上世纪九十年代，一部名为《高智能方程式》的日本动漫在国内颇为流行。根据故事设定， 男主角风见隼人驾驶着其父亲设计的赛车"阿斯拉达"（ASURADA），谱写了一段高智能方 程式史上最年轻冠军的传奇。 在这部动画中，最让笔者印象深刻的是这辆名为"阿斯拉达"的赛车拥有一个同名的高性能电脑，它 能将赛车直接拟人化，与男主角分享汽车的状态、马路的状态，甚至还能在驾驶者碰到困扰的时 候，扮演一个知心好友。 "阿斯拉达"首次亮相 来源：高智能方程式 在当时，如果说想拥有一个"阿斯拉达"，大多数人可能会觉得异想天开。但在三十年后的今天，我 们也许很快能迎来属于自己的"阿斯拉达"。这一切，主要感谢计算芯片和大模型技术的进步。 联发科引领Agentic AI重塑座舱 过去两年，关于大模型进入智能座舱，已经有了很多讨论。在半导体行业观察发布的文章 《 AI 座 舱芯片大变局，联发科上演技术式超车》 中，我们就对此进行了介绍。文章中，我们还深入探讨 了智能手机芯片巨头如何通过提供领先的芯片，让大模型更好地赋能汽车。不过，在智能座舱方 面，过去的大模型上车，更多的是类似ChatGPT或豆包的被动问答式的语音助手。 但在Agentic ...

Core Viewpoint - C3.ai, Inc. has experienced a significant decline in stock value, underperforming key industry benchmarks, and facing challenges in sustaining recent performance levels due to economic uncertainties and operational losses [1][3]. Group 1: Stock Performance - C3.ai shares have dropped 40.8% year to date, compared to a 14.9% decline in the Zacks Computer & Technology sector and a 17% decrease in the Zacks Computers - IT Services industry [3]. - The current stock price reflects a 54.8% discount from its 52-week high of $45.08 and a 19.7% premium to its 52-week low of $17.03 [4]. Group 2: Financial Estimates - The Zacks Consensus Estimate for C3.ai's fiscal 2025 and 2026 loss per share has improved to 45 cents (from 62 cents) and 46 cents (from 55 cents), respectively, indicating a positive shift in analysts' sentiment [7]. - For fiscal 2025 and 2026, the sales growth estimates are 29.7% and 22.4%, respectively [8]. Group 3: Strategic Partnerships - C3.ai's partnerships with major tech companies like Microsoft and Amazon are crucial for revenue generation, with 71% of fiscal third-quarter agreements facilitated through partner engagements [9][11]. - The collaboration with Microsoft has led to 28 new deals across nine industries, with sales cycles shortened by approximately 20% [9]. - The partnership with Amazon's AWS focuses on delivering advanced enterprise AI solutions, while a new relationship with McKinsey's QuantumBlack aims to combine strategic consulting with C3.ai's technology [10]. Group 4: Revenue Growth - C3.ai reported total revenues of $98.8 million for the third quarter of fiscal 2025, marking a 26% year-over-year increase, with subscription revenue growing 22% to $85.7 million [15]. - Revenue from software demonstration licenses reached $28.6 million, significantly contributing to overall revenue [15]. Group 5: Valuation and Investment Outlook - C3.ai is trading at a slight premium relative to its industry but at a discount to historical metrics, with a forward 12-month price-to-sales (P/S) ratio of 5.71X, compared to the sector's 5.39X [16]. - The current stock decline presents an attractive buying opportunity, supported by strong fundamentals, strategic partnerships, and leadership in Generative and Agentic AI [18].

SAN FRANCISCO, April 24, 2025 (GLOBE NEWSWIRE) -- Redis, the world's fastest data platform, and UiPath (NYSE:PATH), a leading enterprise automation and AI software company, today expanded their collaboration toward furthering agentic automation solutions for customers. Redis, a UiPath Technology Alliance partner, has played a crucial role in increasing the speed and efficiency of UiPath's Automation Suite. Redis powers UiPath Orchestrator's high-availability add-on (HAA), an on-premises offering that improv ...

Harvey 绝对是法律场景落地最成功的 AI 企业了。 成立于 2022 年，客户数量从 2023 年的 40 家增长到 2024 年的 235 家，遍布 42 个国家；在美国《法律周刊》评选的前 100 家律所中，有 28 家正在使用 Harvey。 2024 年 ARR 达到 5000 万美元，今年年初预计 8 个月内将达到 1 亿美元，2 月份拿到了红杉资本领投的 3 亿美元 D 轮融资，公司估值达到 30 亿美元 。 简单说的话， Harvey 现在的收入与 AI 搜索当红炸子鸡 Perplexity 相当。 在前不久 福布斯发布的 2025 AI 50 榜单 中，Harvey 是法律领域上榜的为数不多的公司之一。Harvey 目前已经能够实现自动处理从文件审查到客户沟通 的整个法律流程，几乎能替代一整个初级律师团队。 近段时间，Harvey 的创始人兼 CEO Winston Weinberg 以及产品负责人 Aatish Nayak 接受了多家播客节目的访谈，在访谈节目中，两人详细地分享了包 括 Harvey 的顶层战略方向设计、法律类 AI 产品如何实现商业化落地、法律类 Agentic wo ...

