logo elektroda
logo elektroda
X
logo elektroda
Advanced Search
logo elektroda

Open source projects are closing their gates as a result of AI. A flood of low-quality code

p.kaczmarek2 1020 5

TL;DR

  • Open source maintainers at curl, ghostty, and tldraw are tightening contribution rules because AI-generated code has become a major source of noise and frustration.
  • The main complaint is “vibe coding”: plausible-looking patches that often include wrong assumptions, fake vulnerabilities, or cosmetic changes with no real value.
  • Maintainers say they must manually filter through hundreds of pointless proposals, turning supposed productivity gains into extra work.
  • curl shut down bounty programmes for vulnerability reports, ghostty bans outsiders from using AI, and tldraw blocks all PRs from external users.
Generated by the language model.
ADVERTISEMENT
Treść została przetłumaczona polish » english Zobacz oryginalną wersję tematu
Report a violation of the law
Reply Cool? Ranking DIY | New topic
Notify about new articles
📢 Listen (AI):
  • Open source projects such as curl, ghostty and tldraw are fed up with ubiquitous code-generating AI tools. Curl shuts down bounty programmes for vulnerability reports, ghostty bans outsiders from using AI, and tldraw goes one step further - blocking all PRs (change proposals) from external users. What was not long ago portrayed as a revolution to increase developer productivity has today become a source of frustration, extra work and a deterioration in the quality of collaboration for many repository maintainers.

    The common denominator of the problem is so-called 'vibe coding' - automatically generated patches, features and submissions that often look plausible, but in practice contain incorrect assumptions, non-existent vulnerabilities, or cosmetic changes with no real value. Instead of speeding up development, artificial intelligence is increasingly slowing down project maintainers, forcing them to manually filter through hundreds of pointless proposals. As a result, some developers are coming to the conclusion that the only effective strategy is to set hard boundaries.

    This is well illustrated by Mitchell Hashimoto's statement on X (formerly Twitter), who speaks directly about the drastic decline in the quality of PRs.


    Hacker News (https://news.ycombinator.com/item?id=46641042) users are also sceptical about AI tools. Some point out that LLMs-based systems are not bad in themselves, rather the problem is the way they are used - especially the mass sending of low-quality pull requests without understanding the code and context of the project. AI is not a substitute for knowledge and does not make anyone a good programmer; on the contrary, AI creates a flood of 'weak programmers', working against the open source scene.

    Examples of submissions:
    https://github.com/ghostty-org/ghostty/pull/10316
    Screenshot of a GitHub comment where Copilot AI suggests moving a hardcoded API key from code to environment variables.
    https://github.com/ghostty-org/ghostty/pull/10270
    Screenshot of a GitHub comment criticizing a code change as AI-generated and low quality.
    https://github.com/ghostty-org/ghostty/pull/10205
    Screenshot of a GitHub comment criticizing low-quality AI-generated pull requests

    Sources:
    curl:
    https://arstechnica.com/security/2026/01/over...-bug-bounties-to-ensure-intact-mental-health/

    tldraw:
    https://github.com/tldraw/tldraw/issues/7695
    https://news.ycombinator.com/item?id=46641042
    https://tldraw.dev/blog/stay-away-from-my-trash

    ghosts:
    https://github.com/ghostty-org/ghostty/pull/10412

    Have you encountered the "ai slop" problem in a programming context? I've already encountered it and it's in my project - a nonsense PR meant to add rotary encoder support was created contrary to the basic logic and organisation of my project....

    Cool? Ranking DIY
    Helpful post? Buy me a coffee.
    About Author
    p.kaczmarek2
    Moderator Smart Home  
    Offline 
    p.kaczmarek2 wrote 14387 posts with rating 12308, helped 650 times. Been with us since 2014 year.
  • ADVERTISEMENT
  • #2 21819112
    andrzejlisek
    Level 31  
    Posts: 3635
    Help: 82
    Rate: 705
    Board Language: polish
    I wonder when developers will learn that asking AI for a demonstration example of how to implement a particular issue is not the same as having AI write or rewrite an entire programme.

