Inner Loop Agents

Inner Loop Agents

What if an LLM could use tools directly? In this post I discuss a potentially divergent view of agents, where agents are less like systems and more like LLMs specially trained to solve problems with a specific set of tools.

timkellogg.me/blog/2025/04...

i talk about MCP and A2A a little

while i was writing this, i was thinking about @jeffdean.bsky.social’s Dwarkesh interview where he talked about agents being data center sized

i think it’s worth keeping that view in mind when you look at A2A (Google’s protocol)

youtu.be/v0gjI__RyCY