Glean：全球首家企业级生成式 AI 公司

Glean: The World's First Enterprise Generative AI Company

阿尔温德·贾恩 Arvind Jain · Composio · 2026-08-07 · 约 48 分钟 · 原视频 ↗

本期速览 · Overview

Arvind Jain 探讨 Glean 从企业搜索到全面 AI 平台的历程，强调 AI 在企业中的巨大潜力。

Arvind Jain discusses Glean's journey from enterprise search to a comprehensive AI platform, emphasizing the vast potential of AI in the enterprise.

AI 仍处于起步阶段，当前的使用量与未来 5-10 年相比微不足道。
AI is in its infancy; current usage is minuscule compared to the next 5-10 years.
初创公司最大的挑战是执行力和信任，而非竞争。
Startups' biggest challenge is execution and trust, not competition.
开源模型将在 12-18 个月内处理超过 90%的企业 AI 任务。
Open source models will handle over 90% of enterprise AI tasks within 12-18 months.

我们正处于AI革命的非常早期阶段，当前AI的使用量与5到10年后相比微不足道。
We are in the very early stages of the AI revolution, and current AI usage is minuscule compared to what it will be in 5-10 years.
初创公司最大的挑战不是竞争，而是执行力、打造高质量产品以及赢得客户信任。
The biggest challenge for startups is not competition but execution, building quality products, and earning customer trust.
企业需要通过关注部门级指标（如案例解决时间或每位律师处理的合同数）来衡量AI的投资回报率。
Enterprises need to measure ROI of AI by focusing on departmental metrics like case resolution time or contracts handled per lawyer.
开源模型已经达到能够处理超过90%企业AI任务的程度，未来12到18个月我们将看到大部分推理转向开源。
Open source models have reached a point where they can handle over 90% of enterprise AI tasks, and we will see majority of inference shift to open source in the next 12-18 months.
AI成本是企业最关心的问题，需要大幅降低才能实现广泛价值。
The cost of AI is the number one concern for enterprises, and it needs to come down significantly to realize widespread value.
AI应该像软件一样边际成本可忽略，未来我们将以更低的成本用AI做更多工作。
AI should become like software with negligible marginal cost, and we will do more work with AI at lower cost in the future.

竞争不值得担心；即使所有AI公司合并，也满足不了企业需求的10%。
Competition is not something to worry about; even if all AI companies merged, they would still meet less than 10% of enterprise demand.
AI将消耗一半运营支出的说法是有问题的；价值实现必须先行。
The narrative that AI will consume half of opex is problematic; value realization must come first.
代币支出不应与员工人数比较；将AI视为与劳动力成本的权衡是荒谬的。
Token spend should not be compared to headcount; it's ridiculous to frame AI as a trade-off with labor costs.
用MCP构建的智能体常常失败，因为它们缺乏资深员工的深度上下文；投资于上下文图谱至关重要。
Agents built with MCP often fail because they lack the deep context of tenured employees; investing in context graphs is crucial.
对代币的过度关注将会消失；我们从不谈论CPU周期，几年后我们也将不再谈论代币。
The fixation on tokens will go away; we never talked about CPU cycles, and we will stop talking about tokens in a few years.
即使是大型企业也发现由于动态竞争很难承诺给模型提供商；像Glean这样的平台吸收了这种风险。
Even large enterprises find it extremely hard to commit to model providers due to the dynamic race; platforms like Glean absorb that risk.