    I do use AI, but I never mindlessly insert AI-generated code. I always look at what it has written and manually adapt it to the program I am creating. Having the program automatically corrected has never even been tried.
  • ADVERTISEMENT
  • #3 21819119
    p.kaczmarek2
    Moderator Smart Home  
    Posts: 14387
    Help: 650
    Rate: 12308
    In such a situation, often the boss or there principal recognises that you are inefficient, and that competitors can add features faster. In my experience, many people in senior positions do not know what technology debt is.
    Helpful post? Buy me a coffee.
  • ADVERTISEMENT
  • #4 21819506
    marweg1967
    Level 14  
    Posts: 131
    Help: 1
    Rate: 39
    Board Language: polish
    p.kaczmarek2 wrote:
    you are inefficient

    It's called 'braking'.

    p.kaczmarek2 wrote:
    In my experience, many people in senior positions do not know what technology debt is

    I agree. I get the impression that they don't think about what's going to happen in a week's time, let alone the longer term. Previously I thought they just wanted to prove themselves to their superiors, but unfortunately this applies to business owners too. Except that it's more about the heirs who came "at the ready" and not those who created the company in question.
  • ADVERTISEMENT
  • #5 21828363
    p.kaczmarek2
    Moderator Smart Home  
    Posts: 14387
    Help: 650
    Rate: 12308
    Mitchell Hashimoto already wants to pretty much shut down the option of PRs (proposals for change, simplifying) in his open source projects:
    Helpful post? Buy me a coffee.
  • #6 21858594
    Sam Sung
    Level 33  
    Posts: 2000
    Help: 227
    Rate: 570
    Board Language: polish
    andrzejlisek wrote:
    I use AI, but I never mindlessly insert code generated by AI. I always look at what it has written and manually adjust it to the program I am creating. Having the program corrected automatically has never even been attempted.
    If there is more such AI assistance, the pleasure of writing a program will turn into the anguish of reading the excrescences not even of a colleague, but of a tireless robot.
    Maintainers of the aforementioned OpenSource projects also do not want to insert AI-generated code mindlessly, but the patience to read with understanding what a bot (poorly prompted by someone) has written is running out.
    p.kaczmarek2 wrote:
    Mitchell Hashimoto already wants to pretty much shut down the option of PRs (change proposals, simplifying) in his open source projects
    For now, he is "first considering"; and in the meantime, projects "anti-slop", "slopcannon", ... are developing.
Reply Cool? Ranking DIY | New topic
Notify about new articles
📢 Listen (AI):
Report a violation of the law

FAQ

TL;DR: 3 open‑source projects changed policies due to AI “slop”; “drastic decline in the quality of PRs.” [Elektroda, p.kaczmarek2, post #21819040]

Why it matters: Maintainers need clear rules to reduce low‑value AI PRs while keeping real contributors engaged.

Who this is for: repo maintainers, contributors, and engineering managers asking how to handle AI‑generated code without burning review time.

Quick Facts

  • curl paused bug bounties, citing AI‑generated noise and maintainer mental health concerns. [Ars Technica, 2026]
  • tldraw temporarily blocked all external PRs to stabilize quality and velocity. “Stay away from my trash.” “Stay away from my trash”
  • Ghostty restricted AI‑assisted contributions after low‑value PRs appeared. [Elektroda, p.kaczmarek2, post #21819040]
  • “Vibe coding” = plausible‑looking but incorrect auto‑generated patches; maintainers must sift through them. [Elektroda, p.kaczmarek2, post #21819040]
  • Community sentiment: tools aren’t bad; mass low‑understanding PRs are. [“Hacker News discussion”]

What’s actually happening with AI and open‑source PRs?

Maintainers report a surge of “vibe‑coded” PRs that look right but misunderstand project context. These add review load, not features. Some projects responded with stricter policies or temporary PR freezes. The goal is to protect velocity and reduce reviewer burnout. “AI is not a substitute for knowledge.” [Elektroda, p.kaczmarek2, post #21819040]

Which projects publicly changed policy?

