Stanford AI Research | Microsoft AI Agent Coworkers | Workday AI Bias Lawsuit | Military AI Goes Big

Episode notes

Happy Friday, everyone! This week I’m back to my usual four updates, and while they may seem disconnected on the surface, you’ll see some bigger threads running through them all. All seem to indicate we’re outsourcing to AI faster than we can supervise, are layering automation on top of bias without addressing the root issues, and letting convenience override discernment in places that carry life-or-death stakes. With that, let’s get into it. ⸻ Stanford’s AI Therapy Study Shows We’re Automating Harm New research from Stanford tested how today’s top LLMs are handling crisis counseling, and the results are disturbing. From stigmatizing mental illness to recommending dangerous actions in crisis scenarios, these AI therapists aren’t just “not ready”… they are making things worse. I walk through what the study got right, where even its limitations point to deeper risk, and why human experience shouldn’t be replaced by synthetic empathy. ⸻ Microsoft Says You’ll Be Training AI Agents Soon, Like It or Not In Microsoft’s new 2025 Work Trend Index, 41% of leaders say they expect their teams to be training AI agents in the next five years. And 36% believe they’ll be managing them.

• Stopping the Agent Sprawl: Why Cranking The Dial on Autonomy is Financial and Operational Suicide65 / 100
• The Rise of Tokenmaxxing: Analog Organizational Risks Just Got an Expensive AI Upgrade61 / 100
• (Special Episode) AI Agent Sprawl, Lost Agency, & Rethinking Resistance with Travis Hahler
• Vibe-Coded Catastrophes: Accelerating Intimacy in Digital Apps is Your Security Nightmare
• Fortifying Organizational Fragility (Part 3): Business Physics and the Frictionless Fallacy