AI risks: wage pressure, training cost, morale
Multiple experts warn AI’s impact isn't only job loss — it can depress wages, shift bargaining power, and erode engineers’ self‑efficacy; researchers and industry voices say teams are paying hidden costs when skilled staff train models or over‑rely on automation argued, reported, and reported.
A peer‑reviewed experiment (N = 269) and follow‑up survey (N = 270) found passive use of AI—copying AI output—reduced workers’ self‑efficacy, ownership, and meaning, while active human–AI collaboration preserved those ties to work (Scientific Reports). (nature.com) A cost‑modeling paper tracing frontier training runs found amortized training costs rose roughly 2.4× per year since 2016 and projected the largest runs could exceed $1 billion by 2027, with AI accelerator chips and staff costs each accounting for “tens of millions” in major runs (arXiv). (arxiv.org) Former Salesforce AI CEO Clara Shih warned publicly that AI can lower wage floors by reducing barriers to tasks, a dynamic Business Insider reported in March 2026, and an HBR survey of 1,006 global executives in December 2025 found many layoffs have been driven by anticipated—not yet realized—AI gains. (businessinsider.com) Open a leadership note with SCQA (Situation, Complication, Question, Answer) as defined by Barbara Minto, leading with the recommendation and then three grouped supporting points—Minto’s Pyramid Principle is the same top‑down pattern taught at McKinsey for rapid decisioning. (managementconsulted.com) Make three hard metrics the core of any AI‑risk slide: hours of senior engineer time devoted to dataset curation or model tuning (staff costs are a major line item in frontier runs, per the arXiv model), expected turnover cost per departure (SHRM estimates replacement ranges from 50%–200% of annual salary), and incremental training/ops spend set against corporate training budgets (U.S. companies spent about $102.8 billion on training in 2025). (arxiv.org) Surface governance and residual risk with one page: a RAID log (Risks, Assumptions, Issues, Dependencies) for near‑term automation hazards plus a RACI matrix to fix decision rights on model deployment and compensation changes—both are standard project governance tools used to prevent ambiguous handoffs in cross‑functional programs. (thedigitalprojectmanager.com)
