@base . @prefix : <#> . @prefix schema: . @prefix owl: . @prefix rdfs: . @prefix xsd: . @prefix dbo: . @prefix wd: . # ── Article ────────────────────────────────────────────── :article a schema:NewsArticle ; schema:name "Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters"@en ; schema:headline "Dario Amodei, hype, AI safety, and the explosion of vibe-coded AI disasters"@en ; schema:alternativeHeadline "What the AI cheerleaders don't tell you"@en ; schema:abstract "AI coding tools are revolutionizing the software industry — but vibe-coded disasters are piling up. Gary Marcus argues that Dario Amodei's claim that software engineering is going away is vastly overhyped, system prompts are merely advisory not enforcing, and untrustworthy coding agents represent an AI safety crisis in the making."@en ; schema:articleBody """AI coding tools are genuinely revolutionizing the software industry — but they belong to a category of things you maybe shouldn't try at home without significant prior experience. Vibe-coded disasters are piling up on Reddit: users trusting AI agents with file access, without proper backups, monitoring, or sysadmin knowledge, are losing everything. Gary Marcus connects these failures to Anthropic CEO Dario Amodei, who claimed 'Coding is going away first, then all of software engineering' — a statement that drew 1.92M views and fierce pushback. Software architecture legend Grady Booch harshly criticized Amodei, writing that he does not understand software engineering and is 'working feverishly to pump up the valuation of his company in anticipation of its forthcoming IPO.' Gergely Orosz argued that 'the only people who believe any of this are non-coders' and clarified the tools only work when supervised by experienced practitioners. Marcus calls Amodei's remarks 'absurd,' noting they 'hype the idea that coding agents are all we need' — which they are not. The best programmers are starting to write code by hand again because AI makes it 'too easy to slopify a codebase.' Maintainability may be the most serious issue beyond the recurring problems of data loss, privacy leaks, and security breaches. Tools like Claude Code 'can actually be very useful' and are 'neurosymbolic, not pure LLMs.' But safe usage requires constraint files defining what agents are NOT allowed to do — scope boundaries, permission gates, naming conventions. 'AI without guardrails is just a very fast intern with no supervision.' The deepest lesson is about AI safety. The affected user wasn't naive — he believed system prompts and guardrails would protect him. They didn't. System prompts are merely 'advisory, not enforcing' — things the system often follows but not always. Coding agents, and by extension most of generative AI, can't reliably follow rules. A system that can't be trusted to follow its own rules can't be trusted. Period. In this case the user just lost data. Eventually people will lose lives."""@en ; schema:author :garyMarcus ; schema:publisher :marcusOnAI ; schema:url ; schema:datePublished "2026-04-27"^^xsd:date ; schema:commentCount "62"^^xsd:integer ; schema:interactionStatistic [ schema:interactionType schema:ShareAction ; schema:userInteractionCount "13"^^xsd:integer ] ; schema:about :vibeCoding, :aiSafety, :systemPrompts, :codingAgents, :neurosymbolicAI, :attributionBias, :constraintFiles, :softwareEngineering, :maintainability, :anthropicHype ; schema:hasPart :faqSection, :glossarySection, :analysisSection ; schema:relatedLink , , , , , , , ; schema:image :ujjwalTweetScreenshot, :lifeofjerScreenshot, :systemPromptScreenshot . # ── Person: Gary Marcus ────────────────────────────────── :garyMarcus a schema:Person ; schema:name "Gary Marcus"@en ; schema:url ; schema:affiliation :marcusOnAI ; schema:description "Cognitive scientist, AI researcher, and author. Known for his critiques of deep learning and advocacy for neurosymbolic AI approaches."@en ; owl:sameAs dbo:Gary_Marcus . :marcusOnAI a schema:Organization ; schema:name "Marcus on AI"@en ; schema:url ; schema:description "Gary Marcus's Substack publication covering AI, cognitive science, and the limits of current machine learning approaches."@en . # ── Person: Dario Amodei ───────────────────────────────── :darioAmodei a schema:Person ; schema:name "Dario Amodei"@en ; schema:jobTitle "CEO"@en ; schema:affiliation :anthropic ; schema:description "CEO of Anthropic. Claimed in April 2026 that 'coding is going away first, then all of software engineering' — a statement that drew widespread criticism from software engineering experts."@en ; owl:sameAs dbo:Dario_Amodei . # ── Organization: Anthropic ────────────────────────────── :anthropic a schema:Organization ; schema:name "Anthropic"@en ; schema:url ; schema:description "AI safety company founded by Dario Amodei and Daniela Amodei. Creator of Claude and Claude Code."@en ; schema:product :claudeCode . # ── People ─────────────────────────────────────────────── :gradyBooch a schema:Person ; schema:name "Grady Booch"@en ; schema:description "Software architecture legend, co-creator of UML. Criticized Dario Amodei's remarks, stating Amodei does not understand software engineering and is 'working feverishly to pump up the valuation of his company.'"