Link Instacart, Canva, or YouTube Music and AI Mode fills a cart, drafts a design brief, or builds a playlist straight from a query. The search box is turning into an agent, and every commerce and media app now needs a Search integration the way sites once needed a share button. TechCrunch →

The inference provider closed a $1.505 billion Series D on July 16 and reports over $1 billion in annualized revenue, up 5x year over year, per its own announcement. Inference is the new cloud layer: every LLM app needs a latency-optimized backend the foundation labs do not sell. The announcement →

City Attorney David Chiu issued cease-and-desist orders on July 17 aimed at the app stores, not just the developers. The theory makes platform operators liable for AI-generated content they distribute, and California's SB 942 takes effect August 2. TechCrunch →

Leaked code shows 113,879 hours pulled from YouTube Music, plus thousands more from Deezer and Genius, without authorization. Unlike a sources-tell-us copyright suit, this is documented code, and the same breach exposed paying customers' data. 404 Media →

The Google DeepMind CEO published an essay stating AGI is probably only a few short years away and called for a Western-led international coalition to vet frontier models before deployment. The call landed days after China launched WAICO with 29 founding nations: China now has a concrete organization; the West's leading AI researcher is still calling for one to be formed. Coverage on ExplainX →

LM Studio Bionic runs a free local coding agent on your own hardware. We compared it against Cursor and Cline: it wins on cost and privacy, and gives ground on large-context refactors.

Kimi K3 tops agentic coding benchmarks per Moonshot AI and undercuts Fable 5 on cache-heavy loops. Here is the pricing math on which teams should switch before the July 27 weights drop.

ClinePass costs $9.99 a month for open-weight coding models. We compared BYOK, ClinePass, and Enterprise token costs so you land on the right tier instead of overpaying.

98 percent of assessed production agents mix private data, untrusted input, and outbound actions in one session, per Adversa AI. We break down the GuardFall exploit and the 6-point checklist that closes the hole.