Reading 03 · Enterprise
AI for Business & the Business of AI
LLMs show real promise on perceptual tasks — but reasoning limits, organizational silos and legacy infrastructure decide what actually ships. In the enterprise, constraints trump capability.
The thesis
Near-term value is real but modest; the true blockers to enterprise AI are organizational, not technological — and most of the infrastructure to manage AI at scale does not yet exist.
The argument, in five moves
What really decides enterprise adoption
Quality has a ceiling
Elemental tasks — text processing, document generation, dialog management — land around 60–70% quality, insufficient for complex reasoning without human oversight.
Organization trumps technology
Adoption hinges on legacy IT architecture, data silos, process fragmentation and workforce expertise gaps as much as on model capability.
Bottom-line now, top-line later
Near-term value comes from automating document workflows and customer touchpoints; durable growth requires embedding AI into the products themselves.
The infrastructure is unfinished
Compliance, governance, model-evaluation frameworks and lifecycle management remain largely unsolved at enterprise scale.
The labour paradox
Despite the automation promise, maintaining AI systems demands higher-caliber expertise — a workforce transition, not a simple headcount reduction.
“Reasoning,” in practice, redefined as doing things in stages, with intermediary pauses.
The Nyāya lens · व्यवहार
The tradition distinguishes abstract truth from vyavahāra — knowledge as it lives inside practice and institution. AI's enterprise problem is a vyavahāra problem: capability means little until it survives real workflows, real data and real people. And as a disintermediator, it unsettles the knowledge gatekeepers who once owned those workflows.
Distilled from the original essay on engkraft.com.