6 votes

Who cleans up after the vibe-coding party?

10 comments

  1. Kremor
    Link
    Archive

    Archive

    Earlier this year, a group of researchers led by Miklós Koren, an economist at Central European University in Vienna, released a paper titled “Vibe Coding Kills Open Source”. [...] In experiments with six coding models, the researchers found that open-source software packages that were frequently recommended by models saw large increases in download numbers — but that activity did not translate into the engagement that typically sustains maintainers. The paper concludes that, “under the traditional business model, where developer revenue depends entirely on direct user engagement, the open-source ecosystem cannot survive widespread AI adoption”.

    If everybody is using the same tools for writing code anyway, what is the value of collaboration, especially with those who don’t know their way around your project? In the past, a new contributor would gradually build knowledge about your project and hopefully one day become a trusted maintainer. But if they simply point a tool at it, there’s no reason to believe that accepting their contribution will lead to any growth in their knowledge of, or investment in, your project. And why should other developers, maintainers and potential employers assign any value to those contributions?

    Comeau worries about a generation of developers whose coding education is supplanted by interactions with LLMs, which often try to solve problems directly and narrowly, rather than by providing broader context. “AIs will know how to answer the question that you have, but you don’t know what questions you should be asking that you’re not asking,” he said. The lack of a bigger-picture understanding “seems like a serious skill deficit that’s going to become a problem when the current generation of developers, who learnt before these tools existed, start to retire or move on”.

    “Until very recently,” Rich Harris told me, “the idea that you would effectively pay rent to a Silicon Valley company for the privilege of being able to write software would have been considered completely absurd.”

    “One of the beautiful things about software development is that it’s completely permissionless,” Harris said. “Anyone can do it, and the tools that we use to create software are themselves fully available open-source software. Now, suddenly, it’s completely normal to have to pay $100 or $200 per month in order to do software development.”

    “In general, writing code the first time was never the problem for any project,” says cURL’s Stenberg. “The challenge for any project is maintaining it over time, fixing bugs over time. If you don’t understand it, you have to rely on the AI to fix all of the problems, going forward.”
    He continued: “They are not that good at fixing the problems; they’re much better at finding the problems.”

    Nearly 60 years ago, the artist Mierle Laderman Ukeles wrote an essay titled “Manifesto for Maintenance Art 1969!”. She argued that the work of maintenance is often unappreciated and devalued — “a drag; it takes all the fucking time” — next to the more celebrated work of creation. “After the revolution, who’s going to pick up the garbage on Monday morning?” asked Ukeles, who has served as the artist in residence for the New York City Department of Sanitation since 1977. She argued that society “confers lousy status” on those who do maintenance.

    This is perhaps truer than ever in the era of the “creator economy” and tech industrialists who burn and build and move fast and break things. To the “creators” go wealth and prestige while the maintainers work in the background, scraping rust and justifying their existence to a society that takes them for granted.

    8 votes
  2. [3]
    hamstergeddon
    Link
    At work we've got a number of initiatives vibe-coded by executives or product people who quickly shat out a proof-of-concept and then handed it off to real developers to build upon their...

    At work we've got a number of initiatives vibe-coded by executives or product people who quickly shat out a proof-of-concept and then handed it off to real developers to build upon their "efforts". And these are alarmingly being fast-tracked into production for client usage. It's nothing critical, fortunately, but still.

    AI enables the most toxic of executive (and product) traits -- I want something, I want it now, and I don't want to put the time/effort into doing it right.

    7 votes
    1. zod000
      Link Parent
      You are describing the exact situation I am going through. My CEO direct has been handing off his slopications to us saying "it's basically all done". It is going as well as you imagine it is.

      You are describing the exact situation I am going through. My CEO direct has been handing off his slopications to us saying "it's basically all done". It is going as well as you imagine it is.

      2 votes
    2. winther
      Link Parent
      AI is definitely very efficient at giving other people work. On average, I think AI has made me less productive, because the non-developers in the company are moving very fast, which is impressive...

