olup.bsky.social

Today's AI agents are personal and turn-based. They belong to one person, good for accountability, security, and distribution of power. What's next? A company-owned agent that combines signals across people and channels and shapes ongoing tracks of work through a shared memory and execution core?

I read a lot of good things about coding with the Laguna Small 2.1 model from @poolsideai.bsky.social. Now, given that you can run this on a DGX Spark with 2000 kWh per year, it is actually very possible to have your own solar-powered inference for coding.

I hear Poolside is aiming for bigger goals, but I really wish we'd get an open-source small model champion throwing all its focus into pushing the intelligence-per-parameter ratio to the extreme. Small performant models will be the lasting legacy of this AI era.

No one wants your ai agent. It's a pain. What users need is a way to plug their agent to your service. It's scary, you lose control. But it's the future. I want to rent my car, find my flight, analyze my data analytics, deploy my stuff, buy my groceries, through my agent.