At least three named in the thread: curl paused bug bounties, tldraw blocked external PRs, and Ghostty restricted AI‑assisted submissions. This illustrates a defensive posture to keep quality high amid noisy contributions. [Elektroda, p.kaczmarek2, post #21819040]

What is “vibe coding”?

“Vibe coding” refers to AI‑generated changes that appear reasonable but embed wrong assumptions or trivial edits. They often waste maintainer time without improving functionality or security. The thread lists examples and the resulting frustration for reviewers. [Elektroda, p.kaczmarek2, post #21819040]

How do I set an AI‑code policy for my repository?

  1. State what AI assistance is allowed (e.g., summaries yes, code no) in CONTRIBUTING.md.
  2. Require authors to declare AI use and provide reproduction steps and tests.
  3. Enforce via PR template and automated checks, then iterate based on review metrics. [Elektroda, p.kaczmarek2, post #21819040]

Are AI tools always bad for open source?

No. Contributors note AI is useful for examples, drafts, and refactors when humans validate and adapt outputs. The problem is unvetted code pasted into PRs. “I never mindlessly insert AI‑generated code.” [Elektroda, andrzejlisek, post #21819112]

What risks do maintainers face if they leave PRs open to AI slop?

Risks include review backlogs, missed real bugs, and demoralized maintainers. Managers may misread slower merges as developer inefficiency, ignoring growing tech debt from noisy submissions. That dynamic pressures teams to ship risky changes. [Elektroda, p.kaczmarek2, post #21819119]

Is there a real example of a bad AI PR?

Yes. The thread author cites a “nonsense PR” to OpenBK7231T_App that attempted rotary‑encoder support contrary to the project’s architecture. Such PRs consume review time without adding value. [Elektroda, p.kaczmarek2, post #21819040]

How should contributors use AI responsibly?

Use AI for drafts, test scaffolds, and ideas. Then verify logic, align with project patterns, and write passing tests. Do not ask maintainers to debug AI output. Declare AI use in the PR description. [Elektroda, andrzejlisek, post #21819112]

What trade‑offs come with blocking all external PRs?

Quality and focus can improve, but drive‑by fixes may slow. Communication must invite issues, design discussions, or curated patches later. This policy is a circuit breaker, not a permanent wall. [“Stay away from my trash”]

How can managers avoid mislabeling careful review as inefficiency?

Track review load, PR rejection rates, and defect escape. Show how filtering low‑quality submissions protects roadmap delivery. Educate stakeholders on tech‑debt costs from rushed merges. [Elektroda, p.kaczmarek2, post #21819119]

What is OpenBeken?

OpenBeken (OpenBK7231T_App) is an open‑source firmware project referenced by the thread author; it targets IoT devices and accepts community contributions. The cited PR shows how misaligned AI changes can appear. [Elektroda, p.kaczmarek2, post #21819040]

What is Tuya?

Tuya is an IoT platform and device ecosystem used by many smart‑home products; developers often reflash Tuya‑based devices with custom firmware. It’s relevant when discussing open firmware like OpenBeken. [“Tuya”]

What is Arduino Nano?

Arduino Nano is a compact, breadboard‑friendly microcontroller board for embedded projects. Contributors may prototype fixes or repro cases on Nano‑class boards before submitting patches. [“Arduino Nano”]

What is CAN bus?

CAN bus is a robust, differential serial network used in vehicles and industry. It appears in open‑hardware discussions when integrating firmware with sensors or controllers. [“Controller Area Network”]

Any stats that show the scale of change?

A minimum of three prominent projects in one thread changed policies in response to AI‑generated noise. That indicates a measurable shift in governance to protect maintainer time. [Elektroda, p.kaczmarek2, post #21819040]

Edge case: what if an AI PR passes tests but is conceptually wrong?

Reject it. Tests can miss architectural violations or undefined behavior. Require design notes and link to prior issues. Ask for targeted benchmarks or property tests to validate intent. [Elektroda, p.kaczmarek2, post #21819040]
Generated by the language model.
ADVERTISEMENT