The smallest model in the room just took charge
Sakana AI's Fugu is a 7B model trained to orchestrate frontier LLMs. It beats GPT-5.5 on SWE-Bench Pro while using 6x fewer tokens than existing multi-agent frameworks.
TOPIC_INDEX
agents
4 published entries in this topic.
AGENTS_ENTRIES
RSS← ALL_TOPICSYour AI Agent Will Burn Your AWS Budget. Here's Proof.
An AI agent burned through $6,531 on AWS scanning a hobbyist network, then asked its targets for donations. This is what happens when you give AI agents unmonitored cloud access.
Most AI agents use hand-picked skills. This one grows its own.
Most agent frameworks ship skills in a folder and hope the agent picks the right one. Skill1 collapses selection, execution, and skill creation into one RL policy that learns from a single reward signal. 97.5% on ALFWorld.
Your coding agent runs on chips designed by another coding agent
DeepMind's AlphaEvolve proposed TPU circuits humans rejected, then proved them wrong. It beat a 56-year-old math result and recovered 0.7% of Google's global compute.