Where are the third places for learners?

A developer and speaker at SummerCon 2026 argues that open-source projects are struggling to maintain quality and community as AI-driven contributions surge, citing policies from Debian, Zig, Servo, and the Rust rewrite of Bun using Claude Code. The author leans against using AI for coding, referencing studies linking excessive ChatGPT use to procrastination, memory loss, and lower GPAs among students.

Where are the third places for learners? Table of Contents A detailed commentary of the lack of middle spaces between application of high-level knowledge and learning low-level topics. Editor's note ❤️🩹 editor-s-note-heartadhesive-bandage In the spirit of improving third spaces, I wanted to write up a small opinion piece ahead of posting my code for my recent talk at SummerCon 2026. YouTube . If you haven't seen it, I would love for more eyes to be across the wacky ways you can make malicious code run on MacOS via DYLD Quick landscape on AI in open-source I’ve just begun properly contributing to open-source to deepen my understanding of systems programming beyond pwn challenges. In fact, I find researching low-level topics sometimes makes grasping high-level languages with weak types like Python a real pain. Boo My favourite way to stay motivated beyond tutorials is tackling real problems scaffolded by others’ work until I become more knowledgeable. So, if you’ve been reading up , many open-source project policies have strengthened due to the surge in AI-driven contributions. Notable examples to me are : Debian's conditional policy on AI contributors lwn.org, J. Brockmeier https://lwn.net/Articles/1061544/ Zig's banning of AI usage to preserve maintainer bandwidth ziglang.org https://ziglang.org/code-of-conduct/ Servo restricting AI-assisted code servo.org https://book.servo.org/contributing/getting-started The Rust rewrite of Bun using Claude Code GitHub https://github.com/oven-sh/bun/ RustFS's explosive growth in popularity with AI in mid-2025 GitHub https://github.com/rustfs/rustfs They're just examples I pulled from my search history. Developers are trying to reach a synthesis of how maintainers can create code, review it, and still allow an active community to provide both input in features and PRs. To put my cards on the table, I lean towards not using AI at the moment for clearly documented effects on cognition for those without underlying skill-sets paper https://link.springer.com/article/10.1186/s41239-024-00444-7 . If you're still skilled - it might be making the eggheads of us dumber too. paper https://www.thelancet.com/journals/langas/article/PIIS2468-1253 25 00133-5/abstract Our findings suggested that excessive use of ChatGPT can have harmful effects on students’ personal and academic outcomes ... students who frequently used ChatGPT were more likely to engage in procrastination ... these students also reported memory loss .. and had a poor CGPA. Muhammad Abbas et. al, 2024 First, penance - I have found it hard at times to not use it when really I should fail first. It's the human desire to have shit done for you. I'm told it's the new hotness, and I have used it to where I could be using it to go much faster than reading a textbook. It's probably got that book in there, anyway. wired.com https://www.wired.com/story/anthropic-settles-copyright-lawsuit-authors/ Let's just append this prompt response to my commit... The AI Contributor Epoch There have been so https://www.coderabbit.ai/blog/ai-is-burning-out-the-people-who-keep-open-source-alive many articles https://www.ft.com/content/cec8df9e-b43b-4cd1-8feb-c07e804e8d33 from the maintainers on these projects https://news.ycombinator.com/item?id=43078690 regarding I refer to the above and the policies created to stop the sludge. maintainer burden https://github.blog/open-source/maintainers/rethinking-open-source-mentorship-in-the-ai-era/ , correctness to code styling or what is idiomatic to them and possible copyright issues that may implicate their licenses.Open-source projects have been here before many times, and now we have a term for this - "contributor poker", where a maintainer will have to understand whether their own finite time is worth the investment into a new contributor https://kristoff.it/blog/contributor-poker-and-ai/ . The economic forces of open-source is tight, and further now AI tools have made 'lines of code' ever cheaper to produce. Commits are larger, the language space is concentrating and the laggard is usually the review step, and possibly the iteration of a CI/CD cycle. Personally as a person moving from business into security, I saw this especially with a recent review on Bun's rewrite that went over that specific promise in AI. Moreover, Bun is now stuffed with a backlog of PRs made by Claude Code https://lockwood.dev/ai/2026/07/27/how-is-the-bun-rewrite-in-rust-going.html , meaning more-and-more of the engineering cycle is handed off to a frontier model and the company that owns said models' employees. Sounds fine enough. But what about a person introducing themselves to development? Aside from Andrew's technical opinions, what happened to people who saw this as a novel inroad to Zig? Actually, what did Bun look like before the rewrite for an