When AI prophecy fails (permalink)
🔗 **Source:** [When AI prophecy fails (permalink)](https://pluralistic.net/2025/10/29/worker-frightening-machines/)
> [!note]+ Document Notes
> tags: [AI, Amazon, labor]
> author: Cory Doctorow
> source: pluralistic.net
> reference: [https://pluralistic.net/2025/10/29/worker-frightening-machines/]
> date created: 2025-10-29 00:00
>
> # 🏷️ Title: When AI prophecy fails (permalink)
>
<!-- 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)
### 🔗 References
- Include links cited in or relevant to the article.
## 📝 Summary
- What happened: Amazon posts ~$35B profit, plans 14k–30k layoffs while touting AI readiness and massive AI capex.
- Why it matters: Signals a marketing-led push to validate AI replacement of labor despite weak ROI.
- Key stakeholders: Amazon workers, customers, AI vendors (e.g., Anthropic), cloud investors, regulators.
- Risks and tradeoffs: Service degradation, safety issues, sunk-cost AI spending, market power enabling “too big to care.”
- Bottom line: Layoffs function as theater to sell AI and buy time for overextended AI bets.
## 📖 Main Content
- Core arguments:
- AI layoffs are a performative signal to markets, not proven productivity gains.
- Most AI deployments show poor or negative returns; firms misattribute failure to implementation.
- Amazon already uses AI as a “digital whip” for labor control; replacing engineering is unrealistic due to context limits.
- Enshittification lets dominant firms degrade quality without losing customers, sustaining the AI narrative.
- Context/definitions:
- Enshittification: platforms worsen for users/suppliers as they extract value.
- MLM analogy: failing sellers blamed themselves; similarly, managers buy more AI training/consulting.
- Uncertainties:
- Timeline to economically viable, context-robust AI for complex engineering.
- Investor patience for ongoing AI losses.
### 🗒️ Notes
- Amazon: big profit, bigger AI spend (~$120B this year, ~$180B claimed overall); layoffs marketed as “AI ready.”
- Bain cited: sector needs ~$2T revenue by 2030 to break even.
- Evidence cited: 95% of AI projects produce no or negative ROI; little impact on wages/hours.
- Engineering reality: code-generation ≠ software engineering; context maintenance is the bottleneck.
- Customer service: firms that already underperform are likelier to swap humans for chatbots with no real capability loss.
- Cloud reliance: Amazon’s AI customers prop up AWS; those AI firms bleed cash, need investor belief to persist.
- Prime lock-in: K-shaped consumption + prepaid shipping reduce switching; enables quality degradation during peak season.
- Conclusion: layoffs are marketing theater to sustain the AI investment story and delay reckoning.
## 📚 References
- $2 trillion in new AI revenue needed by 2030 (Bain & Company): https://www.bain.com/about/media-center/press-releases/20252/$2-trillion-in-new-revenue-needed-to-fund-ais-scaling-trend—bain–companys-6th-annual-global-technology-report/
- MIT Tech Review analysis on AI ROI: https://www.technologyreview.com/2019/01/25/1436/we-analyzed-16625-papers-to-figure-out-where-ai-is-headed-next/
- Study: no significant impact on earnings/hours/wages: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5219933
- Ed Zitron on AWS dependence on AI spend: https://www.wheresyoured.at/costs/
- CNN on layoffs to be “AI ready”: https://www.cnn.com/2025/10/28/business/what-amazons-mass-layoffs-are-really-about
- Guardian on Amazon degradation: https://www.theguardian.com/technology/2025/oct/05/way-past-its-prime-how-did-amazon-get-so-rubbish
- Doctorow on delivery “digital whip”: https://pluralistic.net/2025/10/23/traveling-salesman-solution/#pee-bottles
- Doctorow on hype/layoffs dynamic: https://pluralistic.net/2025/08/05/ex-princes-of-labor/#hyper-criti-hype
## 🗣️ Paraphrase
- Paragraph A: Amazon is shedding tens of thousands of jobs while pitching AI as justification, despite enormous, unproven capex.
- Paragraph B: Most AI initiatives don’t pay off; companies blame themselves, buy more consulting, and double down.
## 📰 Newsletter
AI reality check: Amazon’s “AI-ready” layoffs show the gap between hype and delivery. For WordPress builders, focus on measurable value—ship docs, performance dashboards, and user-centered workflows. Useful reads: WordPress Performance Team, Core Editor Handbook, “Performance Lab” plugin, and WP-CLI.
## ✅ Action Items
#### DevRel
- Publish “AI in the loop” examples with measurable benchmarks and rollback plans.
#### OSPO
- Define AI-assisted contribution policy and provenance checks for generated code.
#### Community
- Run maintainer clinics on evaluating AI PRs and security pitfalls.
#### WordPress
- Core: add guardrails for AI-generated patches (tests, lint, ownership).
- Core Performance: track perf regressions from AI-suggested changes.
- Core AI: document responsible use, eval sets, red-team reviews.
#### The WPOCC
- Pilot microgrants for testing AI claims with open benchmarks.
### FAIR
- Require signed provenance for AI outputs; fund sunset clauses for pilots.
## 🧪 Hypotheses
- Firms replacing customer support with chatbots see lower CSAT and higher churn within 2 quarters.
- AI code generation increases defect rates unless paired with senior review and tests.
- Prime lock-in reduces elasticity to delivery delays during peak season.
