Meet the Developers Rejecting AI  🔗 **Source:** [Meet the Developers Rejecting AI](https://leaddev.com/ai/meet-the-developers-rejecting-ai) > [!note]+ Document Notes > tags: [ai-adoption, developer-experience, open-source, ethics, mental-health, engineering-management] > author: LeadDev (byline not captured) > source: LeadDev > reference: https://leaddev.com/ai/meet-the-developers-rejecting-ai > date created: 2026-09-21 > > > [!info] Generated analysis — not the author's words. Sections below are machine-written > > from the captured source. Only text inside `> [!quote]` blocks is verbatim from the author. > > Do not quote or attribute any other line to the named author. > > # 🏷️ Title: Meet the developers rejecting AI > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) · [📖 Main Content](#-main-content) · [🗒️ Notes](#️-notes) · [📚 References](#-references) · [🗣️ Paraphrase](#️-paraphrase) · [📰 Newsletter](#-newsletter) · [✅ Action Items](#-action-items) · [🧪 Hypotheses](#-hypotheses) · [📊 Methodology](#-methodology) · [❓ Questions](#-questions) · [🔮 Future Research](#-future-research) · [💡 Implications](#-implications) · [🧠 Learning Styles Assessment](#-learning-styles-assessment) · [📊 Accessibility Assessment](#-accessibility-assessment) · [🌍 DEIB Assessment](#-deib-assessment) · [✅ Logical Validity](#-logical-validity) · [🛡️ DARVO Analysis](#️-darvo-analysis) · [🎯 Conclusion](#-conclusion) · [🧱 Refactor Appendix](#-refactor-appendix) · [🧰 Template Library](#-template-library) ## 🔗 References (in-page) - Source: https://leaddev.com/ai/meet-the-developers-rejecting-ai (LeadDev, published 2026-09-21; the author byline was not captured in the Reader copy) - Brett Chalupa's video: https://www.youtube.com/watch?v=2ZU3j4GQ4K8 - arXiv study on the psychological costs of AI adoption: https://arxiv.org/pdf/2609.03456v1 - Handcrafted Code directory: http://handcraftedcode.org ## 📝 Summary **What happened.** LeadDev profiles four developers who have stopped using AI coding tools, or refuse to start, even at a cost to their careers. The hook is Brett Chalupa, a developer with 20 years of experience, whose YouTube video announcing he is done with AI passed 900,000 views and 10,000 comments. The other three are Anthony Manning-Franklin, who left a CTO role after the founder said he wanted no engineers coding by hand; Pieter Groenendijk, a new computer engineering graduate in the Netherlands; and Jared White, a freelancer who lost a long-time client and a full-time offer over his "no-genAI policy." **Why it matters.** The piece names four reasons for refusing, and every developer it quotes gave all of them: loss of craft, eroded skills, ethics (environmental cost, training on scraped code without consent, erosion of social trust), and mental health. It anchors the mental-health claim in an arXiv study of one software services company a year after its AI rollout, which found "craft identity disruption, meaning and satisfaction erosion, cognitive and workload intensification, and uncertainty distress." **Stakeholders.** Engineering managers who mandate AI adoption; developers deciding whether to comply; hiring teams screening for AI fluency; open-source maintainers whose code sits in training sets; developer-community builders. **Risks.** Four hand-picked people with the same view can't show how common the view is. The article says as much ("Chalupa is not alone") without giving a number. The 900,000 views measure attention, not agreement. **Bottom line.** The refusal is real, it's costing the people who make it jobs and clients, and at least one study backs the emotional cost it describes. How widespread it is, the article can't say. ## 📖 Main Content The article opens on Chalupa telling the internet, not his boss, that he's finished with AI. He'd used it in earnest for about 18 months after his employer mandated it, liked it at first, and then found the work left him depressed and without a sense of purpose. He concedes he shipped more code with AI and argues the tradeoff is still wrong: developers lose track of how the codebase changes, and he feels no ownership of code a model wrote, which he says is bad for business. His defense of the craft is the article's emotional center — the "in the zone" rhythm of micro-decisions, against which reviewing model output is "the opposite." Manning-Franklin supplies the technical complaint. He gave AI a fair trial — he built neural networks from scratch, with no libraries, to understand them — and found it produced too much code. He cites refactoring 