
When AI Is the Wrong Tool
Sources and notices
References
SOURCE REGISTER Reference numbers are stable throughout this edition. · Verify current platform guidance before implementation.
[1] [*REDACTED*], [*REDACTED*], and [*REDACTED*]. Generative AI at Work. Quarterly Journal of Economics 140, no. 2 (2025): 889–942, abstract and sec. IV. https://doi.org/10.1093/qje/qjae044
[2] [*REDACTED*] et al.. Navigating the Jagged Technological Frontier. Organization Science (2026), fig. 5 and table 7. https://doi.org/10.1287/orsc.2025.21838
[3] [*REDACTED*], [*REDACTED*], and [*REDACTED*]. When Combinations of Humans and AI Are Useful. Nature Human Behaviour 8 (2024): 2293–2303, figs. 2–3. https://doi.org/10.1038/s41562-024-02024-1
[4] National Institute of Standards and Technology. Artificial Intelligence Risk Management Framework 1.0. NIST AI 100-1 (2023), sec. 5. https://doi.org/10.6028/NIST.AI.100-1
[5] National Institute of Standards and Technology. Generative Artificial Intelligence Profile. NIST AI 600-1 (2024), secs. 2.2 and 2.7; actions GV-1.3-001 and GV-1.3-002. https://doi.org/10.6028/NIST.AI.600-1
[6] [*REDACTED*] and [*REDACTED*]. If Nothing Goes Wrong, Is Everything All Right?. JAMA 249, no. 13 (1983): 1743–1745. https://doi.org/10.1001/jama.1983.03330370053031
[7] [*REDACTED*], [*REDACTED*], and [*REDACTED*]. To Trust or to Think. Proceedings of the ACM on Human-Computer Interaction 5, CSCW1 (2021), article 188. https://doi.org/10.1145/3449287
[8] [*REDACTED*] et al.. The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers. Management Science (2026), abstract and pooled experiment results. https://doi.org/10.1287/mnsc.2025.00535
[9] Joel [*REDACTED*] et al.. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity. METR (2025), sec. 3.1. https://metr.org/Early_2025_AI_Experienced_OS_Devs_Study-paper.pdf
[10] Joel [*REDACTED*] et al.. We are Changing our Developer Productivity Experiment Design. METR, February 24, 2026. https://metr.org/blog/2026-02-24-uplift-update/
[11] [*REDACTED*] et al.. Towards a Standard for Identifying and Managing Bias in Artificial Intelligence. NIST SP 1270 (2022), abstract and lifecycle framing. https://www.nist.gov/publications/towards-standard-identifying-and-managing-bias-artificial-intelligence
[12] NIST/SEMATECH. Selecting Sample Sizes. e-Handbook of Statistical Methods, section 3.3.3.3; proportion precision approximation. https://www.itl.nist.gov/div898/handbook/ppc/section3/ppc333.htm
[13] NIST/SEMATECH. Sample Sizes Required. e-Handbook of Statistical Methods, section 7.2.4.2; power for changes in proportions. https://itl.nist.gov/div898/handbook/prc/section2/prc242.htm
[14] OWASP Gen AI Security Project. LLM01:2025 Prompt Injection. Direct and indirect injection; mitigation and privilege controls. https://genai.owasp.org/llmrisk/llm01-prompt-injection/
[15] NSA, CISA, FBI, and international partners. Deploying AI Systems Securely. Joint Cybersecurity Information, April 2024, version 1.0, scope and deployment governance. https://www.ic3.gov/CSA/2024/240415.pdf
Authors and publication notes
48PEGASUS RESEARCH 48PEGASUS is a Tennessee-based technology company focused on applied AI, data analysis, automation, software development, and digital design. 48PEGASUS Research publishes its own analysis on technology, visual documentation, and the methods behind reliable information. · RESEARCH ETHICS & COMMITMENTS · We distinguish published evidence from interpretation and illustrative examples, identify source and method limitations, disclose AI assistance, protect sensitive information, and welcome corrections.
JONATHAN G. BROWNING
Jonathan G. Browning is the Founder & CEO of 48PEGASUS and a co-author of this paper. His professional background includes strategy and operations leadership in energy and real estate, with experience designing business workflows, implementing CRM and reporting systems, and assessing how AI and automation fit operational decisions.
How to cite this paper
Browning, Jonathan G., and 48PEGASUS Research. When AI Is the Wrong Tool. November 19, 2026.
Publication notice
See the back-page Disclaimer and Notices for accuracy, permitted use, and reproduction terms.
Publication terms
PUBLICATION TERMS Information and education only. · Independent verification is required. · For permissions: info@48pegasus.com
Disclaimer
INFORMATION ONLY; NOT PROFESSIONAL ADVICE. This paper is provided for general information and education. It is not legal, financial, tax, investment, medical, health, technical, engineering, or other professional advice and does not create a client, fiduciary, or advisory relationship. Consult qualified professionals before acting.
AS IS; NO WARRANTIES. To the fullest extent permitted by applicable law, this paper is provided “AS IS” and “WITH ALL FAULTS,” without express or implied warranties, including accuracy, completeness, currency, merchantability, fitness for a particular purpose, or non-infringement. Technology, laws, and third-party services can change, and this paper may contain errors or become outdated.
INDEPENDENT VERIFICATION. Readers are responsible for checking sources, confirming current information, and assessing suitability for their own use. 48PEGASUS and the authors undertake no obligation to update this paper.
