One real reason AI isn’t delivering: Meatbags in manglement

Summary

The article argues that AI’s poor business impact is not a technology failure but a management one. Despite heavy investment and widespread pilots, studies (including an MIT report) show only around 5% of custom AI initiatives scale to production and many organisations report no measurable returns. The core problem is treating AI as a plug‑in piece of software instead of a new form of labour that needs context, memory and workflow redesign.

Key Points

Mass adoption of GenAI has not translated into ROI: many pilots stall and few projects reach production.
Organisations often treat AI as conventional software; AI behaves more like a worker requiring training, context retention and integration into processes.
Pilots usually fail when grafted onto existing workflows rather than redesigning work to complement AI outputs and catch errors.
Statefulness matters: successful AI systems accumulate context and improve over time rather than being stateless after each interaction.
Top success factors include process designers, workflow architects and domain experts translating AI into day‑to‑day operations.
Collaborating with external partners and bottom‑up, employee‑led experiments raise the odds of scaling AI successfully.
Biggest real ROI is often in back‑office automation (finance, operations, supply chain), not the visible front‑office projects favoured by execs.

Why should I read this?

Short version: if your shiny AI pilots never leave the lab and budgets keep evaporating, this piece tells you why — and it’s not the model’s fault. Read it if you want a blunt, practical nudge to stop buying tools and start building capabilities.

Context and relevance

This is important for leaders and practitioners wrestling with AI adoption. The article reframes the debate from chasing bigger models to changing how work is organised: hire or contract process experts, make AI stateful where needed, favour bottom‑up proofs that solve real problems, and hunt ROI in often overlooked back‑office areas. The insight is timely given continued heavy investment in GenAI and persistent reports of limited measurable benefit.

Source

Source: https://go.theregister.com/feed/www.theregister.com/2025/12/24/reason_ai_isnt_delivering/