1,000-line files to about 100 lines and 3,000-line solutions to 250, and describes incident response inside AI-written code that "no one wrote and no one actually understands." He also can't square AI use with his environmental values. Groenendijk and White extend the ethics case. Groenendijk notes that about 40% of Wikipedia's AI article covers ethical issues and says he doesn't want to outsource his thinking. White tracks the tools closely but won't use them, citing data-center environmental damage, erosion of social trust, and non-consensual training — he has found his own open-source contributions in training datasets. The career-cost section is the most concrete. Chalupa works at a small company that doesn't monitor token usage, so abstaining hasn't hurt him yet, but he expects anyone hiring to require AI. White has already lost a client that went "all-in on agentic coding" and a job that otherwise fit, which he calls a "dictatorial approach" to how developers structure their work and a shift he hasn't seen before. He has rebranded his developer Discord as the "Human Web Collective" and launched handcraftedcode.org, a directory for developers who code by hand. The closing section is about grief. Manning-Franklin no longer enjoys open-source work because he doubts a real person will use what he builds. Chalupa had planned a 30-year career and now wonders if he wasted his life on a skill "a stupid chatbot can do." Groenendijk grieves a career he never got to start, acknowledges that universal healthcare and a financial safety net make his stance possible, and says quitting AI is when he started "getting my soul back." ### 🗒️ Notes - **Key takeaways box:** the viral video passed 900k views; the reasons are craft, ethics, and mental health; some people have already paid with lost clients, missed jobs, and one exit from the industry. - **Chalupa:** 20 years coding; about 18 months of mandated AI use; left because of the mental-health cost; accepts he may be fired or unable to find work — "so be it." His employer doesn't monitor tokens, and colleagues have seen the video. - **Manning-Franklin:** former CTO; left after the founder barred hand-coding; refactoring ratios of roughly 10:1 and 12:1; cites skill erosion and environmental conflict; has "effectively left the industry." - **Groenendijk:** 2026 computer engineering graduate in the Netherlands; the Wikipedia 40% figure; disliked even early Copilot autocomplete; names his own privilege; building his own project and willing to change careers. - **White:** freelancer; researches the tools but won't use them; found his open-source code in training sets; lost a long-time client and a full-time role; founded the Human Web Collective and handcraftedcode.org. - **Evidence cited:** the arXiv study on psychological costs of AI adoption in software engineering (one large services company, one year after rollout). - **Framing:** the article calls the split a "bifurcation," White's word, and says the developers don't expect to reverse the trend. ## 📚 References Primary sources first, as linked in the article's markup: - Brett Chalupa, "I'm done with AI" video — https://www.youtube.com/watch?v=2ZU3j4GQ4K8 - Study on the psychological costs of AI adoption in software engineering — https://arxiv.org/pdf/2609.03456v1 - handcraftedcode.org — http://handcraftedcode.org LeadDev's own related coverage, linked in the body: - Trust in AI coding tools is plummeting — https://leaddev.com/technical-direction/trust-in-ai-coding-tools-is-plummeting - Engineers grieve a job that no longer exists — https://leaddev.com/career-development/engineers-grieve-a-job-that-no-longer-exists - AI adoption has to be driven from the top — https://leaddev.com/ai/ai-adoption-has-to-be-driven-from-the-top - Engineering skills are being rewritten by AI — https://leaddev.com/ai/engineering-skills-are-being-rewritten-by-ai - Devs fear the AI race is throwing ethics to the wind (linked through a google.com redirect) — https://leaddev.com/ai/devs-fear-the-ai-race-is-throwing-ethics-to-the- Not linked: the Wikipedia 40% figure, the Human Web Collective Discord, and the details of the CTO departure are all reported without a source. ## 🗣️ Paraphrase 1. A senior developer announced on YouTube that he's done with AI, and the video hit a nerve with hundreds of thousands of viewers. 2. He's not alone: other developers are refusing AI even if it ends their careers, because for them the craft was the point. 3. One former CTO walked away after being told nobody should code by hand, and says AI code is bloated, hard to debug, and bad for skills and the environment. 