> Beads, which is basically some sort of issue tracker for agents, is 240,000 lines of code that … manages markdown files in GitHub repositories. And the code quality is abysmal.
I'd heard of beads as a lightweight issue tracker for agents, so this gave me a real shock. What could all that code POSSIBLY be doing? Going to the repo and poking around, I truly cannot tell. There's an enormous `docs/` folder with no hierarchy, containing files like `MULTI_REPO_HYDRATION.md`, which "describes the implementation of Task 3 from the multi-repo support feature (bd-307): the hydration layer that loads issues from multiple JSONL files into a unified SQLite database," and `ANTIVIRUS.md`, a 7KB text file about how `bd.exe` sometimes gets flagged as untrustworthy by antivirus software.
I opened a random go file, `detect_pollution.go`. This is a CLI command for detecting and cleaning up test tickets from a production database by (1) scanning ticket titles for testing-related prefixes like "debug," "test," or "benchmark," (2) scanning for short descriptions, (3) scanning for suspicious phrases like "sample ticket," and (4) scanning for batches of tickets that were created all at once. It uses these signals to compute a confidence score for each ticket that determines whether it should be deleted. This command was deprecated and replaced by `doctor_pollution.go`, which reimplements large parts of `detect_pollution.go` and is not, at a glance, substantially different. Two seconds of thought will tell you that this feature is unnecessary, since you can create tickets with a "#test" tag and then delete them by tag.
I don't want to come across as mean, but Steve should be embarrassed by this. It's grotesquely baroque and completely unmaintainable—proof positive that whatever he's doing isn't working.
Re: completely unmaintainable. He's freely admitted that he's never read the code and it's only written by agents. So to the extent that it's maintained, it is done by agents who seems to be doing something, somehow.
If only 'grotesquely baroque & completely unmaintainable' were congruent with 'not working'. A lot of software in production is exactly as you describe, funnily enough.
This is the entire reason the average consumer has lost trust in Software & developers.
I don't understand if it's fun for people(in the software development trade) to see everyone complaining about Software...
I as a software developer honestly feel ashamed in the quality of software we provide out there.
I think LLMs should instead be used to automate grunt work to make software better for edge cases, or where you can use it to get more time to improve software quality.
Generally if a program has a good --help I'd recommend at least evaluating somewhat reasonably what the subcommands are, before trying to go code diving.
> As a maintainer many PRs now look like an insult to one’s time, but when one pushes back, the other person does not see what they did wrong.
I have been volunteering as an advisor to various master's and PhD theses, giving feedback on theses texts and papers. I see people using AI to write their texts more and more, and I feel like my hours are now wasted on improving AI-generated texts instead of helping people hone their writing and thinking skills. Since I cannot constantly analyze and think about who actually wrote the texts, I am thinking about stopping my volunteering.
Thoughtful writing matters so much in an AI world.
The writing and editing is thinking. Trying to figure out whether we communicate our ideas well matters. Giving that over to the AI means giving up on critical thinking.
If we don’t think through the implications of our words, we delegate work without thought of for consequences. That’s especially problematic if AI actually does something for us like code.
> I see people develop parasocial relationships with their AIs, get heavily addicted to it, and create communities where people reinforce highly unhealthy behavior.
In the end, the biggest difference between the enthusiasts and the skeptics might be “do you enjoy talking to robots.” The rest is downstream of whether you find endless prompting fun or annoying.
All I know is that when I watch someone at 3am, running their tenth parallel agent session, telling me they’ve never been more productive
... okay, I'll bite. What is actually being made here?
These people are so productive, running 10 checkouts of a repo with Claude or whoever... Code must be flying out. I'm sure github is seeing a rise in lines pushed faster than ever.
I am not seeing an explosion of products worthy of any cents out of this, though, at least nowhere near what is being evangelised by the "trust me bro, we're productivity gods now".
Where is the output of all these tokens going, when you wake up the next morning?
