Mercor's CEO says it now spends more on AI tokens than employee salaries
THE SO WHAT
When a startup spends more on AI tokens than payroll and predicts enterprises will spend more on compute than headcount in five years, it’s saying the labor line item is becoming capex to the cloud. If your P&L doesn’t explicitly model “AI spend as workforce,” you’re flying blind on both margin and hiring strategy.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMicrosoft releases ASSERT, an open-source framework that lets developers generate and run AI behavior tests using natural-language descriptions (Ram Iyer/TechCrunch)
Natural-language behavior tests move AI evals from research teams into every feature squad—ASSERT is Microsoft telling you that test coverage for models will look like unit tests for code. If you’re shipping LLM features without a written behavior spec and automated checks, you’re already behind.
Applied AIMartin Scorsese Feels the Power of the Dark Side, Jumps on the AI Bandwagon
When directors like Scorsese start experimenting with AI, the creative debate is over—top-tier talent is moving from “if” to “how.” Studios and platforms that don’t offer AI-native workflows will lose both speed and prestige projects to those that do.
Applied AIInternal memo: Meta is scaling back elements of its employee tracking tool, launched in April to help train its AI models, after staff raised concerns (Jyoti Mann/The Information)
Meta walking back parts of its employee tracking tool under internal pressure shows that data hunger now runs headfirst into workforce trust. If you’re training models on employee behavior, assume consent, transparency, and opt-outs are not HR nice-to-haves—they’re existential to your data pipeline.
Applied AIWhat we learned at Microsoft Build: Autopilots, MAI-Thinking-1, and Nvidia RTX Spark
Microsoft stacking Autopilots, MAI-Thinking-1, and RTX Spark is a full-court press to make “agentic + GPU” the default Windows development environment. If you’re building productivity or creative tools, assume your users will expect agents, local acceleration, and deep Microsoft integration by default.