Skip to main content

FediForum October 2026

Merits/Drawbacks of AI-Assisted Coding on the Open Social Web

Topics AI Developer experience

Convener: Julian Lam (@julian@activitypub.space)

Participants who chose to record their names here:

Notes

  • Terms:

    • Necessary to define because some believe all AI assisted code is vibe code

    • Vibe-coding: the process of creating software through the use of LLMs, but intentionally or unintentionally avoiding reading of the code, and iterating on end-user testing only

    • AI Assisted coding: leveraging the abilities of AI/LLMs to enhance existing coding ability

  • NodeBB uses AI in some coding.

  • Where do we draw the line between vibe-coding and AI assisted coding?

  • Jaz

  • Seth

    • FediDeck - Dabbling in the code, AI helps me build it. Would not be able to write it completely without help.

  • Christoph

    • Callback to steam engines and people choosing not to use them

    • It’s a tool that’s so powerful that not using it leaves you behind

  • Jaz: Some thoughts on declaration of use, drawing the line, definitions, framing:

  • Stefan

    • Disclosure is super important

    • “because of bad sentiment, I see people are afraid to [disclose]”

  • Christoph

    • OpenPeeps also uses AI assisted coding

    • “at the end it’s your code and you’re responsible for what you do [with it]”

    • OpenAI should be held responsible for their “rogue agents”

  • Stefan

    • “I also hope this discussion becomes dated”

    • Hoping disclosure helps build bridges

  • Anca

    • Local LLMs and open-source models

    • We have an opportunity to train a model for Open Source coding. What if we do that?

  • Evan

    • https://cosocial.ca/@evan/116217855307998592 “Is it OK to run software written with the help of AI/LLMs on the Fediverse?”

    • Over 50% solid no

    • Noted persistent use of the “infection” analogy in relation to use of LLMs in codebase – an interesting an absolutist idea

    • “Institutional vs populist” divide – software developers tend to be moderate/enthusiastic, passionate anti-AI contingent on the fediverse

      • continues to be a problem: brigading, attacked

  • Mattias

    • As an open-source project, you cannot guarantee that there is no AI code in your codebase

    • Even if you as the main dev don’t use it, you can’t guarantee no AI code

    • Question is not: good vs bad, but rather how can we have the discussion over AI without the shame aspect

  • Johanna B

    • I’m at the intersection of non-coders/coders

    • In this circle of fediverse developers/users, there is a high degree of trust

    • A lot of the opposition is based from a world where everything is “AI” (used as a marketing term)

  • Jaz

    • Recalls story of a hired employee who handed all his work to his buddy on the west coast

    • You can’t trust anything from anybody anyway – this is nothing new

    • 95/98/01 – world wide web; lots interested, but lots fearful

      • These conversations are powered by an incredibly fast-paced moment in time – all of these viewpoints will change

      • The q of Declaration is not whether we should or not, it’s a PR question

      • … and what does that declaration look like?

  • Georgia

    • Story: Worked in tech for many years – In the last 10-15 years of teaching pole dancing, the public opinion toward it has changed completely

    • Where the industry might go: analogy to cars – when they were first invented, only the wealthy had them. Eventually, affordability was achieved, leading to much disaster, and road rules were introduced, etc. mechanics, licensing

    • Explosion of vibe coding will go somewhere like that – we’re in the early adoption phase

    • Would love to see some stewardship or mentorship from fediverse developers to help people use ai assisted coding safety and effectively

    • Anca (in chat): If the people selling us vibe-coding weren’t obviously evil we would be more excited about it.

    • Johanna (in chat): The rise of car culture is a good analogy, and also a warning! We are backsliding now with respect to the safe, secure, and regulated use of cars … because they became regulated and perceived as safe, we are now neglecting safety in new design and deployment of cars (largely driven by tech companies getting into the auto industry).

    • Stefan (in chat) cars even remain dangerous even after all those rules. so if there were anti car people they were not wrong, even after the laws an seat belts.

    • Georgia: “thats a really good point johanna and stefan re the car analogy - hopefully we can learn from both the successes and failures there when it comes to stewarding in the era of ai assisted coding”

  • Olivia

    • It is not ineffective to resist against AI. Just because it feels inevitable does not mean it is.

    • Just because you can’t tell if a contributor is using AI in their PR does not mean it is okay. If they are doing it in a project that has explicitly said they do not want LLM contributions, they are violating the rules of that community. Do we want to support that. The resistance to LLM usage is not unreasonable.

    • https://blog.codeberg.org/protecting-our-floss-commons-from-llms.html Codeberg recently ran a poll and updated their policies to disallow AI generated code

  • Jim DeLaHunt: caution about the financial unsustainability of the AI/LLM industry

  • Evan:

    • Makes it a point to thank people for contributing to the fediverse when they disclose AI use

    • “I hate wading into it, you’ll get told to go **** yourself; nature of the process”

    • Having explicit AI policies helps a lot

  • Julian

    • I’ve noticed changes in the collaborative aspects of coding - I don’t know if I’m talking to human when I get a response. Sometimes people respond w/ an AI-generated response, which feels like they didn’t even value my question enough to make a human response.

    • previously limited by time, energy, interest - now with the aid of ai assisted coding more people can build and prototype and imagineeer

  • Johanna

    • Observed changing policy at university - from 2024 “do what you feel is right” to now “disclose what you’re doing, and don’t use AI on things where you’re trying to develop skills”.

    • suggestion to focus beyond the “ai” complaint to the actual complaint underneath - eg it is disingenuous engagement, is it misinformation, what is the actual policy or standard or norm we want to enforce

  • Christoph

    • OpenPeep’ss principle: “Don’t build your house on someone else’s land.” That includes relying on someone else’s tools with no alternative of one’s own.