AI digest: loops, chips, and cyber warnings
Agentic AI goes fully autonomous, Groq bounces back with $650M, and Five Eyes intelligence agencies say offensive AI threats are months away.
Big week for infrastructure and agents. The direction of travel is becoming very clear: more autonomous, more hardware-hungry, and more concerning to governments.
The AI world is going loopy, and that’s not a joke
The latest shift in agentic AI is the move to “loops”, where swarms of agents run continuously in the background without waiting for a human prompt. This is a meaningful step beyond one-shot agents. The question of how you supervise something that never stops running is one nobody has properly answered yet. Read the full piece on TechCrunch.
Five Eyes says offensive AI threats are months away
The Five Eyes intelligence alliance has warned that frontier AI models capable of disrupting governments and businesses are only months from being viable weapons. That is not a vague long-term risk statement, that is a near-term operational warning from the agencies who should know. Worth taking seriously. More from The Decoder.
Groq raises $650M after Nvidia’s not-acqui-hire
Groq confirmed a $650M raise and is rebuilding its exec team after Nvidia’s unusual deal that took its talent without a full acquisition. The company is doubling down on its neocloud business, which is interesting given how crowded that space is getting. Good sign that investors still see an independent Groq as worth backing. TechCrunch has the details.
Vibecoding is now a due diligence tool
Bain & Company is using AI to vibe-code replicas of acquisition targets to test how defensible their software actually is. If an AI can rebuild your product in hours, your moat is not what you thought it was. This is going to change how SaaS companies are valued. The Decoder covers it here.
Sakana AI’s Fugu coordinates multiple LLMs to match frontier benchmarks
Japanese lab Sakana AI released Fugu, a system that orchestrates several models together to reach performance levels comparable to Anthropic’s top offerings. The angle of reducing dependence on any single provider is smart. Multi-model orchestration as a strategy rather than a workaround is worth watching. Full story at The Decoder.