企业采用 AI Agent 的主要瓶颈是可靠性、数据访问，还是组织变革？

“Aaron Levie enumerates them in 20VC: regulatory constraints (SEC, FINRA), fragmented legacy data estates, annual budget cycles tied to EPS commitments, and the sheer organisational change management required. Even a simple task like finding all contracts across ten systems can fail because agents need curated, standardised data.”

“Even when AI tools and models are technically ready, the biggest hurdle identified at the 2025 Spark Private Company Conference was getting employees to actually use them. The phrase “You can lead a horse to water, but you can’t make it drink” encapsulates the challenge.”

“At a CIO conference described in the October 2025 Thoughts on the Market episode, only 4 out of 150 CIOs (2.7%) raised their hand when asked if they had a good understanding of agentic AI. This extremely low level of comprehension highlights that: The technology is still in an early exploratory phase.”

“Contrary to many expectations, Ian Livingston and Joel De La Garza predict that enterprises will adopt AI agents at scale before consumers, for several reasons: Massive internal ROI – Agent-enabled workflow optimization directly improves earnings efficiency; this is a top-level business objective, not a developer productivity side-effect.”

“Jesse Zhang observed that many large organisations are now framing AI integration as a “transformation” imperative, with customer service identified as the lowest-hanging fruit because it delivers clear, measurable ROI (e.g., 60–70% contact-center cost reductions). This dynamic makes enterprises far more willing to trial startup solutions than in the past, opening markets that were previously slow to adopt.”

“Enterprise AI adoption is thus at a crossroads: bottom-up enthusiasm is creating rapid productivity gains, but data privacy, cost, and governance issues remain unresolved. Chamath Palihapitiya raised a critical challenge: using public LLM endpoints (e.g., ChatGPT) can waive attorney-client privilege and expose confidential data.”