ecosystem? If you don't wish to open it up, it's a quick screen-grab of Bun's commits at November 2025 against July 2026. The breadth of contributors that are publicly visible against a mash of robobun/autofix-ci. What if someone liked Bun and wanted to learn from smarter folks in open-source against it? Do I now just talk to Claude by-proxy on how Bun works? Do we all fork? For all the comments on Bun being bad a bit funny because the nature of comptime is quite well explained by Zig and in Ziglings - easy to comment when you're not making big projects though , it had people working on it that loved it and it is ~~enterprise-ready~~. Frankly, I think that means it has gone to the dogs and is an amazing piece of marketing, engineering, and an example of us further losing space for open-source software and instead simply having public source code . I would wager that the much older Servo is a much better place to learn - it is for all it's faults , a much more active community shown in both watchers to the repository that explicitly desire the daily outcome of the project and the unique number of co-authors. A very valid argument is that this is simply a linear trend to time of a given project, and I do use Servo as a tongue-in-mouth example against RustFS somewhat, but I want to cast it against the attention economy that GitHub now runs on regarding stars and actual eyes-on-README. This measurement of stars meaning 'software is good' makes me ask so many questions on what this is all for lately, more than venture capital jousting. Thank God We All Know GitHub repository stars are rubbish At this point stars should just be considered what truly are at this point on GitHub - just like any other social media - it is reflective of how this project fits into algorithmic sharing, paid marketing campaigns and influencers. It is not a stamp of quality regarding content or community. It is the paid followers on X, it's view-botting YouTube, it's the algorithm and ad-tech economy letting you serve you lunch. Going to who runs the new dogs on the block, we could review contributors for these AI-native repositories. The contributors do not co-author as often, the actual author is opaque, it is successfully run by much smaller core teams. Because these teams are powered by AI, they also tend to submit much larger diffs between each commit. This on paper is amazing, I think we should give it that - truly a sign that the price of code has truly collapsed, and you know what... the code looks OK a fair bit of the time considering the floor we have for standards But that question again... where do learners or curious people approach this? How do they develop this eluded 'taste'? In fact, what if this prospective learner or even mid-level engineer , looking to up-skill, does not have a Claude Code or Codex plan? Could they 'keep up'? What if they are seeking to unlearn anti-patterns through working on free software with other humans on public problems and projects? What if they are working with a chatbot by proxy? This ended up being fine in many other eras in history, like Grace Hopper terming compilers from assemblers https://scalibq.wordpress.com/2013/02/28/assemblers-are-not-compilers/ , but I think we need to pump the brakes a little to consider what we are throwing away in this fever dream era. Have we fully considered the economic implications of an open-source project like this? Consider some arguments of how economic forces of low-cost code might shape open-source. - Will a new human contributor still incur costs against maintainers as they learn? - If the price for code inference is zero, will the contributor/public bear the cost of developing a frontier AI agent? How are these contracts arranged? - Alternatively, will these projects eventually be supported by fully automated, paid, proprietary systems? Do these respect the projects license? - Stretching out, will the public face a significant shortage of new contributors to address a growing number of computing problems exacerbated by AI? I don't think these are considered nearly enough and often the argument is that every anti-AI post is some inconsiderate Luddite when I really think we should already start mourning that there is no middle ground between our private projects and corporate enterprise projects and it's much more foreboding than any Rage Against The Machine lyric. We have lost the middle space open-source promised to hold onto. We are ruining our own third spaces Contributor poker is probably the phrase that has stuck with me the most in 2026. These foundations and open-source teams that create tools to solve the biggest of problems in computing often run on tight margins, if any margins at all, apart from their own time. They are often supported now by contracts from corporations to remunerate their hard work with support contracts or feature development. As such, these are now beholden to large companies that oversee funding for these projects - good in that people do get to walk away being able to eat. This move in open-source was commented on being one of the first nails in the blending of free and open source software with commercial spaces. Better people have commented on this when Firefox reorganised in 2020 as