- AI-heavy cloud revenue is disproportionately concentrated among loss-making customers.
### 📊 Methodology
- Diff-in-diff on CSAT/churn pre/post chatbot rollout.
- Repository mining: defect density vs. AI-tool adoption + reviewer seniority.
- Panel data: Prime households’ switching behavior vs. delivery SLAs.
- Segment AWS revenue by customer profitability; stress-test exposure.
### ❓ Questions
- How long can investors subsidize AI losses?
- What governance signals would trigger AI capex pullbacks?
- Which tasks show persistent net gains from AI without quality loss?
### 🔮 Future Research
- Longitudinal studies on AI-in-the-loop engineering outcomes.
- Benchmarks for customer service resolution quality with/without AI.
- Labor safety impacts under algorithmic management.
### 💡 Implications
- Practice: treat AI as assistive, not replacement, with audits.
- Policy: require service quality disclosures during mass AI-driven layoffs.
- Security: scrutinize AI-generated code provenance and SBOMs.
- Funding: tie AI spend to staged, independent evaluations.
### 🧠 Learning Styles Assessment
- Reading level estimate: Flesch-Kincaid ≈ Grade 11–12.
#### 👀 Visual learners
- Improve: architecture/context diagrams of AI-in-the-loop workflows.
#### 🔊 Aural learners
- Improve: 3–5 minute audio summary of findings.
#### 🗣️ Verbal learners
- Present: vivid analogies (MLM, “digital whip”).
#### 🧩 Kinesthetic learners
- Improve: hands-on checklists for evaluating AI outputs.
#### 🧠 Logical learners
- Present: causal chains; add KPI tables/decision trees.
#### 🧑🤝🧑 Social learners
- Improve: moderated Q&A and cohort reviews.
#### 🧘♀️ Solitary learners
- Improve: self-check quizzes on AI claims.
#### 🌿 Naturalistic learners
- Improve: lifecycle cost/energy analyses of AI workloads.
### 📊 Accessibility Assessment
- Improve: descriptive alt text for images; transcripts for media; avoid autoplay.
- Ensure 4.5:1 contrast, visible focus, skip links, responsive at 200% zoom.
- Map gaps to WCAG 2.2 AA.
### 🌍 DEIB Assessment
- Improve: global labor perspectives, plain-language summaries, glossary of AI terms.
- Balance quotes across roles; add content warnings for labor/safety topics.
### ✅ Logical Validity
- Strong reasoning: causal link between dominance and tolerance for service degradation; supported by Prime lock-in dynamics.
- Fallacies: watch for hasty generalization from select studies; remedy with broader meta-analyses.
- Tone: critical, polemical, uses analogies to illustrate systemic incentives.
### 🛡️ DARVO Analysis
- Deny: “AI works—you’re implementing it wrong.”
- Attack: Imply worker redundancy/incompetence to justify cuts.
- Reverse victim/offender: Portray firms as victims of labor costs while workers bear harms.
## 🎯 Conclusion
- Bottom line: Amazon’s layoffs are market theater to sustain AI narratives despite weak returns; dominance cushions the fallout.
- Next steps: Demand transparent metrics, stage-gate AI pilots, protect service quality, and strengthen labor and consumer safeguards.
## 🧱 Refactor Appendix (append after 🎯 Conclusion)
<!-- See instructions in the comment for how to use these pads and templates. Keep this block intact for tooling. -->
```markdown
### 📚 Literature Split Pad
- Purpose: Stage Konik CEE units that become per-source Literature notes.
- Split level: H3.
- Include verbatim quotes under 25 words and a [view highlight] link.
### [H1] 5–10-word present-tense claim
> [!quote] When AI prophecy fails (permalink) (https://pluralistic.net/2025/10/29/worker-frightening-machines/) by [[Cory Doctorow]] via [[pluralistic.net]]. Published 2025-10-29. Accessed .
> “Layoffs function as marketing to sell AI.”
Two lines: context and use. Cite data if relevant.
[view highlight](https://pluralistic.net/2025/10/29/worker-frightening-machines/#H1) #xref #articleseed ID:rw_[MISSING]
### [H2] 5–10-word present-tense claim
> [!quote] When AI prophecy fails (permalink) (https://pluralistic.net/2025/10/29/worker-frightening-machines/)
> “AI coding tools struggle with system context.”
Two lines: context and use.
[view highlight](https://pluralistic.net/2025/10/29/worker-frightening-machines/#H2) #xref ID:rw_[MISSING]
---
### 🔖 Atomic Split Pad (Zettel Candidates)
- Purpose: Promote reusable claims into Atomic notes.
- Split level: H4.
- Each item: one idea, present tense, specific, testable.
#### [Z1] Dominant firms tolerate quality degradation
- Peak-season e-commerce under lock-in.
- Links: [[Enshittification]] [[Switching-Costs]]
#### [Z2] AI codegen increases defects without review
- Repo mining with senior review variable.
- Links: [[Software-Quality]] [[AI-Assist]]
---
### ✏️ Fleeting Queue (Extract by Selection)
- [F1] Draft AI pilot stage-gate checklist — context/https://pluralistic.net/2025/10/29/worker-frightening-machines/ #fleeting #inbox
- [F2] Collect CSAT + churn datasets for chatbot rollouts — context/… #fleeting #inbox
- [F3] Map AWS exposure to loss-making AI customers — context/… #fleeting #inbox
---
<!-- deep-read:sections end -->