4. A new graduate won't use it because of the ethics and because he wants to do his own thinking. 5. A freelancer won't use it because of environmental harm, social trust, and training on code taken without consent — including his own. 6. After 18 months of mandated use, the YouTuber quit because it was hurting his mental health, and a study of one company after its AI rollout found the same kinds of harm. 7. He admits AI made him faster but thinks developers lose their grip on the codebase and stop caring about code they didn't write. 8. Refusing has already cost the freelancer a client and a job, and he's started a community and a directory for hand-coders. 9. All of them describe grief — for a career, a craft, or a future that now looks different. 10. The graduate knows his safety net makes refusal possible, and says quitting AI gave him his soul back. ## 📰 Newsletter **Subject: The developers opting out** LeadDev talked to four developers who've decided they're done with AI coding tools. The one you've probably seen is [Brett Chalupa's video](https://www.youtube.com/watch?v=2ZU3j4GQ4K8): 20 years of coding, 18 months of mandated AI use, and a public "so be it" if it costs him his job. 900,000 views later, it clearly landed. What makes the piece worth your time is that the refusers don't sound like people who never tried the tools. One former CTO built neural networks from scratch to understand them, then spent his days cutting 3,000-line AI solutions down to 250 lines. His description of an incident inside code "no one wrote and no one actually understands" should make every engineering manager a little uneasy. The career cost is already real. Jared White lost a long-time client and a job offer he was a great fit for, both over AI policy, and has started [handcraftedcode.org](http://handcraftedcode.org) as a directory for people who still code by hand. The WordPress angle writes itself. Our plugin directory runs on contributors who do this for the love of the craft. If a meaningful share of them feel the way these four do, "AI-first" mandates in agencies and product shops cost more than productivity metrics show. There's also [an arXiv study](https://arxiv.org/pdf/2609.03456v1) backing up the mental-health part, which is more than most AI-adoption pitches bring. One caution: four people who agree don't make a trend. But four people willing to give up jobs over it is a signal worth tracking. ## ✅ Action Items **DevRel** - When you promote AI tooling, name the craft and mental-health concerns directly instead of treating resistance as a knowledge gap. The four developers here understood the tools and rejected them anyway. - Think about an "AI optional" track for workshops and content, so developers who code by hand aren't excluded from the community. **OSPO** - The training-data consent complaint (White found his own open-source contributions in datasets) belongs in license and contribution-policy conversations. Track how projects are responding, including opt-out signals. **Community** - The Human Web Collective and handcraftedcode.org are a new community segment. Watch whether hand-coding identity becomes a real affinity group that events and meetups should make room for. **WordPress subteams** - *Core AI* — the strongest link. Contributor sentiment on mandated AI deserves an explicit place in AI-feature discussions; the team shouldn't assume everyone welcomes the tools. - *Core* / *Core Performance* — Manning-Franklin's bloat complaint (10:1 refactors) suggests reviewing AI-assisted patches for size and clarity, not just correctness. - *Community* — contributor retention: grief and lost joy in open source ("I don't even enjoy contributing to open source anymore") are warning signs for volunteer programs. - *Training* — Learn content on AI should include when *not* to use it, and should be honest about the tradeoffs. - *Plugins* — the directory's review process could track the share of AI-generated submissions and whether they're harder to review. - *Core Program Team*, *Mobile*, *Playground*, *Marketing*, *Support*, *Design*, *Documentation*, *Photos*, *TV*, *Polyglots*, *Accessibility*, *Meta*, *CLI*, *Hosting*, *Tide*, *Openverse*, *Themes* — no direct action; the concerns apply wherever contributors are asked to adopt AI. **The WPOCC** - A governance question worth raising: should community-run projects set contributor-facing AI policies that protect people who choose not to use it? **FAIR** - Training-data consent connects directly to provenance: a distribution network that records origin and license makes "was my code used without consent?" answerable. ### → Things 3 - [ ] Read the arXiv psychological-costs study (2609.03456) and note its sample and method - [ ] Look at handcraftedcode.org as a possible community segment for DevRel ## 🧪 Hypotheses 1. **Mandates drive refusal more than the tools do.