1. Deskilling and upskilling with AI systems
Author: Kevin Crowston & Francesco Bolici Year: 2025 Type: Academic article, conference paper (iConference 2025) Publisher: Information Research, 30 URL: https://doi.org/10.47989/ir30iconf47143 (open access)
What this source contributes
A model of when working with an AI system builds skill and when it erodes it. The distinction turns on what the person does with the system’s output: engaging with it as material to understand builds skill, accepting it without understanding the underlying task erodes it. Peer-reviewed; it carries the entry’s Deskilling sentence, the claim that improving AI can deskill AI proficiency itself.
Related entries
- AI-Ninja — the conditions under which the ninja’s own practice hollows out
- Deskilling — the general mechanism, with AI as the case
2. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers
Author: Hao-Ping (Hank) Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks & Nicholas Wilson Year: 2025 Type: Academic article, conference paper (CHI 2025) Publisher: Microsoft Research and Carnegie Mellon University; Proceedings of the CHI Conference on Human Factors in Computing Systems URL: https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/
What this source contributes
A survey of 319 knowledge workers with 936 first-hand examples of AI use at work. The more confidence a worker had in the AI, the less critical thinking they reported doing; the more confidence in their own ability, the more. Critical thinking shifted towards verifying, integrating and stewarding output rather than producing it. Used in Why This Matters for the question of whether what was built during the ninja’s window transferred into understanding.
Related entries
- AI-Ninja — the measured direction of the entry’s closing question
- AI Dependency — trust in the system as the predictor of reduced thinking
3. You Don’t Need Prompt Engineering Anymore: The Prompting Inversion
Author: Imran Khan Year: 2025 Type: Preprint (not peer-reviewed) Publisher: arXiv, 2510.22251 URL: https://arxiv.org/abs/2510.22251
What this source contributes
A single-author preprint testing three prompting strategies across three OpenAI model generations on one benchmark (GSM8K, arithmetic word problems). A rule-heavy method the author calls Sculpting raised GPT-4o’s score from 93 to 97 percent and lowered GPT-5’s from 96.4 to 94.0: the constraint that had been a guardrail became a handcuff, in the author’s words. One benchmark and one researcher, so the entry cites it with that qualification; its value is that it measures the entry’s claim on the newest models rather than the 2022 ones Zhou et al. used.
Related entries
- AI-Ninja — the most recent measurement of the skill turning against its owner
- Prompt Engineer — the same finding from the profession’s side
4. AI Prompt Engineering Is Dead
Author: Dina Genkina Year: 2024 (6 March) Type: Journalism Publisher: IEEE Spectrum URL: https://spectrum.ieee.org/prompt-engineering-is-dead
What this source contributes
The article that gave the discourse its phrase. Reports research by Rick Battle and Teja Gollapudi at VMware, in which automatically generated prompts beat human-written ones on mathematical problems, sometimes with prompts no person would have written, and Intel Labs’ NeuroPrompts for image generation. Its conclusion is more careful than its title: the job survives as part of broader model operations. Useful as the public form of what Zhou et al. showed in the literature.
Related entries
- AI-Ninja — the moment the skill’s obsolescence became a headline
- Prompt Engineer — the profession the headline was about
5. Job Posts With These Words Are Twice As Likely to Be Filled by a Woman
Author: Fortune, reporting on Textio and Paradigm Year: 2016 (11 November) Type: Journalism, reporting on a data analysis Publisher: Fortune URL: https://fortune.com/2016/11/11/hire-women-textio
What this source contributes
Reports Textio’s analysis of its database of fifty million job postings: which phrases attract and which deter women applicants. Ninja and rockstar sit among the deterring terms, alongside aggressive and dominant; gender-neutral postings also filled two weeks faster on average. This is the source that connects Gaucher, Friesen and Kay’s mechanism to the word the entry is about; the paper itself never names ninja. A commercial analysis reported by a magazine, so secondary, and cited for the word list rather than for the mechanism.
Related entries
- AI-Ninja — the word placed on the list of words that turn women away
- Broligarchy — the same vocabulary, at the top of the industry
6. AI-geletterdheid Companion — studenteneditie
Author: Vrije Universiteit Amsterdam, Centre for Teaching & Learning Year: 2025 (version 1.1, October) Type: Institutional handbook Publisher: Vrije Universiteit Amsterdam URL: https://vu.nl/nl/student/studievaardigheden/handboek-ai-geletterdheid-studenteneditie
What this source contributes
What a Dutch university now teaches its students under the name AI literacy: how the models work, their environmental cost, their effect on work and on power relations, critical and responsible use, with prompting as one skill among these rather than the skill. Not research; a document of the institutional counter-frame to the ninja, and the one the entry’s own project sits closest to.
Related entries
- AI-Ninja — the counter-frame, as an institution defines it
- AI Literacy — the term, in institutional practice
7. Gender Decoder
Author: Kat Matfield Year: n.d. Type: Web tool Publisher: gender-decoder.katmatfield.com URL: https://gender-decoder.katmatfield.com/about
What this source contributes
A free tool that scans a job advertisement for masculine- and feminine-coded words, built directly on the word lists of Gaucher, Friesen and Kay (2011). Included as an artefact: it shows the primary source turned into practice, and it is the fastest way to see for oneself how a posting that calls for an AI-ninja is coded.
Status
Artefact. The tool is the phenomenon’s countermeasure in product form, not a source of evidence about it; its word lists are the paper’s, and any finding it produces should be traced back there.
Related entries
- AI-Ninja — the paper’s word lists, applied to a job ad