Have the brakes finally been applied in the AI race? 

This week, the AI industry received a major dose of realism from OpenAI co-founder Andrej Karpathy. Speaking on the race towards artificial general intelligence (AGI), Karpathy said we’re still a decade away from seeing it work as intended, calling today’s AI agents ‘slop’. 

For businesses that have held back on their AI investments, fearing that AGI might soon render their efforts obsolete, Karpathy’s comments flip that logic on its head. If AGI is ten years out, the excuse for waiting disappears. 

As host Magnus Oxenwaldt puts it: “The question isn’t when AI will be ready. It’s whether you’re using the AI that’s ready right now, or getting left behind waiting for something that won’t arrive until 2035.”  

Top stories this week 

• Claude adds long-term memory, enabling models to recall user projects and preferences across sessions. 

• Google’s Willow chip achieves quantum advantage, running algorithms 13,000× faster than the top supercomputer. 

• Anthropic signs a $40 billion TPU deal with Google, securing compute capacity for the AI decade. 

• Oracle launches an AI agent marketplace, while Microsoft opens an Agent Store. 

• DeepSeek unveils OCR models compressing context windows by 10×, and GPT-5 sets new SWE-Bench and AIME benchmarks.