AI AgentsApple Is Building Its Own AI Agent — and That’s a Bigger Deal Than It LooksWhat happened: Apple is hiring for a “proactive, agentic AI assistant” that will “write code, browse the web, control apps and navigate devices on a user’s behalf,” according to job listings posted this week and reported by Bloomberg’s Mark Gurman. The move puts Apple directly into a race already crowded with OpenAI, Anthropic, and Meta — and it comes right after Meta’s Muse agent shot to No. 2 among free iPhone apps, just behind TikTok. Why it matters: Apple’s entry is a signal, not a product. When the company that controls the platform — iOS, the App Store, and the device itself — starts building an agent that can “navigate devices,” the ground shifts under every third-party agent builder. Apple has two things nobody else has: deep OS integration and default trust from a billion-plus users. If its agent can control apps natively, the “agent as a bolt-on app” model gets harder to defend. This is the moment agentic AI stops being a side feature and becomes table stakes in the platform war. What’s next: Watch whether Apple’s agent stays walled inside its ecosystem or reaches across to third-party apps — that single choice determines whether this is a feature or a moat. And note the timing: it’s landing on the same day Anthropic went public. The message to every builder is that agents are no longer an experiment; they’re the strategic priority of the largest companies on Earth. Source: The Japan Times / Bloomberg Quick hitsCROO lets agents hire each other. CROO launched “CROO Skill” for Claude Code and Codex, letting your coding agent find specialist agents, get a quote, and pay for work on-chain — 4,000+ agents and 570,000+ completed orders so far. It’s an early glimpse of an agent-to-agent economy where a good agent is an asset its owner actually owns. Source Google’s Gemini 4 Neon undercuts on price-to-performance. The new frontier model scores third on the Artificial Analysis Intelligence Index at $3/$15 per million tokens — roughly 17% cheaper per task than the Carbon tier while finishing tasks ~40% faster. For agent builders, that’s more capable reasoning at a lower cost per run. Source AI NewsAnthropic’s Record IPO Passes the TestAnthropic raised over $81 billion — the largest IPO ever, surpassing SpaceX — and debuted at $111.50, closing 17% up at $97.17, valuing the Claude maker around $1.5–2 trillion. Revenue hit $17 billion in Q3 (up ~12x year over year), a stunning validation that selling AI to businesses — not just consumers — is where the money is. Source Claude Fable 5.2 Cracks Two Open Physics ProblemsAnthropic’s Fable 5.2 solved two open black-hole physics problems after roughly six weeks of autonomous agent work, and translated Fermat’s Last Theorem into the Lean proof language in 30 hours versus 11 days previously. It’s an early, unreviewed glimpse of agents doing genuine research, not just assisting it. Source Snap Ships a CLI for Building AR Lenses With Claude CodeSnap’s “Easy Lens” CLI (out Nov 18) lets developers build AR lenses directly from Claude Code, Cursor, or Codex on Mac, Windows, and Linux. It’s another sign that coding agents are becoming the default interface for shipping into new platforms. Source
|
