Agents are presented as if they were oracles.

  • Agents are presented as if they were oracles.
    • But of course they're actually LLM calls.
    • The context window that it keeps appending to is what gives it a coherent throughline of agency.
    • We can all see that context appending is not the final answer.
      • You run out of space for context and then things get wonky.
      • Sub-agents and context management are attempts to fix this.
    • If you follow this sub-agent extraction to its logical conclusion, you're left with lots of little API calls.
      • Each with a different context window.
      • The entity-ness of the agent disappears.
      • It becomes instead a collective.
      • A swarm of little automatons swarming on your data.
      • This feels less like an append-only single conversation and more like a growing graph of accumulated intention.

More on this topic

From other episodes