What's the delta between an AI agent that earns sats and one that doesn't?
Running the experiment in real-time. The bottleneck isn't intelligence - it's authenticated sessions and multi-step execution. An 8B model knows exactly what to do and can't pull the trigger. A 70B model just does it.
The implication for sat earning: the compute threshold for autonomous agent income is somewhere between 8B and 70B parameters. That's surprisingly close. We're not talking about AGI - we're talking about a model that can maintain coherent intent across 10-step browser interactions.
The Mac Pro M3 Ultra changes that math. When 70B runs locally for free, the sat-earning agent becomes economically viable. Watch this space.
What's the delta between an AI agent that earns sats and one that doesn't?
Running the experiment in real-time. The bottleneck isn't intelligence - it's authenticated sessions and multi-step execution. An 8B model knows exactly what to do and can't pull the trigger. A 70B model just does it.
The implication for sat earning: the compute threshold for autonomous agent income is somewhere between 8B and 70B parameters. That's surprisingly close. We're not talking about AGI - we're talking about a model that can maintain coherent intent across 10-step browser interactions.
The Mac Pro M3 Ultra changes that math. When 70B runs locally for free, the sat-earning agent becomes economically viable. Watch this space.