Financial Data and Key Metrics Changes - Total revenue for Q1 2025 was $263 million, up 3% year-over-year [36] - Cloud revenue increased 21% to $94 million, while services revenue declined 8% to $121 million [37] - RPO ended the quarter at approximately $1.9 billion, up 25% year-over-year and 6% sequentially [38] - Adjusted operating profit was $91 million with an adjusted operating margin of 34.7%, up over 340 basis points year-over-year [40] - Adjusted earnings per share (EPS) for Q1 was $1.19, up 16%, while GAAP EPS was 85 cents, down 1% [41] Business Line Data and Key Metrics Changes - The company experienced a 25% year-over-year increase in RPOs, driven by strong demand for mission-critical solutions [22] - Approximately 50% of new cloud bookings in Q1 were generated from net new logos, indicating strong demand for products [71] - Services team completed over 100 go-lives for customers in Q1, showcasing operational effectiveness [25] Market Data and Key Metrics Changes - The addressable market is forecasted to grow at a double-digit CAGR for the next several years, indicating robust market potential [15] - Competitive win rates remained consistent at about 70%, reflecting strong market positioning [24] - The company has established healthy footprints across diverse sectors including retail, grocery, life sciences, and technology [23] Company Strategy and Development Direction - The company is focused on organic innovation and capital allocation strategy to expand its addressable market [14] - Investment in sales and marketing is prioritized to drive growth and capitalize on new product offerings [64] - The launch of new products like Enterprise Promise and Fulfill aims to optimize B2B order fulfillment, addressing evolving customer needs [26] Management's Comments on Operating Environment and Future Outlook - Management expressed caution regarding near-term services revenue growth due to macroeconomic uncertainties [21] - The company remains optimistic about its position in the market and long-term growth opportunities despite current challenges [20] - Management reiterated full-year guidance for RPO and total revenue, reflecting confidence in business fundamentals [45] Other Important Information - The company was named Google’s Cloud Business Applications Partner of the Year for supply chain and logistics, highlighting its innovation in the Google Cloud ecosystem [18] - The company ended the quarter with $206 million in cash and zero debt, indicating a strong balance sheet [42] Q&A Session Summary Question: Insights on cloud bookings and RPO dynamics - Management indicated strong pipeline in Q2 and confidence in guidance despite macro uncertainties [59][60] Question: Growth investments and monetization of products - Management plans to invest in sales specialists to drive growth and maintain high win rates [64] Question: Linear progression of bookings and sales activity - Management noted a balanced product portfolio and strong demand across verticals, with high expectations for Q2 [68][71] Question: Resilience of cloud bookings drivers - Management stated that no particular segment is more resilient, but all channels remain important for growth [78][80] Question: Visibility on multi-year ramps and growth trajectory - Management confirmed strong visibility into contract durations and ramp processes, supporting confidence in growth [95][97] Question: FX impact on guidance - Management indicated that FX swings represent less than 1% impact on revenue guidance [104][105] Question: Strength of large deals in the pipeline - Management reported favorable conditions in the pipeline, with confidence in closure rates compared to last year [111][112]

但 LangChain 创始人 Harrison Chase 对于 OpenAI 在文中的一些观点持有异议，尤其是「通过 LLMs 来主导 Agent」的路线，迅速发表了一篇长文回 应。 Harrison Chase 认为，并非要通过严格的「二元论」来区分 Agent，目前我们看到大多数的「Agentic 系统」都是 Workflows 和 Agents 的结合。理想 的 Agent 框架应该允许从「结构化工作流」逐步过渡到「由模型驱动」，并在两者之间灵活切换。 相比 OpenAI 的文章，Harrison Chase 更认同 Anthropic 此前发布的如何构建高效 Agents 的文章，对于 Agent 的定义，Anthropic 提出了「Agentic 系 统」的概念，并且把 Workflows 和 Agents 都看作是其不同表现形式。 总的来说， 这是大模型派（Big Model）和工作流派（Big Workflow）的又一次争锋， 前者认为每次模型升级都可能让精心设计的工作流瞬间过 时，这种「苦涩的教训」让他们更倾向于构建通用型、结构最少的智能体系统。而以 LangGraph 为代表的后者，强调 ...