I've used AI quite a lot. Enough to know that an inference state machine is an inference state machine.
I want to see it, I want to believe! Show me the goods! Stop telling everyone how productive you are and show the finished work.
At least the post seems to be rightfully conclusive that people are going to go _insane_.
Vibecoding slop every night, waking up the next morning, starting again, and again. Without any meaning or end; I suspect these people will quit and move on to something else. I've been programming, probably averagely, for over 25 years -- because I like computers -- not because I like being a productivity junkie, shooting on dopamine.
I've been using AI to systematically go through my (extremely) long list of pet projects/fixes/gripes/things that should exist, and it's been fun, but I agree with the sentiment that it has to be for the sake of improving _something_.
> We should also remember that current token pricing is almost certainly subsidized. These patterns may not be economically viable for long. And those discounted coding plans we’re all on? They might not last either.
In The Resilient Farm and Homestead by Ben Falk, you'll find a sidebar on page 28: Oil to Soil--Use it or Lose It: Leveraging the Cheap-Oil Window for Maximum Effect. He says, "we have made the conscious decision to take advantage of the small window of time still remaining with which to develop intergenerational land and infrastructure systems, which greatly enables long-term production of the site without any oil input for hundreds if not thousands of years."
I think of subsidized LLM tokens like this. Use them to build developer tools. Ideally, these developer tools will work with and without further LLM use. Then it won't matter if token prices fall forever, or if the subsidies end and nobody can afford AI-assisted development.
> You can use Polecats without the Refinery and even without the Witness or Deacon. Just tell the Mayor to shut down the rig and sling work to the polecats
A funny read with a grain of wisdom too, I suppose. Is the following the future of open source, I wonder? And will GitHub face the same slop-destiny as mainstream social media?
> I'm not sure how we will go ahead here, but it’s pretty clear that in projects that don’t submit themselves to the slop loop, it’s going to be a nightmare to deal with all the AI-generated noise.
> Some projects no longer accept human contributions until they have vetted the people completely.
Also reminds of the following recent piece that talked about increasing (or exploding?) verification debt:
> will GitHub face the same slop-destiny as mainstream social media
At the very least because it's now human + coding agent, separating out the human input from the machine output in pull requests becomes necessary in my book. There are dramatic differences in prompting styles that can have completely different qualities of output and it's much easier to tell it apart from the prompts than from the outputs given that it's basically an amplification problem.
> separating out the human [input] from the machine
I was thinking more generally and thus put the noun input in parenthesis in the quote. With agents and slop, the value for humans being there may quickly spiral down. There are also a lot of bad stuff already there, including malware and such.
If you have your own infrastructure instead of a mega-platform, you can control these things more easily.
The value in open source code was never the code. It was the trust that was created around it that it becomes a place for useful innovation, for trust, for vetting, for keeping dependencies low.
I can build my own curl in a week, but the value that curl gives me is that it's a multi decade old library, by a person that has dedicated his live to keeping the project there, keeping a quality bar etc.
The value of curl is not curl, it's the human behind it.
The human behind it, the community using it critically, and the years of battle hardening.
The great open source tools out there have handled, worked around, or influenced away many many bugs and edge cases out in the real world, many of which you wont think of when initially designing your own. The silent increase in stability and productivity resulting from this kind of thing is as vast as it is hard to see/measure. It feels like the quote about expertise saying someone "has forgotten more than I'll ever know about [subject]".
Thank you to everyone powering our collective work.
> Beads, which is basically some sort of issue tracker for agents, is 240,000 lines of code that … manages markdown files in GitHub repositories. And the code quality is abysmal.
I'd heard of beads as a lightweight issue tracker for agents, so this gave me a real shock. What could all that code POSSIBLY be doing? Going to the repo and poking around, I truly cannot tell. There's an enormous `docs/` folder with no hierarchy, containing files like `MULTI_REPO_HYDRATION.md`, which "describes the implementation of Task 3 from the multi-repo support feature (bd-307): the hydration layer that loads issues from multiple JSONL files into a unified SQLite database," and `ANTIVIRUS.md`, a 7KB text file about how `bd.exe` sometimes gets flagged as untrustworthy by antivirus software.
