On May 4, 2026, David Gewirtz published an analysis arguing that organizations face a stark strategic choice as agentic AI matures: pursue safe, incremental automation or chase transformational, tenfold improvements that carry significant operational and financial risk. He frames the decision around two concrete pathways — aiming for modest human‑capital savings on the order of 10% versus launching an aggressive program designed to multiply business outcomes — and cautions that the latter can backfire without disciplined controls.
Gewirtz lays out practical implications for builders and executives. Not every product labeled "agentic AI" delivers autonomous, goal‑directed behavior; many early efforts are experiments or proofs of concept that may not scale. He stresses the importance of anchoring initiatives to measurable outcomes rather than hype, and treating experiments as learning vehicles rather than automatic deployment candidates. The analysis highlights common failure modes that can turn ambitious projects into operational and reputational liabilities. Poor prompting and so‑called rogue agents can cascade into functional breakdowns and business loss. Vendor marketing practices he describes as "agent washing"—rebranding assistants, RPA, script‑based services and chatbots as agentic — further muddle procurement decisions and inflate expectations.
Industry forecasts underpin the stakes of that gamble. KPMG estimates agentic AI could unlock roughly $3 trillion in annual productivity gains, while Accenture characterizes agentic systems as a new form of capital. Gartner adds urgency, advising organizations they have a three‑ to six‑month window to define an agentic AI product strategy as the sector hits an inflection point.
Gewirtz points to concrete risk signals uncovered in research. Gartner warns that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. He also cites vendor market analysis that finds fewer than 13% of the thousands of suppliers positioning agentic solutions actually deliver genuinely agentic products. On the technical and economic front, Gewirtz underscores heavy dependence on external large language models for cognitive processing, naming providers such as OpenAI and Google. That reliance produces steady, high recurring costs and exposure to third‑party pricing and availability, elevating the economic risk of broader rollouts if usage and guardrails are not tightly managed.
His bottom line is caution coupled with discipline: pursue proof‑of‑value metrics, verify vendor claims of agentic functionality, and enforce cost and safety guardrails before scaling. Experiments will fail; the objective is to let measurable ROI guide expansion so that unchecked ambition does not convert potential upside into canceled programs and financial loss.
Sources
Replies (0)
No replies in this topic yet.