** Developers who were required to use AI will report more distress and more refusal than those who adopted voluntarily, controlling for exposure. Testable by surveying teams with and without mandates. 2. **AI-generated code is larger.** For the same task, AI-assisted code will be several times larger than hand-written code before refactoring. Testable by measuring lines of code on matched tasks. 3. **The refusal is concentrated in experienced developers and new graduates.** Mid-career developers under financial pressure will refuse less often. Testable against the age and tenure distribution in developer surveys. 4. **Refusers lose work.** Developers with public no-AI stances will see fewer hiring callbacks in AI-forward markets. Testable by a matched-résumé audit study. 5. **Craft loss predicts open-source withdrawal.** Contributors who report lost enjoyment will reduce contribution volume within a year. Testable with contribution data plus a survey. ## 📊 Methodology **Design:** feature journalism based on four interviews and one viral video. There's no sample frame; the subjects were chosen because they refuse AI. **Evidence:** - Four first-person interviews — rich and specific, with a built-in selection bias. - View and comment counts for the video — measure reach, not prevalence. - One arXiv preprint, linked — a single-company case study one year after rollout. The article doesn't give its sample size, and preprints aren't peer reviewed. - The Wikipedia 40% figure — an unlinked, unverifiable aside, and a page-length proportion says little about the ethics. - The refactoring ratios — self-reported anecdotes, not measured. **What's missing:** a counter-voice (no developer who adopted AI happily is quoted), prevalence data, and any employer perspective on the mandates. **How much weight it can bear:** strong as a qualitative portrait of one camp; weak as evidence of how large that camp is. ## ❓ Questions - What share of developers would refuse AI if it had no career cost? The article can't tell. - Is the problem AI itself or the mandate? Chalupa liked the tools until they were required and measured. - How does the arXiv company's experience compare with teams that adopted AI voluntarily? - If hand-coding becomes a niche, who pays for it — and does it become a premium or a penalty? - What does "no engineers coding by hand" mean in practice at the company that pushed Manning-Franklin out? - How should an open-source project treat contributors who won't touch AI tools? ## 🔮 Future Research - A prevalence survey of AI refusal across seniority, geography, and employment type. - Longitudinal follow-up on the arXiv study's company: do the psychological effects fade or persist? - A controlled comparison of code size, defect rate, and time-to-understand for AI-written and hand-written code. - Hiring audits measuring the career cost of a public no-AI stance. - Qualitative study of the Human Web Collective as an emerging professional identity. ## 💡 Implications **Practice.** Managers who mandate AI should expect some strong engineers to leave and should weigh that against the productivity gain. Letting people opt out, and making code size and comprehensibility review criteria, both reduce the cost. **Policy.** Mandated tooling that affects mental health is an HR and duty-of-care issue, not only a productivity decision. The training-data consent complaint feeds ongoing copyright and licensing debates. **Security.** Manning-Franklin's point bears directly on incident response: code no one wrote and no one understands is slower to debug under pressure. That's operational risk. **Education.** New graduates like Groenendijk are entering a field that changed while they studied. Curricula need to teach judgment about when to use AI, not just how. **Funding.