This piece is ignorant at best. It hinges on the argument that current models are too large, and too expensive. It assumes nothing will change with regards to 1

This piece is ignorant at best. It hinges on the argument that current models are too large, and too expensive. It assumes nothing will change with regards to 1) the intelligence of smaller, open source models and 2) the price of larger, more capable models. They are very clearly trending towards "intelligence too cheap to meter", as smaller research labs release highly capable local models (see: ModernBERT for small models, DeepSeek v3 for large models) and larger companies decimate each others margins with continual price decreases (see: the amount of large models that can be accessed for free on OpenRouter). As we build larger, more capable models, and distill them into smaller, more affordable models, intelligence will only grow in the long term. At that point, creating intelligent agents that operate within your environment is an engineering problem, not a research one.