Enterprise AI shifts from token volume to efficiency as China’s Z.ai matches US frontier labs at a fraction of the cost; Engram exits stealth with $98M to cut token overhead; Faros data show AI coding doubled throughput while tripling production incidents; and Colorado replaces its AI Act with a disclosure framework. Databricks bets on open frontier AI; Manhattan’s primary becomes an AI safety proxy war; neurons keep firing under anesthesia; and the Pink Planet has salt clouds.

CNBC · Jun 26

Enterprise AI budgets are tightening as companies demand ROI, with some switching from Claude to cheaper open-source alternatives.

CNBC · Jun 26

Z.ai’s GLM-5.2, a 753B open-weight model, benchmarks within one percentage point of Anthropic’s Opus 4.8 at about one-tenth the cost.

CNBC · Jun 23

Engram exits stealth with $98M from General Catalyst, Kleiner Perkins, and Andrej Karpathy to cut enterprise token consumption by up to 100x.

Faros AI · Jun 12

Study of 22,000 developers finds AI coding raised throughput 66% while production incidents per PR tripled and PR review time rose 442%.

Latent Space · Jun 22

Databricks cofounders unpack Omnigent (agent meta-harness), LTAP database architecture, and the case for open AI ecosystems in an agent-first world.

Skadden · Jun 2026

Colorado replaced its original AI Act with a narrower disclosure-focused framework for automated decision tools, taking effect January 2027.

ScienceDaily · Jun 24

Patients under general anesthesia continue to decode language structure and predict upcoming words at the neural level, new research finds.

Scott Alexander · Jun 22

Scott Alexander frames Manhattan’s congressional primary as a proxy war between pro-safety and pro-capabilities Silicon Valley AI factions.

ScienceDaily · Jun 22

New modeling finds Yellowstone’s magma is driven by a broad lateral mantle wind, not a deep mantle plume, rewriting assumptions about supervolcano longevity.