an open, but explicitly more corporate entity. https://www.boringcactus.com/2020/08/13/post-open-source.html 'No AI' often means 'I still care' and you need to realise that Codeberg now has a global no-LLM/AI policy in place https://codeberg.org/Codeberg/org/pulls/1253 issuecomment-19820434 , and I think it is good. I also think any technical definition on 'vibe-coded' is a category error at the level of an entire site and an attempt to misses the point so much so I assume you like making arguments online for money. Just don't vibe-code there. Assuming a bit here, but I imagine a large section of Codeberg's or other sympathetic users argue that the policy is trying to limit factors that further erode collaboration between anyone making software, of any skill-set, from meeting each other with a common goal of making reliable and interesting work. A 'vibe', if you would so kindly. I want to draw attention to the fact that these spaces in-fact boom not out of some reactionary stance to AI in code as a main ethos, but a new line in the sand on what has been free and open software capitulating to open-source software in spirit and action . AI and it's effects on projects break the social contracts necessary to maintain open-source projects, and these erode out the third spaces, a sociological term for a social surrounding apart from home and workplace https://en.wikipedia.org/wiki/Third place , which I think in GitHub is essentially eroded. I would doubt it exists to help open-source in the long run as much as it is maintains a ground for public source code, the next evolution of this flywheel . These massive depots of code are less places we actually engage and do meaningful work ourselves, or find and contribute to code ourselves. Public Source Code may be where Open Source dies Without caution, I think we will get more of the same - where we can view the source code but can't meaningfully modify it for all. We can't do much apart from creating atomic versions via forks of the same software. It'll probably be worse than the original because we won't have Claude Mega Max or whatever. I fear that students and newcomers are already incentivised to engage outside of open-source projects - it is easier to leverage AI beyond what is useful for their own technical development but engage in the pre-AI trend of using public spaces like GitHub for portfolio projects in of itself, rather than tools end users actually want, or what actually develops nerd credit in this field. Some interesting developments to fight in the viper pit are underway - it may sound contrapositive, but I actually like the idea of Mitchell Hashimoto's Vouch, where people must come recommended before interacting with projects, it is configurable by part, and bad users can be explicitly revoked from this trust system. https://github.com/mitchellh/vouch But what the hell happens when we hit the trolley problem of someone entering into the industry and is only equipped with a text editor, a textbook and gumption? Do we really expect no new blood in this field? AGI in 5 years 1 Computer History Museum Repository Erosion Bouncing off the suggestion that students could leverage AI to spam portfolio boilerplate, we can all agree the amount of repositories on GitHub has exploded. I would assume many already understood before AI that GitHub was becoming a commercial enterprise that promoted the idea of public-facing work rather than any prospect of open-source software, let alone FOSS. GitHub is often assumed by its users and by investors to be a public square to display skill for private job positions more than anything about creating binaries for actual end-users . I circle back to show that this erosion of what open-source is, and that these sites now reward becoming something else. GitHub total repositories, ranked by largest economies from Q1-2020 - Q1-2026. GitHub This shouldn't need to go too deep into the business story of GitHub's rocky habits now, handling this AI-enabled usage - being plagued by uptime https://isgithubcooked.com , the poor tooling for CI https://www.iankduncan.com/engineering/2026-02-05-github-actions-killing-your-team/ , and the purchase of the biggest corpus of code to now train models without explicit consent of the authors https://www.linkedin.com/posts/nrao95 github-just-tried-to-pull-off-the-biggest-activity-7444808251981918209-pmFJ/ . The privacy tab for a given GitHub repository having opt-out dialogues for AI model training. LinkedIn Not too sure that GitHub fully understands how to harness their move to public source code after open-source, either. What is to be done? If it's not obvious, I think that FOSS Free and Open Source Software is dead. It lost the title position of what internet-facing source code was a long time ago to open-source. Open-source projects posited a new contract that instead promised maintainers the world, and instead gave them implicit contracts into unpaid labour for hyperscalers. And now, I think we could be seeing it die but keep the name, but for now. Without being too nihilistic, we may look back on this age as the most egregious example of public facing software licenses being discarded en-masse as a simple prelude to a great loss of skill, novel ideas and