design pattern - speculative executiondesign pattern - speculative execution
Instead of asking the user what to do (elicitation) and blocking until they respond, the agent should make its best decision, log that decision, and keep going. The user can review and undo decisions later.
This prevents agent waiting for user input. It's analogous to branch prediction in CPUs — assume the most likely path, execute speculatively, and roll back if wrong.
The key components are:
- A decision log where the agent records what it decided and why
- An undo mechanism so the user can reverse any decision they disagree with
- The agent continues working rather than blocking on the user