Investment Rating - The report does not explicitly provide an investment rating for the industry Core Insights - The clinical research arena is experiencing transformative advancements due to the successful application of Generative AI (GenAI) tools, enhancing efficiency and quality in clinical trials [4][6] - Regulatory agencies, including the FDA, are beginning to establish guidelines for the responsible use of AI in clinical studies [6] - The report emphasizes a multi-pronged approach to safeguard efficiency and quality in clinical trials through various AI methodologies [19] Overview - The report highlights the increasing anticipation among industry professionals regarding the potential of AI to improve clinical trials and healthcare [4] - There are numerous opportunities to leverage AI technologies across the clinical trial ecosystem, including design, patient engagement, and regulatory submissions [5] AI Methodologies - Distinction is made between Generative AI, which generates responses based on training data, and agentic AI, which independently handles complex problems [10][11] - A holistic approach is necessary for developing AI frameworks, emphasizing the importance of training, ethical considerations, and human oversight [12][13] Safeguards for AI in Clinical Trials - Five critical categories of safeguards are identified to ensure the safe and efficient use of AI in clinical studies: curating and containerizing data, integrating "human-in-the-loop," harmonization of response, objectivity, and recognizing uncertainty [19][20] - Curating training data is essential to avoid poor-quality responses and ensure reliability in clinical operations [24][25] - The integration of human oversight is crucial to optimize quality and prevent erroneous outputs from AI systems [26][29] Use Cases of AI in Clinical Trials - The report discusses successful applications of AI, including a scientific Q&A chatbot used in a Phase III trial, which improved the efficiency of protocol clarifications and reduced the burden on medical monitors [39][40] - The chatbot's success was attributed to rigorous training, harmonized responses, and the ability to recognize knowledge gaps [41][42]

作者：yongxin 编辑：Siqi Agentic AI 的 3 要素是：tool use，memory 和 context，围绕这三个场景会出现 agent-native Infra 的 机会。 01 . 为什么 Search API 很重要 按照场景和信息需求类型，搜索行为大致可以被分为四类： • 第一类，高频快速查询， 指的是一两步内就能完成的查询。Google 大部分的 query 还是以几个单词为 主，用户得到答案后马上离开，不会进行深入的查询。对于这类查询 Google、Bing 还是最好的应用，新 玩家几乎没有挑战的机会。 • 第二类，研究性质的深入查询， 用户可以和搜索工具反复交流，获取知识。这一类搜索是 LLM 和 LRM 带来的新场景，对应的代表性产品形态分别是 Chatbot 和 Deep research。 Agent 所获取到的信息质量是 agent 推理的起点，虽然 LLM 带来了 perplexity 为代表的 AI answer engine，提供了完全不同于传统搜索引擎的体验，但这些产品仍旧面向的是人类用户，产品逻辑是围 绕人类行为设计的。 在我们 MCP 的研究中发现， ...

随着我们迈入 Agentic 时代，对于各组织机构而言，若要对模型进行扩展以实现高效推理，他们将需要 在从训练到推理的每一个步骤中都遵循扩展流程。在 NVIDIA GTC 2025 上，黄仁勋的愿景以及所发布 的消息聚焦于在从企业信息技术、云计算到机器人技术等各个行业中构建 " AI工厂"。 为了让AI工厂取得成功，NVIDIA持续创新，并提供完整的AI技术栈，包括芯片、系统和软件，以最高 的效率来加速和扩展AI。该公司的方法涵盖了Agentic AI 和 Physical AI 领域。NVIDIA在其整个技术栈 方面做出了以下发布内容： 图片来源：NVIDIA & Counterpoint Research 芯片方面：从计算路线图到硅光子学领域 都有重大消息发布 图片来源：NVIDIA NVIDIA 的芯片产品组合涵盖了中央处理器（CPU）、图形处理器（GPU）以及网络设备（用于纵 向扩展和横向扩展）。 NVIDIA 发布了其最新的 " Blackwell超级AI工厂" 平台 GB300 NVL72，与 GB200 NVL72 相比，其 AI性能提升了 1.5 倍。 NVIDIA 分享了其芯片路线图，这样一 ...

2025年，AI产业化元年已经是箭在弦上，只欠关键场景的"东风"。 iTerms Pro，一个为法务"真干活"的AI智能体。 从越来越多的AI应用自主接入DeepSeek，到Agentic AI（智能体）的崛起，技术和产品形态狂飙，AI 平权化时代来临，似乎也将各行各业的智能化转型推 向了临界点。 在法律 科技 行业，国内头部电子签名产品与解决方案提供商法大大，也以十年积淀给出了自己的答案—— 基于自研法律大模型，发布法务AI智能体产品 iTerms Pro。 深究AI产业化的本质，不难发现：AI的价值不在于炫技，而在于让技术穿透专业场景， 尤其是容错率极低的法务岗位，好的法务AI产品需要做的是 放大专 业价值，从而实现人机协同的效应。 所以，让AI产品化不难，难的是让AI应用帮法务"真干活"。 合规挑战下，智能体"破局" "法务数字化在中大型客户群体中已经落地了很久，但人与AI协同去驱动业务流转的愿景，却还有很长的路要走。" 法大大产品负责人梅容告诉36氪，目前国内 大概有 一半以上的企业都已经完成了合同的数字化管理。 但与此同时，其中的大多数仍困在"流程线上化"的浅水区，大量的非标合同依赖人工审查和驱动， ...