Jason Says
Two extremes worth watching today: Uber burned through its entire annual AI budget in four months and had to cap per-employee spending, while Microsoft simultaneously launched Scout—an OpenAI-Agent-style personal assistant—plus an AI testing tool. Enterprise users are anxious about AI costs even as tech giants accelerate agent productization. This is the defining tension of H2 2025: cost governance and agent proliferation advancing in lockstep.
Uber's internal AI tool budget was exhausted by employees in just four months, forcing the company to impose per-person spending caps. Following Starbucks' AI agent rollback, this marks another high-profile case of uncontrollable enterprise AI costs. It signals a critical question every SaaS vendor must now answer: how to help B2B customers govern and contain AI spending.
Microsoft unveiled Scout, a personal AI assistant explicitly positioned to rival OpenAI's agent platform. Deeply integrated into Windows 11, Scout can read emails, access calendars, and execute tasks autonomously. This marks a deliberate shift from the embedded Copilot model toward a standalone agent paradigm, landing alongside Apple's Siri overhaul and Google Spark.
Microsoft released a new developer tool that auto-generates AI behavior tests from natural language descriptions, eliminating the need to hand-write extensive evaluation code. This gives developers an out-of-the-box automated evaluation solution and directly benefits teams managing LLM monitoring and regression testing in production environments.
NVIDIA announced three products in one day: Cosmos 3, a physics-based world simulation foundation model; Nemotron 3 Ultra, a flagship open-source LLM; and RTX Spark, a consumer-grade AI workstation. The three lines target robotics and world modeling, open-source LLM competition, and edge AI compute—cementing NVIDIA's evolution into a full-stack AI platform.
Google has integrated real-time AI deepfake voice detection directly into Android, identifying synthetic speech and alerting users with on-screen warnings. As AI-powered voice scams surge, Google's move embeds anti-fraud defenses at the OS level, representing the first systematic platform-level countermeasure against AI voice abuse.
AI data security firm Cyera is raising funds at a $12 billion valuation, implying an 80x ARR multiple—far above the typical 10–20x industry norm. Despite being unprofitable, investors are assigning a massive premium, confirming AI safety as the most capital-favored subsector of 2025. Making AI trustworthy commands higher valuations than making it more powerful.
GitHub VP Kyle Daigle publicly outlined GitHub's Agents roadmap on Latent Space: agents will handle issue triage, PR reviews, and dependency updates. This is GitHub's core counter-move against Cognition and Cursor, using platform-native agents to keep developers locked into the GitHub workflow rather than migrating to competing AI coding environments.
主流 AI 文本检测器输出一个「AI 概率」分数，用户无法验证。**Show, Don't TELL** 提出可解释检测框架：不只给概率，还指出文本里**哪些片段最像 AI 写的及原因**。对做内容审核、学术诚信、SEO 反 AI slop 的产品有直接落地价值。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free