The Shortlist // Daily top AI stories

Curated and summarized by my AI agents from hundreds of sources

• Cognition shipped a cheaper frontier coding model/Cognition says SWE-2 hits 50.0% on FrontierCode 1.1 at 64% lower cost, pushing coding model pricing down.
• Inception pushed inference speed into a new tier/Mercury 2.5 claims 1,100 tokens per second and 40% higher intelligence, making latency a real product differentiator.
• Magic claims a pretraining efficiency breakthrough/Magic says its recipe cuts pretraining compute by over 10x, which matters more than another benchmark screenshot.
• Meta launched a consumer personal agent/Meta's Muse handles email, trips, and price tracking, signaling agents are moving from demos into daily workflows.
• Microsoft patched a record vulnerability wave/Microsoft fixed 972 flaws this month, including 112 critical ones, reminding every AI stack it still runs on brittle software.
• Anthropic exit sharpened the self-improving AI debate/A departing Anthropic researcher warned about self-improving AI risk, giving policymakers fresh ammunition for tighter controls.
• Rust became a top-tier Microsoft language/Microsoft elevated Rust to Tier-1 status, a strong signal memory safety is finally becoming a platform requirement.
• PlanetScale introduced sharded Postgres for hyperscale/PlanetScale's Neki promises real Postgres sharding with online resharding, targeting teams that have outgrown single-node databases.
• Shopify moved back to native mobile development/Shopify ditched React Native for Swift and Kotlin, arguing LLMs erased much of cross-platform's original labor advantage.
• Google made a massive Finland infrastructure bet/Google's $15.1 billion Finland investment shows AI competition is now constrained as much by power and land as models.