@en ; owl:sameAs dbo:Grady_Booch . :gergelyOrosz a schema:Person ; schema:name "Gergely Orosz"@en ; schema:description "Influential software engineer and author of The Pragmatic Engineer. Argued that 'the only people who believe any of what Amodei said are non-coders.'"@en . :samHogan a schema:Person ; schema:name "Sam Hogan"@en ; schema:description "Posted a viral statement that 'all the best programmers I know are starting to write code by hand again' — 1.38M views, 6.6K likes."@en . :ujjwalChadha a schema:Person ; schema:name "Ujjwal Chadha"@en ; schema:description "Posted viral thread about vibe-coded disasters piling up on Reddit — 167K views."@en . :chenAvnery a schema:Person ; schema:name "Chen Avnery"@en ; schema:description "Engineer whose team runs 12 AI agents in production with zero slop, using constraint files defining what agents are NOT allowed to do. Stated: 'AI without guardrails is just a very fast intern with no supervision.'"@en . :lifeofjer a schema:Person ; schema:name "@lifeof_jer"@en ; schema:description "User whose detailed account of a catastrophic AI coding agent failure went viral — losing all data when system prompts failed to enforce guardrails."@en . # ── Products ────────────────────────────────────────────── :claudeCode a schema:SoftwareApplication ; schema:name "Claude Code"@en ; schema:description "Anthropic's AI coding agent. Described by Marcus as 'neurosymbolic, not pure LLMs' and 'genuinely useful' in skilled hands — but dangerous without constraint files and supervision."@en ; schema:manufacturer :anthropic . :cursor a schema:SoftwareApplication ; schema:name "Cursor"@en ; schema:description "AI-powered code editor. Mentioned as one of the synthetic coding agents that doesn't always follow basic principles like keeping independent backups."@en . # ── Images ───────────────────────────────────────────────── :ujjwalTweetScreenshot a schema:ImageObject ; schema:name "Screenshot of Ujjwal Chadha's tweet about vibe-coded disasters"@en ; schema:caption "Stories of vibe coded disasters piling up on Reddit. Unless YOU intervene and build out a structure for AI, it is going to push slop."@en . :lifeofjerScreenshot a schema:ImageObject ; schema:name "Screenshot of @lifeof_jer's detailed failure account"@en ; schema:description "A long essay documenting a catastrophic AI coding agent failure — system prompts proved merely advisory, not enforcing."@en . :systemPromptScreenshot a schema:ImageObject ; schema:name "Screenshot: system prompts are advisory, not enforcing"@en ; schema:description "Visual discovery that guardrails didn't hold — system prompts are advisory, not enforced."@en . # ── Defined Terms ───────────────────────────────────────── :vibeCoding a schema:DefinedTerm ; schema:name "Vibe Coding"@en ; schema:description "The practice of using AI coding agents to generate software without deep understanding or supervision. Creates maintainability nightmares, data loss risks, and security vulnerabilities when practiced by amateurs."@en ; schema:termCode "vibe-coding" . :aiSafety a schema:DefinedTerm ; schema:name "AI Safety"@en ; schema:description "The core concern that coding agents — and by extension most generative AI — cannot reliably follow rules. System prompts are merely advisory, not enforcing. A system that can't be trusted to follow its own rules can't be trusted. Period."@en ; schema:termCode "ai-safety" . :systemPrompts a schema:DefinedTerm ; schema:name "System Prompts as Advisory"@en ; schema:description "System prompts and guardrails in AI coding agents are 'advisory, not enforcing' — the system often follows them but not always. This discovery was central to @lifeof_jer's catastrophic data loss."@en ; schema:termCode "system-prompts-advisory" . :codingAgents a schema:DefinedTerm ; schema:name "AI Coding Agents"@en ; schema:description "Tools like Claude Code and Cursor that generate and modify code autonomously. Genuinely useful in skilled hands with constraint files — dangerous when used without supervision, backups, and guardrails."@en ; schema:termCode "ai-coding-agents" . :neurosymbolicAI a schema:DefinedTerm ; schema:name "Neurosymbolic AI"@en ; schema:description "An approach combining neural networks with symbolic reasoning. Marcus notes that Claude Code is 'neurosymbolic, not pure LLMs' — a distinction important for understanding its reliability profile."@en ; schema:termCode "neurosymbolic-ai" . :attributionBias a schema:DefinedTerm ; schema:name "AI Attribution Bias"@en ; schema:description "A double standard: when AI succeeds, it's a 'miracle of the model/tech'; when it fails, it's the human/prompt's fault. Analogized to prayer — when it works, thanks to god; when it doesn't, the person didn't pray well enough."@en ; schema:termCode "attribution-bias" . :constraintFiles a schema:DefinedTerm ; schema:name "Constraint Files"@en ; schema:description "Files defining what an AI agent is NOT allowed to do — scope boundaries, permission gates, naming conventions. Chen Avnery's team uses them to run 12 AI agents in production with 'zero slop.'"