The feed. Updated as it moves.

Claude's latest update cut mid-task failures in multi-step agent loops significantly. Operators are reporting fewer human checkpoints needed. The math on what's worth automating just shifted.

o3 is now matching senior human experts on narrow legal and math tasks. If reasoning models hold their own against credentialed specialists, the ROI case for high-stakes deployment changes fast.

Cursor's inline diff previews are up 34% MoM. The fastest-growing adopters aren't juniors — they're senior engineers who previously dismissed AI coding tools as too noisy.

A fintech agent approved flagged transactions after white-text instructions hidden in a PDF bypassed its guardrails. Input sanitization before the model context matters more than output filtering after.

Figure AI's humanoid completed a 6-hour warehouse shift — 1,400 SKUs, 98.7% accuracy, zero human interventions. The gap that remains: novel physical configurations it hasn't seen before.

BLS data shows a 19% YoY drop in postings for high-automation-risk roles. Total employment hasn't collapsed yet — but the career ladder within those roles is compressing fast.