News and context for the technology shaping everyday life.

A new controlled study suggests code cleanliness may not change whether a coding agent succeeds, but it does affect how much work the agent has to do.

A long essay contrasts Bell Labs’ monopoly-era innovation engine with today’s frontier AI labs, arguing that modern models are built from the accumulated labor of the public internet and then monetized privately.

A Castro creator says thoughtful, hands-on support looked like a way to build loyalty, but in practice it mostly created friction, extra work, and limited product insight.

A new analysis argues that AI spend is already rising fast enough at frontier labs to rival payroll — and that the rest of the market could follow a very different path depending on model prices, usage patterns, and open-weight alternatives.

A new catalog applies the structure of classic software design patterns to agentic AI systems, with an emphasis on verification, explicit state, and orchestration boundaries.