** Open-source sustainability programs should notice contributors losing motivation because they doubt real people will use what they build. ## 🧠 Learning Styles Assessment - **Visual** — weak: a text feature with a hero image; no charts, no data visualization. - **Aural** — indirect: the story originates in a video, which audio-first learners can watch. - **Verbal** — strong: quote-driven narrative with vivid lines ("damp spaghetti," "getting my soul back"). - **Kinesthetic** — minimal: no exercises. The nearest action is joining the directory or the Discord. - **Logical** — moderate: four reasons are named and supported with examples, but prevalence isn't quantified. - **Social** — strong: built entirely from personal stories and community responses. - **Solitary** — strong: invites reflection on your own relationship to the craft. - **Naturalistic** — not applicable to this source; no natural-world or classification content. The environmental concern is mentioned but not explored. **Reading level:** about grade 10. Short paragraphs, plain vocabulary, a few technical terms (LLM, token usage, agentic coding). ## 📊 Accessibility Assessment Mapped to WCAG 2.2 AA, from the Reader capture; visual checks need the live page. 1. **Text alternatives (1.1.1)** — the hero image alt text isn't in the capture; an event ad image ("NYC 27 Pre sale ticket block image") appears as raw alt text mid-article, which interrupts screen-reader flow. *Partial.* 2. **Structure (1.3.1)** — clear section headings ("Why these devs are rejecting AI," "Loss and grief") and a key-takeaways box. *Pass.* 3. **Link purpose (2.4.4)** — several links use vague anchor text ("AI coding" twice, pointing to different articles; "using AI"). Duplicate anchor text for different destinations fails the purpose-in-context test. *Fail.* 4. **Contrast (1.4.3)** — not assessable from the capture. 5. **Readable language (3.1.5)** — plain language throughout. *Pass.* 6. **Consistent navigation (3.2.3)** — site chrome is outside the capture; not assessable. 7. **Media (1.2.x)** — the linked YouTube video's captions aren't evaluated here. **Top fix:** make link text unique and descriptive, and give the inline ad image an empty alt so screen readers skip it. ## 🌍 DEIB Assessment **Cultural breadth.** Mostly US-framed, with one Dutch voice. Groenendijk's point that universal healthcare makes refusal possible is the article's best moment of breadth, and it shows that the ability to refuse depends on where you live. **Inclusive language.** Respectful. No derision of AI users, though the piece doesn't give their view. **Neurodiversity.** Not discussed, and relevant: the "in the zone" flow Chalupa describes, and the cognitive load of constant review, affect neurodivergent developers in particular ways the piece doesn't explore. **Representation.** All four developers appear to be men. Given the companion LeadDev piece on AI widening the gender gap (women are penalized more for *using* AI), refusal may carry different costs by gender, and the article can't see that. **Psychological safety.** The piece treats grief as legitimate, which is humane. It also shows that psychological safety is missing at mandate-driven companies: White calls it "dictatorial," and Manning-Franklin was pushed out. ## ✅ Logical Validity ### Strong reasoning - The article states its evidence's limits in part ("not everyone who feels this way is able to risk their livelihood"). - The arXiv study is linked, so the mental-health claim can be checked. - Chalupa's concession that he shipped more code with AI makes his argument more credible, not less. ### Fallacies 1. **Hasty generalization / selection bias.** *Where:* "Chalupa merely said out loud what many developers are feeling" and "Chalupa is not alone." *Why it fails:* four people selected for sharing a view can't establish "many." *Fix:* cite a prevalence figure or frame it as "some developers." 2. **Popularity as validation.** *Where:* the 900k views and 10k comments. *Why it fails:* views measure attention; comments are self-selected. *Fix:* present them as reach, not agreement. 3. **Weak proxy.** *Where:* "40% of the Wikipedia page for AI is dedicated to ethical issues." *Why it fails:* page proportion reflects editorial structure, not the size of the ethical problems. *Fix:* cite a substantive ethics source. 4. **Anecdote standing in for data.** *Where:* the refactoring ratios. *Why it fails:* self-reported, no task comparison. *Fix:* frame as experience, not measurement. ### Tone Sympathetic and restrained. The article lets its subjects speak and mostly avoids arguing their case for them. There's no opposing voice, which makes it one-sided in effect even though the tone is fair. ## 🛡️ DARVO Analysis **Does the source perform DARVO?** No. It doesn't deny, attack, or reverse roles; it reports the developers' experiences. **Is DARVO described in the source?