文发论点是企业对 AI agents 的高投入往往在早期部署阶段停滞，关键原因不是“智能不足”，而是缺少人类监督与问责来保证安全与有效运行。MediaMint CEO Rajeev Butani 指出，问题从更早的工作流定位与责任划分就开始。Gartner 的一项调查进一步量化了这种犹豫：只有 15% 的 IT 应用负责人表示正在考虑、试点或部署“完全自主”的 AI agents（不需要人类监督的目标驱动工具）。

文中用多组比例描述信任缺口与规模化障碍。Prosper Insights & Analytics 的消费者调查显示，39% 的消费者认为 AI 工具需要更多人类监督，尤其在高风险或情绪敏感情境下；在组织内部，38.9% 的高管与 32.7% 的员工表示 AI 系统需要人类监督才值得信任。与此一致，Xebia 的 Data & AI Monitor 指出，当系统缺乏透明性、情境支撑与清晰问责时，团队会犹豫采用；在真实运营中，缺少护栏会使 agents 在边缘案例、含糊指令与需求变化下失稳。文章把 agents 类比为“初级成员”：若缺少决策规则与持续指导，就会误读数据、做出错误假设并采取偏离业务目标的行动。

成功路径被概括为 human-plus-AI：agents 承担重复、重体力的执行，人在最后一层做判断与校准，例如广告运营里 agents 可在几分钟内生成媒体计划初稿，由策略师再结合语境与目标精炼。该组合被认为能降风险、提质量并减少低价值劳动；Prosper 数据也显示 15.9% 的员工与 13.9% 的高管会因 AI 感到焦虑，支持用人类引导来降低压力并提高信心。Butani 将这种把 agents 嵌入工作流并由训练有素的操作员验证与共担结果的模式称为 Service-as-a-Software；文章建议领导者以数据质量、人类在环监督、明确所有权结构与可见的早期用例来实现可持续扩展。

Published at 2026-01-08 10:00:00 (EST), the article argues that heavy enterprise investment in AI agents often stalls at early deployment because the core failure is not “lack of intelligence,” but missing human oversight and accountability needed for safe, effective operation. MediaMint CEO Rajeev Butani says the breakdown starts earlier, in how companies frame agents inside workflows and assign responsibility. A Gartner survey quantifies the hesitancy: only 15% of IT application leaders say they are considering, piloting, or deploying fully autonomous AI agents (goal-driven tools that do not require human oversight).

Several percentages are used to map the trust gap and scaling barriers. A Prosper Insights & Analytics consumer survey reports that 39% of consumers say AI tools need more human oversight, especially in high-stakes or emotionally sensitive scenarios; inside organizations, 38.9% of executives and 32.7% of employees say AI systems require human oversight to be trusted. Consistent with this, Xebia’s Data & AI Monitor is cited for the idea that adoption hesitates when systems lack transparency, contextual grounding, and clear accountability; in real operations, weak guardrails make agents struggle with edge cases, ambiguous instructions, and evolving needs. The article treats agents like junior team members: without decision rules and continuous guidance, they misread data, make incorrect assumptions, and act out of alignment with business goals.

The proposed success model is human-plus-AI: agents do heavy, repeatable execution while people provide final judgment and refinement, such as an advertising agent drafting a media plan in minutes and a strategist adjusting it for context and client goals. This pairing is framed as reducing risk, improving quality, and cutting low-value work; Prosper data also shows 15.9% of employees and 13.9% of executives say AI makes them anxious, supporting human guidance to lower stress and raise confidence. Butani labels the workflow-embedded, operator-validated, shared-accountability approach Service-as-a-Software; leaders are urged to focus on data quality, human-in-the-loop oversight, clear ownership structures, and early use cases with visible wins to scale responsibly.