I opened a random go file, `detect_pollution.go`. This is a CLI command for detecting and cleaning up test tickets from a production database by (1) scanning ticket titles for testing-related prefixes like "debug," "test," or "benchmark," (2) scanning for short descriptions, (3) scanning for suspicious phrases like "sample ticket," and (4) scanning for batches of tickets that were created all at once. It uses these signals to compute a confidence score for each ticket that determines whether it should be deleted. This command was deprecated and replaced by `doctor_pollution.go`, which reimplements large parts of `detect_pollution.go` and is not, at a glance, substantially different. Two seconds of thought will tell you that this feature is unnecessary, since you can create tickets with a "#test" tag and then delete them by tag.
I don't want to come across as mean, but Steve should be embarrassed by this. It's grotesquely baroque and completely unmaintainable—proof positive that whatever he's doing isn't working.
Re: completely unmaintainable. He's freely admitted that he's never read the code and it's only written by agents. So to the extent that it's maintained, it is done by agents who seems to be doing something, somehow.
If only 'grotesquely baroque & completely unmaintainable' were congruent with 'not working'. A lot of software in production is exactly as you describe, funnily enough.
This is the entire reason the average consumer has lost trust in Software & developers.
I don't understand if it's fun for people(in the software development trade) to see everyone complaining about Software...
I as a software developer honestly feel ashamed in the quality of software we provide out there.
I think LLMs should instead be used to automate grunt work to make software better for edge cases, or where you can use it to get more time to improve software quality.
Generally if a program has a good --help I'd recommend at least evaluating somewhat reasonably what the subcommands are, before trying to go code diving.
The --help page lists 70+ commands, most of which are intended for use by agents.
> As a maintainer many PRs now look like an insult to one’s time, but when one pushes back, the other person does not see what they did wrong.
I have been volunteering as an advisor to various master's and PhD theses, giving feedback on theses texts and papers. I see people using AI to write their texts more and more, and I feel like my hours are now wasted on improving AI-generated texts instead of helping people hone their writing and thinking skills. Since I cannot constantly analyze and think about who actually wrote the texts, I am thinking about stopping my volunteering.
Thoughtful writing matters so much in an AI world.
The writing and editing is thinking. Trying to figure out whether we communicate our ideas well matters. Giving that over to the AI means giving up on critical thinking.
If we don’t think through the implications of our words, we delegate work without thought of for consequences. That’s especially problematic if AI actually does something for us like code.
> I see people develop parasocial relationships with their AIs, get heavily addicted to it, and create communities where people reinforce highly unhealthy behavior.
In the end, the biggest difference between the enthusiasts and the skeptics might be “do you enjoy talking to robots.” The rest is downstream of whether you find endless prompting fun or annoying.
These people are so productive, running 10 checkouts of a repo with Claude or whoever... Code must be flying out. I'm sure github is seeing a rise in lines pushed faster than ever.
I am not seeing an explosion of products worthy of any cents out of this, though, at least nowhere near what is being evangelised by the "trust me bro, we're productivity gods now".
Where is the output of all these tokens going, when you wake up the next morning?
I've used AI quite a lot. Enough to know that an inference state machine is an inference state machine.
I want to see it, I want to believe! Show me the goods! Stop telling everyone how productive you are and show the finished work.
At least the post seems to be rightfully conclusive that people are going to go _insane_.
Vibecoding slop every night, waking up the next morning, starting again, and again. Without any meaning or end; I suspect these people will quit and move on to something else. I've been programming, probably averagely, for over 25 years -- because I like computers -- not because I like being a productivity junkie, shooting on dopamine.
Make it count.