** Partly, as employer behavior. The founder who barred hand-coding and the interviewer's "if you don't fit… you're not welcome" framing cast the refusing developer as the problem to be managed. That echoes the **Reverse** move (the person raising a concern becomes the one who doesn't fit), though the article doesn't show an employer denying anything or attacking anyone, so it's structural pressure rather than DARVO. **Assessment:** low. No DARVO pattern is demonstrated; there's a hint of role reversal in how mandates treat dissent. ## 🎯 Conclusion **Bottom line.** A careful, humane profile of developers opting out of AI coding, grounded in specific costs — a lost client, a lost job offer, a CTO who walked away — and backed by one linked study on the psychological toll. It's reliable as testimony and silent on scale: four refusers and a viral video can't tell you how many developers feel this way. **Next steps.** 1. Read the arXiv study to see its sample and method before citing the mental-health finding. 2. Pair this with LeadDev's gender-gap piece: the two together show that both using and refusing AI carry career penalties that fall unevenly. 3. Watch handcraftedcode.org and the Human Web Collective as signals of a lasting split. ## 🧱 Refactor Appendix ### Literature Split Pad **L1 — Refusal at a career cost.** Four developers stop or refuse AI coding despite lost clients, jobs, or a CTO role. Source: LeadDev, 2026-09-21. **L2 — Four reasons, cited by all.** Craft, skill erosion, ethics (environment, consent, social trust), and mental health. **L3 — The mandate as trigger.** Chalupa liked AI until his employer required it; Manning-Franklin left over a no-hand-coding rule; White lost work over his own policy. **L4 — Bloat and comprehension.** Refactoring ratios of 10:1 and 12:1; incidents in code "no one wrote and no one actually understands." **L5 — Refusal is a privilege.** Groenendijk names healthcare and a safety net as what make refusal possible. #### Atomic Split Pad **A1 — A mandate turns a tool into an identity question.** Chalupa liked AI until it was required and measured. Coercion changes how people experience a tool. Links: [[Voluntary vs mandated adoption]]. **A2 — Code nobody wrote is code nobody can debug.** Comprehension, not generation speed, sets the cost of an incident. Links: [[Operational risk of generated code]]. **A3 — The ability to refuse depends on a safety net.** Opting out of a dominant tool depends on healthcare, savings, and local labor law. Links: [[Privilege in professional dissent]]. **A4 — Attention metrics aren't prevalence.** Views and comments measure reach, not how common a view is. Links: [[Popularity is not evidence]]. **A5 — Both using and refusing AI can carry a career penalty.** Pairs with the gender-gap piece. Links: [[AI competence penalty]]. #### Fleeting Queue - Author byline not captured; confirm on the live page before citing. - Check the arXiv study's sample size and whether it's since been peer reviewed. - The "devs fear the AI race" link goes through a google.com redirect — note it as a markup defect. - Is there a counterpart piece profiling developers who embraced AI? Balance. <!-- deep-read:sections end --> 📄 **Full text:** [[Full Document Contents/Articles/Meet the Developers Rejecting AI.md|Meet the Developers Rejecting AI by LeadDev]] ## Highlights - “I have, far too many times, refactored a 1,000-line file down to almost 100 lines, or a 3,000-line solution down to 250 lines. Digging through 3,000 lines of damp spaghetti during an incident is depressing enough, but it’s much worse when it’s brand new code that no one wrote and no one actually understands,” he says. ([View Highlight](https://read.readwise.io/read/01m31yqa58n091143hvhmhz3mf)) - Note: Atom A2 — code nobody wrote is code nobody can debug. Comprehension, not generation speed, sets the cost of an incident. The refactoring ratios are self-reported, not measured. - Tags: developer-experience, ai-adoption, atom/atomic/h1 - “I have been grieving a little bit. Sometimes I’ve just been very down thinking, what am I gonna do?” he says. “Eventually I overcame that, mostly because I decided I should stop using it. When I made the choice, that’s when I kind of started getting my soul back.” ([View Highlight](https://read.readwise.io/read/01m31yqcpn10x3tnhec5ag2esj)) - Note: Atom A3 — the ability to refuse depends on a safety net. Groenendijk names universal healthcare and financial security as what make this choice possible. - Tags: atom/atomic/h2, mental-health, ai-adoption