Brian Nowak compares compute rental, model services on owned infrastructure and model services using rented capacity. Pricing power and processing efficiency drive the resulting economics. The returns discussed are scenario estimates rather than realized results; the scale of capital expenditure alone cannot establish investment success.
Morgan Stanley argues that agents require identity controls covering purpose, duration and permitted actions, with decisions made during execution across systems. This could expand identity-security demand. Market-size estimates are analyst forecasts, not realized demand or evidence that a particular security product prevents agent misuse.
Goldman Sachs combines adoption surveys with occupational employment data and finds weaker hiring in more exposed sectors, with potentially greater pressure on junior workers. Economy-wide effects remain limited in its estimates. These associations and model-based comparisons do not attribute every departure from employment trends to AI.
Anthropic reviews misuse it detected and disrupted between December 2025 and August 2026, including workflow execution, fraud and surveillance. The cases motivate ongoing monitoring. They are selected observations within one provider’s visibility, not prevalence estimates or evidence that humans have left the command chain.
Bain’s survey of 951 companies identifies a gap between savings targets and reported outcomes, with many agents still requiring human approval or exception handling. The authors recommend reflecting oversight, data integration and workflow redesign in investment cases. Survey associations and consulting experience do not establish a single cause of success or failure.
Bain separates AI’s industry impact into productivity, innovation and shifts in competitive share. Its $4.7 trillion estimate concerns profit pools over 2025–2035 and includes creation and redistribution of profits. It is a forward-looking analytical estimate, not realized net gains or a certain forecast.
About half of 70 executives surveyed by BCG reported observing deskilling. The authors argue that judgment, problem framing and creativity need continued practice within everyday work. These findings reflect a small executive sample’s perceptions, not direct measurement of workers’ cognitive decline.
BCG’s AI Radar surveyed 2,360 executives, including 640 CEOs. Nearly three-quarters of CEOs described themselves as their organization’s main AI decision maker. The report frames AI as a strategy, workforce and governance issue. Investment intentions and confidence in returns are expectations, not realized financial results.
BCG argues that business-built CRM agents introduce risks around customer records, consequential decisions and consumption costs. It recommends shared data architecture, agent governance and cost controls while moving development closer to business users. This is practice-based governance guidance, not a universally validated deployment template.
BCG describes decision agents that assemble cross-functional evidence, identify missing data and compare scenarios under different assumptions. The proposed value is better-informed human deliberation, not automatic transfer of strategic accountability. This consulting perspective does not establish a general causal improvement in boardroom decision quality.