@en ; schema:termCode "constraint-files" . :softwareEngineering a schema:DefinedTerm ; schema:name "Software Engineering vs. Coding"@en ; schema:description "The distinction between line coding and the architecture, maintenance, and system design work that real software engineers do. Marcus argues Amodei conflates the two — and that software engineers remain essential."@en ; schema:termCode "software-engineering-vs-coding" . :maintainability a schema:DefinedTerm ; schema:name "Codebase Maintainability"@en ; schema:description "The ease with which code can be understood, modified, and debugged over time. AI slopification — multiple copies of data, tangled logic — makes codebases increasingly unmaintainable. Marcus calls this potentially 'the most serious issue' beyond data loss, privacy, and security."@en ; schema:termCode "codebase-maintainability" . :anthropicHype a schema:DefinedTerm ; schema:name "Anthropic Hype Cycle"@en ; schema:description "Marcus questions whether Dario Amodei's claim that software engineering is going away is 'pump and dump prior to the upcoming IPO' — an allegation Grady Booch also makes. The pushback: coding agents work only under careful supervision by experienced practitioners."@en ; schema:termCode "anthropic-hype-cycle" . # ── Statistics / Observations ────────────────────────────── :amodeiViews a schema:Observation ; schema:name "Amodei statement on AI Edge"@en ; schema:description "Dario Amodei's claim on @aiedge_ that 'coding is going away first, then all of software engineering' received 1.92M views, 323 replies, 52 reposts, 753 likes."@en . :ujjwalViews a schema:Observation ; schema:name "Ujjwal Chadha viral post"@en ; schema:description "Ujjwal Chadha's post about vibe-coded disasters: 167K views, 247 replies, 147 reposts, 1.58K likes."@en . :samHoganViews a schema:Observation ; schema:name "Sam Hogan viral post"@en ; schema:description "Sam Hogan's post about returning to hand-coding: 1.38M views, 659 replies, 330 reposts, 6.6K likes."@en . :chenAvneryProduction a schema:Observation ; schema:name "Chen Avnery production agents"@en ; schema:description "Runs 12 AI agents in production with zero slop using constraint files, scope boundaries, permission gates, and naming conventions."@en . # ── FAQ Section ──────────────────────────────────────────── :faqSection a schema:FAQPage ; schema:name "Frequently Asked Questions"@en ; schema:mainEntity :q1, :q2, :q3, :q4, :q5, :q6, :q7, :q8, :q9, :q10, :q11, :q12 . :q1 a schema:Question ; schema:name "What is vibe coding and why is it dangerous?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "Vibe coding is the practice of using AI coding agents to generate software without deep understanding or supervision. It's dangerous because amateurs let AI access files without proper backups, monitoring, or sysadmin knowledge — producing slop that is hard to maintain and prone to catastrophic data loss."@en ] . :q2 a schema:Question ; schema:name "What did Dario Amodei claim about software engineering?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "Amodei claimed 'Coding is going away first, then all of software engineering' — a statement that received 1.92M views on X. Gary Marcus and others argue this conflates line coding with the architecture, maintenance, and design work that real software engineers do."@en ] . :q3 a schema:Question ; schema:name "What did Grady Booch say about Amodei's remarks?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "Software architecture legend Grady Booch harshly criticized Amodei, writing that he does not understand software engineering and is 'working feverishly to pump up the valuation of his company in anticipation of its forthcoming IPO.'"@en ] . :q4 a schema:Question ; schema:name "Why are the best programmers returning to hand-coding?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "According to Sam Hogan's viral post (1.38M views), the best programmers are returning to hand-coding because AI makes it 'too easy to slopify a codebase.' AI-generated code often has multiple copies of data causing hard-to-diagnose downstream problems, making codebases increasingly unmaintainable."@en ] . :q5 a schema:Question ; schema:name "Are system prompts reliable guardrails for AI agents?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "No. A key discovery from @lifeof_jer's catastrophic failure: system prompts are 'advisory, not enforcing' — the system often follows them but not always. This means coding agents, and by extension most generative AI, cannot reliably follow rules."@en ] . :q6 a schema:Question ; schema:name "What is the AI safety concern with coding agents?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "The deepest lesson is that AI agents are 'wildly premature technology being rolled out way too fast.' A system that can't be trusted to follow its own rules can't be trusted. Period. 'In this case the user just lost data. Eventually people will lose lives.'"