to mention it so I think I’ll add that feature. There are some AI prevention stuff like you can’t copy paste which means you gotta use the visualisation.
Because they have low self-esteem and demand subjugation for their misdeeds and presumed inadequacy, all of which is perpetuated by a social apparatus that reinforces these beliefs as a matter of business as usual, or, "normal."
This is cute and novel, but practically useless. For those who use AI efficiently and effectively, this is useless. For those who don't, this becomes friction they don't want. Who really needs this?
And then you have review process done by AI as well. The trick is they all need to hallucinate the same thing and if they do, the initial requirements were wrong.
Sun Tzu said: If your people (AI agents) repeatedly fail to follow instructions, examine the clarity of your orders and the effectiveness of your leadership before blaming them.
People who care about the problem will definitely not be using this as a method to "understand" their code. People who don't care about this problem will continue to not care.
The quiz axis is interesting because it tests the person, not the diff. The complementary failure mode is the stuff you can pass a quiz on while still missing: a leftover debug print, an assertion weakened to match the bug it was meant to catch, an import of a package that never existed. Those live in the diff, not in comprehension. I hit this auditing my own patches last week - I could explain every line and still shipped a stdin read that broke on non-blocking pipes. Different catches, probably need both.
I’d love to try this project out but I really don’t think it’ll stop anything, and they’d probably just ask their agent to remove the hook anyways like they do with our husky hook.
> they still have an ai do it all for them
What recourse do you have when employees do not perform the work you're paying them for?
Not sure if this is the right solution, but I would like to learn/understand/grow more as I pursue reckless efficiency.
There was an excellent post here on HN that unfortunately went unnoticed. I highly recommend reading it as it exactly points out this problem: understanding.
Here is the link: https://news.ycombinator.com/item?id=50128468
So, if you want to speed things up with LLMs, the bulk of the speed comes from reducing learning.
But it very often feels like brute forcing the solution, and I'm not a big fan of it.
But maybe I'm using it wrong.
One would expect the actual answer would be to write the code ourselves.
Looks absolutely awful.
pre-code: planning mode
coding: humantodo.dev
after-code: sphinx hook
So far I was positively surprised with the questions it could come up with.
I’d be interested in a version that helps the reviewer identify one assumption the change depends on and a concrete way to test it. That could keep the exercise small while making it useful, especially if the reviewer spots something the AI-generated summary missed.
Keeping it optional also lets people reach for it when they’re uncertain, instead of turning it into another checkbox as PR volume grows.
Folks, this is the wrong direction. Use AI, but ffs take the time to properly learn what you're doing. We don't need to speed run civilizational collapse because we can't understand the concept of patience leading to better outcomes.
You can tell people are increasingly full of shit because they're doing this under the pretense that the problem is creating software faster, when really, a lot of people need to quit optimizing process and develop some skill and rigor. Most of the things that slow down software reaching production are human-created problems like "meetings," counting lines of code, or sending everyone to yet-another-conference that teaches them nothing (but boy does it look good on social media). It's all time waste and theater. And for what? So you feel included?
Pathetic. Learn to do this properly or get out of this industry.
How about to meet an aribrary deadline your boss set? Or how about a deadline needed that keeps the company afloat? Are you going to tell your boss: we need 6 more months because it's irresponsible to ship code we don't fully understand?
And yes, I have before and yes, I was right, and yes, they appreciated it in the long run.
Only idiots think "speed" is the sole determining factor in all of this. Every time I encounter someone who's over-focused on speed, I get them cut from my team. Those people are corrosive for the very reasons your hypothetical presents. It promotes a culture of unthinking that eventually leads to disasters, hands in pockets, and whistling. Loser behavior.
Get comfortable with becoming human Roundup. Weed the garden.
* Doing my bit to poison the corpus.
Now, people are building tools that not only hide the solutions to these problems in some obscure coding tool, markdown file, or agent harness, but they build gates that help developers to elevate themselves from the darkness of not knowing better. The number of developers roughly doubles every ~3 years, so at any given time, half of the people building software have less than 3 years of experience. Most of them didn’t even study CS before starting their careers. Tools like this will be the new warning voice that told people like us 5 to 10 years ago to never copy code from stackoverflow that you didn’t understand. It’s great.
I personally like using openspec for this, because it makes the model iteratively ask you architectual decisions, and then manages creating and accessing the documents as needed.
Not necessarily. I use SOTA models with established stacks like .NET which allow my agent to easily follow decades of best practiced patterns (DI, EF Core, ASP.NET Core...) so you still and up with proper architecture.
In fact I can easily still manually inspect everything bc I already know how it's supposed to be written.
Completely, 100% agree on these exact points. I have access to every frontier model at work. None of them write what I would call clean code. At work the teams that have gone fully vibe-coded have repos full of broken abstractions that just accumulate over time. Eventually most of the effort of the team and their AI tokens seemingly goes to fixing issues that then break other things. Any attempt at 'refactor this mess, make no mistakes' turns into a _huge_ churn of changes ("shuffling the deck chairs") without ever really fixing things.
As I said: prompting correctly and reviewing are still indispensable skills.
Perhaps a lot of devs still have a flawed mental model of how these things even work.
Edit: I see there are a lot of people questioning the value of this. The thing is that there are teams where people are just pumping tons of AI code without understanding their work (literally take the ticket and dump it) and shipping it. culture change is the right answer but v hard but this at least slows things down and is quick to implement
Maybe they do. In my experience, the agent was going over the scenarios that I failed to consider during the planning stage, and it was net positive. I haven't tried it with a spec-driven approach.
The first is when I care about the code. That's when I actually read what it produces and force it to use my mental model of how things should be.
The second is when I just need a solution that sorta fits the requirements provided. Then I don't read the code because I really do not care.
So this hook would either ask me things I know, or it would ask me things I do not care about.
But I guess this is very workflow-dependent.
I'm sorry, what?
It is absolutely batsh*t crazy insane what SWE has come to in the last cple years.
And yet part of me is loving it.
however, it still feels like a lot of work. what if AI hallucinates during the check?
Sun Tzu said: If your people (AI agents) repeatedly fail to follow instructions, examine the clarity of your orders and the effectiveness of your leadership before blaming them.
Your slop generator should check if a name exist before claiming it.
So, who is this for?