      AI is definitely very efficient at giving other people work. On average, I think AI has made me less productive, because the non-developers in the company are moving very fast, which is impressive in its own right and I understand their excitement as they get to automate internal processes that have normally been down-prioritized by the developers, because we need to focus on the customer facing product. However, their constant new ideas gives us more work for questions, new requests for data access, changes in existing internal tooling and several meetings with compliance on why we can't just give AI agents full access to our database and production system. All makes for less time to do actual development on our product.

      2 votes
  3. papasquat
    Link
    I'm seeing this in the enterprise space too. We've been extremely conservative with ai usage. We only have sanctioned licenses for about 20 of our thousands of users, we're very rigorous with what...

    I'm seeing this in the enterprise space too. We've been extremely conservative with ai usage. We only have sanctioned licenses for about 20 of our thousands of users, we're very rigorous with what we do and don't allow them to do, and we have regular communication and monitoring with them.

    Even still, there's been a proliferation of hundreds of "agents". Most of them already forgotten and abandoned, but some of them becoming parts of business processes, and I have no doubt in my mind that as time goes on they'll become more and more critical to very important business processes. We have no good way of differentiating between the two, and no good long term plan for cleaning them up.

    Its like the ever present shadow IT problem cranked up to 11 and injected with steroids. With traditional shadow IT, at least the outputs were generally deterministic. This stuff is a mess.

    2 votes
  4. [2]
    skybrian
    Link
    I imagine there will be more open source projects that rarely accept pull requests and ask for donations instead, because they’d rather use a coding agent themselves.

    I imagine there will be more open source projects that rarely accept pull requests and ask for donations instead, because they’d rather use a coding agent themselves.

    1 vote
    1. Kremor
      (edited )
      Link Parent
      I think the article asking what's the point of open source at that point then? What will happen to critical projects when their maintainers retire and no one else has a good idea on how they work?...

      I think the article asking what's the point of open source at that point then? What will happen to critical projects when their maintainers retire and no one else has a good idea on how they work?

      Think of it this way, would Linux have become what it is today if only a selected few were allowed to contribute to it? In the age of AI, would something like Vue be possible, where its creator can make a living through Patreon, if AI recommends the more popular tools like React? And would Godot have built enough momentum to become an alternative worth considering?

      3 votes
  5. [3]
    zod000
    Link
    I call "Not it!". Just kidding, I'll definitely be one of the cleaners.

    I call "Not it!".

    Just kidding, I'll definitely be one of the cleaners.

    1 vote
    1. [2]
      hamstergeddon
      Link Parent
      Years ago I did agency web development work and we started to see the rise of wix and all these other WYSIWYG website builders. At first it seemed like a threat, but then we started to get new...

      Years ago I did agency web development work and we started to see the rise of wix and all these other WYSIWYG website builders. At first it seemed like a threat, but then we started to get new clients who had gone that route, hit the limit of their own capabilities, and came to us to build something custom.

      So I'm hoping this ends like that. Where there's just a ton of work for developers when this inevitably crashes, stalls, or prices itself out of viability.

      4 votes
      1. zod000
        Link Parent
        I did that as well at the time, though free lance as a side job, and the rise of Wix/Wordpress/etc made me stop doing it. The reason was that so many of the potential clients had come to the...

        I did that as well at the time, though free lance as a side job, and the rise of Wix/Wordpress/etc made me stop doing it. The reason was that so many of the potential clients had come to the conclusion that web development was now trivial because they had already done so much with Wix or equivalent and they expected extremely quick results for very little compensation. I am feeling the same vibe (no pun intended) with the AI built prototypes. I don't enjoy needing to be the bad guy, but it has become part of my role to directly point out how trivial some of the "applications" we're handed are and that a real application will require a great deal more work. I do think that these can work really well as prototypes to show off ideas and intent, so there is definitely value there, but nothing that has been handled off by an executive has been correct or anywhere approaching something we'd keep and maintain.

        3 votes