💰TechCrunch AI•較早收集於 60m
Aaron Levie 稱科技業執行長正遭受「AI 精神錯亂」
💡來自知名 SaaS 執行長對當前 AI 炒作週期的批判性觀點。
⚡ 30-Second TL;DR
有什麼變化
Aaron Levie 指出科技業執行長存在「AI 精神錯亂」
為什麼重要
此觀點為產業提供了一次現實檢視，鼓勵採取更務實的 AI 整合方式，而非僅是基於炒作的投資。
下一步行動
專注於可衡量的投資報酬率 (ROI) 與具體應用場景，而非投機性的 AI 導入，以避免陷入「精神錯亂」的陷阱。
誰應關注：Founders & Product Leaders
關鍵要點
- •Aaron Levie 指出科技業執行長存在「AI 精神錯亂」
- •批評對 AI 生產力提升的炒作式信仰
- •凸顯市場期望與現實之間的脫節
🧠 深度解析
Web-grounded analysis with 13 cited sources.
🔑 增強重點摘要
- •Aaron Levie attributes 'AI psychosis' to CEOs being too far removed from the practical implementation of AI, often only seeing 'happy path results' in demonstrations without understanding the extensive work required for sustainable, real-world deployment.
- •Industry data supports Levie's concerns, revealing that 42% of companies abandoned most AI initiatives in 2025 (up from 17% in the prior year), and over 80% of AI projects overall fail, which is double the failure rate of non-AI technology projects.
- •Levie advises CEOs to engage more deeply with AI tools to personally encounter the failures, edge cases, and downstream dependencies that are not apparent in polished demos, thereby fostering a more realistic appreciation for both AI's potential and the effort involved in its integration.
- •Box, under Levie's leadership, is actively developing and deploying its own suite of AI agent-based solutions, including Box Agent, Box AI Studio, Box Automate, and Box Extract, to automate complex content workflows and provide secure, context-based insights for enterprises.
- •Levie views the shift to AI agents as the beginning of a 5-to-10-year automation journey, emphasizing that these agents will move beyond simple chatbots to handle complex workflows and decision-making, with interoperability and managing token budgets emerging as critical enterprise challenges.
🛠️ 技術深入
- Box's AI strategy centers on 'AI agents' that leverage advanced reasoning models to understand natural language instructions and complete complex tasks across enterprise content.
- The 'Box Agent' employs an 'agentic loop' that includes upfront planning (breaking down goals), capability selection (choosing tools like searching or data extraction), agent-to-human collaboration (allowing user input and source preview), and autonomous execution (synthesizing information into deliverables).
- Box AI Studio allows administrators to create custom AI agents tailored to specific business rules, knowledge, and data sets without requiring code.
- These custom agents can be configured to use various foundation models, including Gemini, ChatGPT, or Claude, providing flexibility and choice for enterprises.
- The system is designed with enterprise security, governance, and permissions controls, ensuring agents only access authorized files and do not use customer data to train third-party large language models.
- Box Automate integrates with existing Box products (Box AI, Box Extract, Box Apps, Box Sign, Box Hubs, Box DocGen) and features a drag-and-drop builder for workflow creation.
🔮 前景展望AI analysis grounded in cited sources
The 'AI psychosis' phenomenon will lead to a significant re-evaluation of AI investment strategies in the short to medium term.
High failure rates of AI projects (42% abandoned in 2025, 80% overall failure rate) and the disconnect between CEO expectations and implementation reality will force a more pragmatic approach to AI adoption, focusing on 'last mile' reliability and measurable ROI.
The market for AI agent platforms and tools that address 'last mile' implementation challenges will see increased investment and demand.
Levie's critique highlights the need for robust governance, audit trails, and consistent task completion at scale for AI agents, suggesting that companies providing solutions for these challenges will be favored by investors and enterprises.
Enterprises will increasingly adopt a multi-model, interoperable AI strategy rather than relying on a single vendor.
Levie explicitly states that 'interoperability is key' and that enterprises are deploying multiple AI systems, wanting flexibility to swap models and agents to optimize costs and avoid vendor lock-in.
⏳ 時間線
2022-11
ChatGPT release sparks significant increase in AI companies and hype
2023-10
Aaron Levie advocates for 'church and state' separation between deterministic systems and probabilistic AI at TechCrunch Disrupt, citing agent failures
2024-10
Aaron Levie discusses Box's AI strategy, emphasizing AI agents as a 5-10 year journey of automation
2025-06
Aaron Levie discusses AI agents as augmentation and Box's AI Studio for custom agents
2025-09
Box announces agentic solutions like Box Extract, Box Automate, and AI capabilities for Box Apps at BoxWorks
2026-04
Box launches Box Agent and enhances Box AI Studio, enabling custom AI agents and leveraging models like Gemini, ChatGPT, or Claude
2026-05
Aaron Levie coins 'AI psychosis' to describe CEOs' disconnect from 'last mile of work' in AI implementation
📎 來源 (13)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源： TechCrunch AI ↗