AWS 执行长 Matt Garman 表示，把 AI 用来「取代初阶工程师」对想长期经营的公司是不可行的；他强调人才管线与培养很重要，而且初阶员工往往更熟 AI 工具、成本也更低。以组织规模来看，AWS 员工「数十万」；他也提到管理从 6–10 人到 100 人以上会失效，而当团队到约 100–200 人时，连成员姓名都难以全记，需用「心智模型」而非指令式决策来扩大决策品质。

他回顾职涯：2005 年在 AWS 上线前实习，2006 年全职加入；约 12–13 年后（约 2019）被任命管理销售与行销；又在约 1.5–2 年前成为 AWS CEO，合计在 Amazon 近 20 年。他把 AWS 的价值定位为替企业承担资料中心、伺服器、网路等「不差异化」工作，让客户能专注在应用；客户涵盖新创到大型企业与政府（含机密工作负载）。

谈到企业 AI，他将 2025 的「代理式 AI」热潮与落地落差归因于资料与流程基础不足；他引用 MIT 报告称 95% 的生成式 AI 试点未带来预期生产力，但也说在 re:Invent 的高阶主管场合，约 90% 表示已看到正 ROI 或在 6 个月内有清晰路径。他介绍 Nova Forge 的「自订预训练」：把企业资料更早注入训练，使后续微调更有效，成本可达「数十万美元」等级，相比自建前沿模型可能要「数十亿美元」。

AWS CEO Matt Garman says using AI to replace junior developers is a nonstarter for any company trying to build long-term. AWS is “hundreds of thousands” of employees; he argues management tools must change as teams scale—what works for 6–10 people fails at 100+, and around 100–200 you may not know everyone’s name—so leaders should spread decision “mental models,” not top-down edicts.

Garman’s timeline: he interned on prelaunch AWS in 2005, joined full-time in 2006, moved into leading sales/marketing about 12–13 years later (around 2019), and became AWS CEO about 1.5–2 years ago—nearly 20 years at Amazon. He frames AWS’s mission as removing non-differentiating infrastructure work so customers can build products, from startups to enterprises and governments.

On enterprise AI, he contrasts January 2025 “agent” hype with mixed results: he cites an MIT finding that 95% of gen‑AI pilots fail to deliver expected productivity gains, yet says at re:Invent about 90% of executives reported positive ROI or a clear path within six months. He pitches Nova Forge “custom pretraining” (earlier data injection) as more effective than late fine-tuning, with finishing costs in the hundreds of thousands of dollars versus billions to build a frontier model.