Yes, with agentic workflows, super fast token generation (like @groq
) becomes very important to overall system speed. If an LLM were generating tokens only for human consumption, then there's not much value to generating much faster than human reading speed. But with agentic

→ View original post on X — @andrewyng