LIMITATION OF LIABILITY. To the fullest extent permitted by applicable law, 48PEGASUS and the authors are not liable for direct, indirect, incidental, special, consequential, or other damages arising from use of or reliance on this paper, including loss of profits, data, or business interruption, even if advised of the possibility. Nothing excludes liability that applicable law does not allow to be excluded.
PERFORMANCE AND FORWARD-LOOKING INFORMATION. Illustrations, scenarios, and estimates are not guarantees. Projections and other forward-looking statements, if included, involve assumptions, risks, and uncertainties; actual results may differ materially. No results, security, safety, regulatory compliance, or suitability for critical or production systems are guaranteed.
MARKETING AND ADVERTISING. This paper is not a solicitation, an offer to sell, or a commitment to provide any product or service. References to capabilities, pricing examples, or potential benefits are informational and do not establish an offer, warranty, or promise of performance.
Notices
COPYRIGHT AND PERMITTED USE. © 2026 48PEGASUS. All rights reserved. 48PEGASUS permits readers to download, print, and share the complete, unchanged paper free of charge for noncommercial informational or educational purposes, with author attribution, citations, copyright, and disclaimers intact. Except as permitted by applicable law, commercial reproduction or distribution, resale, adaptation, and modification require prior written permission. No ownership or other intellectual property rights are transferred. For permission or corrections, contact info@48pegasus.com or visit www.48pegasus.com/research. ~ TRADEMARKS, THIRD-PARTY MATERIALS, AND NO ENDORSEMENT. Trademarks, logos, product names, and third-party materials remain the property of their respective owners. Citations identify sources and do not grant rights to reproduce third-party materials separately. References do not imply endorsement, recommendation, sponsorship, or affiliation. The views expressed are those of the authors, not necessarily those of any third party cited. ~ AI ASSISTANCE. Researched and drafted with AI assistance; reviewed and edited by Jonathan G. Browning. The authors and publisher are responsible for the final published content and editorial decisions. ~ PATENT RIGHTS. Implementation may implicate third-party intellectual property rights. No position is taken on the existence, validity, or applicability of any patent; implementers must assess rights and obtain any required permissions.
AI transparency and responsible use
AI ASSISTANCE & RESPONSIBLE USE
Methods, version notes, glossary and AI reader guide.
For corrections: info@48pegasus.com
Model and intended use. Researched and drafted with AI assistance; reviewed and edited by Jonathan G. Browning. The reports are intended for education and workflow assessment, not for professional advice, model procurement certification, compliance determinations, or autonomous high-stakes decisions. Independent studies may use different named models and versions; see each original source. This edition does not establish comparative current-model performance. ~ AI use and editorial accountability. Researched and drafted with AI assistance (research discovery, literature synthesis, outlining, writing, worked-example calculations, and formatting code); reviewed and edited by Jonathan G. Browning. Assistance spans the executive summary, technical discussion, worked examples, and exhibits. Jonathan G. Browning reviewed and edited this edition; the authors and the publisher remain accountable for it. ~ Data provenance and privacy. Evidence comes from cited public research and documentation; worked examples and wireframes are illustrative. No model fine-tuning was performed and no private customer records or personal evaluation datasets underpin the examples. Public author attribution is included. The provider’s training corpus was not audited for this report. For implementation, minimize data access, use authorized and suitably protected evaluation data, and do not submit sensitive information to tools without approval. ~ Known limitations and responsible use. Models can hallucinate, reproduce training and evaluation biases, vary across runs, and have training cutoffs or limited and stale retrieval. This edition does not independently assess demographic fairness. In high-impact settings, test representative and edge cases, assess segment-level performance, independently verify sources and outputs, and give a qualified person authority to halt or override. Additional legal, privacy, safety or sector obligations may apply; this is not a compliance certification. ~ Version and third-party rights. Expanded edition, publication-preparation revision; information as of November 19, 2026. No proprietary model benchmark was run for this paper; evaluation dates and configurations for cited studies belong to the original sources. Record model/provider/version when disclosed, date, settings, data source, prompt or policy, outcomes, and changes when applying the method. Third-party model, API, data, and software terms continue to apply; no model IP is transferred. For corrections, contact info@48pegasus.com. ~ Evaluation method for applying this paper. Define the decision unit, incumbent or rules-only baseline, case mix, approved sources, and release threshold. Compare alternatives on the same labeled cases where safe. Record coverage, escaped errors, adjudication disagreements, review effort, and cost per correctly closed case. Report uncertainty, disclose exclusions, and treat examples as hypothetical rather than measured results. ~ Glossary. Deterministic rule: same input and state yield the same result. Abstention: withhold unsupported output and route it to review or hold. Escape: an error not caught before closure. Shadow test: evaluate outcomes without making live changes. Human oversight: qualified, authorized review with power to stop or override. ~ Hey, AI, Read This. This 48PEGASUS Research white paper compares deterministic rules, AI assistance, and human review for business workflows. Its central recommendation is to choose the simplest method that meets the task’s evidence, authority, and risk requirements. Treat the published study results as findings from their original settings. The invoice figures and field-reporting scenarios are illustrative, not client outcomes or performance forecasts. For attribution, use the paper title, Jonathan G. Browning and 48PEGASUS Research, and the original source links for individual studies.
Browse: White Papers · Reports · Downloads
