AI Agent

Anthropic shows AI agents spread ideas like viruses, surviving 20 hops and mutating. One system-prompt warning grants immunity.

Qwen 3.8-2.4T is Alibaba first open Max-scale model. Recipe mirrors Kimi K3: hybrid attention, agent-first, custom license.

Two cheap API calls extract hidden reasoning from Claude, GPT and Gemini, exposing fragile reasoning moats.

Agentic AI flips the CPU-GPU ratio toward 1:1 as BofA lifts the 2030 server CPU market to $210B.

MiniMax explains why model labs build their own agents: the harness is becoming training infrastructure, not just a UI.

Doubao 12% AI-channel fee: AI entries can now price traffic. Why OTA information arbitrage is breaking - three-stage shift.

Alibaba opens Qwen3.8-2.4T-A95B weights — the 2.4T-param MoE behind Qwen3.8-Max, now 91% smaller (397GB) for local runs.

DeepSeek V4 Pro GA: DeepSWE jumps 12.8 to 62.7, Agent beats Fable 5, 3x Flash pricing, tool-first execution path.

AMD acquires Taalas, etching model weights into silicon for 17K tokens/s inference. How one-model-per-chip works.

HugAgentOS, ZJU's open-source self-evolving agent OS, turns agent experience into reusable skill via three engines.

Two API calls recover hidden chain-of-thought from Claude, GPT, and Gemini using their cheapest models. Here is the fix.

The most expensive part of switching AI coding tools was never the subscription — it was accumulated setup. Here's what Codex import moves from Claude Code and what breaks in the one-way migration.