I've been using AI to systematically go through my (extremely) long list of pet projects/fixes/gripes/things that should exist, and it's been fun, but I agree with the sentiment that it has to be for the sake of improving _something_.
Vibe coding is to software development as fad diets are to weightloss.
> We should also remember that current token pricing is almost certainly subsidized. These patterns may not be economically viable for long. And those discounted coding plans we’re all on? They might not last either.
In The Resilient Farm and Homestead by Ben Falk, you'll find a sidebar on page 28: Oil to Soil--Use it or Lose It: Leveraging the Cheap-Oil Window for Maximum Effect. He says, "we have made the conscious decision to take advantage of the small window of time still remaining with which to develop intergenerational land and infrastructure systems, which greatly enables long-term production of the site without any oil input for hundreds if not thousands of years."
I think of subsidized LLM tokens like this. Use them to build developer tools. Ideally, these developer tools will work with and without further LLM use. Then it won't matter if token prices fall forever, or if the subsidies end and nobody can afford AI-assisted development.
> You can use Polecats without the Refinery and even without the Witness or Deacon. Just tell the Mayor to shut down the rig and sling work to the polecats
amazing quote
I have tried using coding agents and was left doing more with myself, so I agree with the broader idea.
https://x.com/gdb/status/2013164524606775544?s=61 But who is gonna tell him.
I don’t understand what these CEO are up to.
They used to call us Rock Stars for writing PHP
Now everyone’s a DJ https://www.youtube.com/live/wc5j-HK4NS8
Not related to the content itself, but people using Psychology terminologies for wrong behavior is not acceptable.
I think one flaw with agentic coding: agents want to please the coder. They don’t push back on our opinions.
Hearing we’re geniuses all the time can create slop loops.
If we fail to be circumspect about a problem, to think through the implications of our decisions, we’ll produce thoughtless slop.
We have to avoid the dopamine high of “velocity” and take our time and ensure we remember all the real constraints for our problem.
the average glazer on HN will tell you that you can just use an agent to do this part as well.
A funny read with a grain of wisdom too, I suppose. Is the following the future of open source, I wonder? And will GitHub face the same slop-destiny as mainstream social media?
> I'm not sure how we will go ahead here, but it’s pretty clear that in projects that don’t submit themselves to the slop loop, it’s going to be a nightmare to deal with all the AI-generated noise.
> Some projects no longer accept human contributions until they have vetted the people completely.
Also reminds of the following recent piece that talked about increasing (or exploding?) verification debt:
https://cacm.acm.org/blogcacm/verification-debt-when-generat...
> will GitHub face the same slop-destiny as mainstream social media
At the very least because it's now human + coding agent, separating out the human input from the machine output in pull requests becomes necessary in my book. There are dramatic differences in prompting styles that can have completely different qualities of output and it's much easier to tell it apart from the prompts than from the outputs given that it's basically an amplification problem.
> separating out the human [input] from the machine
I was thinking more generally and thus put the noun input in parenthesis in the quote. With agents and slop, the value for humans being there may quickly spiral down. There are also a lot of bad stuff already there, including malware and such.
If you have your own infrastructure instead of a mega-platform, you can control these things more easily.
The value in open source code was never the code. It was the trust that was created around it that it becomes a place for useful innovation, for trust, for vetting, for keeping dependencies low.
I can build my own curl in a week, but the value that curl gives me is that it's a multi decade old library, by a person that has dedicated his live to keeping the project there, keeping a quality bar etc.
The value of curl is not curl, it's the human behind it.
The human behind it, the community using it critically, and the years of battle hardening.
The great open source tools out there have handled, worked around, or influenced away many many bugs and edge cases out in the real world, many of which you wont think of when initially designing your own. The silent increase in stability and productivity resulting from this kind of thing is as vast as it is hard to see/measure. It feels like the quote about expertise saying someone "has forgotten more than I'll ever know about [subject]".
Thank you to everyone powering our collective work.
Slop Loops is my favourite expression for the year so far.