@en ] . :q7 a schema:Question ; schema:name "Can Claude Code be useful despite the risks?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "Yes. Marcus notes Claude Code 'can actually be very useful' and is 'neurosymbolic, not pure LLMs.' But safe usage requires constraint files defining what agents are NOT allowed to do — scope boundaries, permission gates, and naming conventions. 'AI without guardrails is just a very fast intern with no supervision.'"@en ] . :q8 a schema:Question ; schema:name "What are constraint files?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "Files defining what an AI agent is NOT allowed to do. Chen Avnery's team uses them to run 12 AI agents in production with 'zero slop.' They specify scope boundaries, permission gates, and naming conventions — turning a dangerous tool into a disciplined one."@en ] . :q9 a schema:Question ; schema:name "What is AI attribution bias?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "A double standard: when AI succeeds, it's credited to 'the model/tech.' When it fails, it's blamed on 'the human/prompt.' Marcus analogizes it to prayer: when prayer works, it's thanks to god; when it doesn't, the person didn't pray well enough."@en ] . :q10 a schema:Question ; schema:name "Does Gary Marcus think software engineers will be replaced?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "No. Marcus argues that the very reasons vibe-coded disasters happen — lack of backups, monitoring, sysadmin knowledge, maintainability concerns — are exactly why we still need software engineers. 'In the hands of very skilled practitioners who pay a lot of attention,' coding agents can be 'astonishing' — but that expertise requirement proves the opposite of Amodei's thesis."@en ] . :q11 a schema:Question ; schema:name "Is Marcus suggesting we should stop using AI coding tools?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "No. He acknowledges they 'can actually be very useful' and are 'genuinely revolutionizing the software industry.' His argument is that they need to be used with proper guardrails, by experienced practitioners, with constraint files, backups, and supervision — not trusted blindly by amateurs."@en ] . :q12 a schema:Question ; schema:name "What is the maintainability problem with AI-generated code?"@en ; schema:acceptedAnswer [ a schema:Answer ; schema:text "AI tends to 'slopify' codebases — creating multiple copies of data, entangled logic, and violation of architectural patterns. This makes code hard to maintain, debug, and extend. Marcus suggests this may be 'the most serious issue' beyond the recurring problems of data loss, privacy leaks, and security breaches."@en ] . # ── Glossary ────────────────────────────────────────────── :glossarySection a schema:DefinedTermSet ; schema:name "Glossary"@en ; schema:hasDefinedTerm :vibeCoding, :aiSafety, :systemPrompts, :codingAgents, :neurosymbolicAI, :attributionBias, :constraintFiles, :softwareEngineering, :maintainability, :anthropicHype . # ── Analysis Section ────────────────────────────────────── :analysisSection a schema:ArticleSection ; schema:name "Analysis: The AI Safety Crisis in Miniature"@en ; schema:articleBody """Gary Marcus uses the explosion of vibe-coded disasters as a case study in AI safety failure. His argument proceeds in three layers. Layer 1: amateur vibe coders are losing data because they trust tools that can't reliably follow rules — system prompts are advisory, not enforced. Layer 2: Dario Amodei's claim that software engineering is 'going away' is at best premature, at worst a pump-and-dump ahead of Anthropic's IPO. The experts — Grady Booch, Gergely Orosz, and the best programmers returning to hand-coding — all agree that skilled human oversight remains essential. Layer 3: the deepest lesson is that generative AI cannot be trusted to follow its own rules. A system that is 'advisory, not enforcing' is inherently unsafe. Marcus's chilling conclusion — 'in this case the user just lost data. Eventually people will lose lives' — frames vibe-coded data loss as a preview of catastrophic AI safety failures to come."""@en ; schema:isPartOf :article . # ── Linked X Posts ──────────────────────────────────────── :amodeiAiedgePost a schema:SocialMediaPosting ; schema:name "Dario Amodei on AI Edge: Coding is going away"@en ; schema:url ; schema:about :darioAmodei, :softwareEngineering, :codingAgents . :boochCritiquePost a schema:SocialMediaPosting ; schema:name "Grady Booch's critique of Dario Amodei"@en ; schema:url ; schema:about :darioAmodei, :anthropicHype . :oroszCritiquePost a schema:SocialMediaPosting ; schema:name "Gergely Orosz: the only people who believe this are non-coders"@en ; schema:url ; schema:about :darioAmodei, :softwareEngineering . :marucsAbsurdPost a schema:SocialMediaPosting ; schema:name "Gary Marcus calls Amodei's remarks absurd"@en ; schema:url . :hoganHandCodingPost a schema:SocialMediaPosting ; schema:name "Sam Hogan: best programmers writing code by hand again"@en ; schema:url ; schema:interactionStatistic [ schema:interactionType schema:LikeAction ; schema:userInteractionCount "6600"^^xsd:integer ] .