eventual entrenchment in the domain of computer science. Not due to the cost of inference, the 'power of AI', but the breath and depth of developers now we have a Grand New Abstraction and how fast we take it on without the public considering it's use cases . We are not considering how the count of repositories could also be that open-source is becoming so atomic that no one wants to, or quickly can't, work together any more. No one wants to be the shitty hand-spun PR. No one wants to review the shitty hand-spun PR. Because it might not be a person anyway. Our ability to understand and own software in any actual practical fashion with others could be in jeopardy. I wish I could tell so many AI boosters that is the actual argument, beyond many other non-digital concerns that actually consider our very fragile planet Pragmatically, sites like GitHub should seriously consider widespread adoption of tools like Vouch or similar vetting. These systems could foster a culture of evaluating people based on skills and interests. GitHub's new star repositories seem to run on this new public source code, atomic forcing paradigm with low-contributor, high-diff teams that use AI agents. These AI-native projects appear to function effectively with financial backing and a dedicated core team. I think this is also horrible for anyone thinking this is a valid space to learn in a way a human mind should. If vetting instead becomes a proxy for "a guy you know" , I think that really is signing away your projects lifespan. It will save your mental health though in these times, probably If you’re just a random lurker browsing these repositories by stars and gawking at them like all other social media - this is not what learning is. Go do something. I’m ordering you to block GitHub and write “I’m an un-creative individual with or without AI” on the chalkboard a hundred times. It's rude but you'll thank me, I promise Not knowing things and failing to be better is eventually fun. Closing thoughts from a n00b; tldr You should still write code and share it. Make projects with the intention that others will use it and it benefits the user. Try to channel your anger constructively or even passionately for/against this post and realise your third space is diminishing either way. If we lose what is left of our third space, I think you’re in trouble. Do not sit here idly reading docs and diffs from a stochastic prediction machine that can replicate the smaller stuff and let it turn you into some mass procrastinator fuelled by either pro- or anti-AI despair . To the skeleton crews working on publicly facing code, don't let your heart turn to stone. Take this on as a novice urging you to forge relationships with those that really do seem to care about how software is made and look on them as people that can carry the torch. Take what wins you can, so this epoch of AI-generated code does not result in open-source software dying, too. And to the Servo crew, sorry about me misreading a package twice on my PR. I did learn so much crap about cargo through that. With time, the budding learners of the world will know what they're doing, hopefully Beyond copyright and the ethical concerns raised by Codeberg, which I cosign, is the growing trend of new and curious contributors adopting a framework like this: "I understand using AI isn’t beneficial for my skill development or mind at the moment and I’m not yet experienced enough with the project. However, I feel compelled to use it to 'not get left behind' and contribute – my body and mind be damned." Other ways we've been here before I reply with this - assume the argument AI and LLMs are the new compiler, assume we will have all that is promised in the mid- and long-term with AI from these companies. Assuming this is the New Abstraction, we still have outstanding questions about LLM tools similar to compiler dilemmas that led to much of the free and open-source software movement. - The most powerful AI models currently are closed-source, not free, and licensed. - Economic forces reward competitors who reverse-engineer, hire soft IP, conduct distillation attacks and other attempts to recover intellectual property. - Given the pace, I doubt we can confidently teach someone 'how to use AI' for a while. - Harnesses are super fascinating I want to know more about the guts of this, genuinely - Do you recall when prompt engineers claimed they could improve weights by talking to a chatbot like an abuser? I wonder if A-0 before C compilers had this cool tech... - There's way too many incentives for those with limited time to invest in something that might significantly hinder their actual skill. - I would also wager soft skills deteriorate at this point too. Chatbots are an antisocial way of creating code, and you genuinely would love being in these hard-ass but welcoming projects like above more than the ass-pat machine. While many people will enjoy the ‘loop’ or ‘plan’, I believe this early stage of implementing a novel technology is laughable and will genuinely degrade the funnel and maintenance of skills. Be honest - what was so wrong with your PRs just being error messages and exceptions for a few months, anyway? ciao the dyld swizzle will be out early next month with a blog