Why I Stopped Letting One Turn Give an Agent Two Orders
Why I Stopped Letting One Turn Give an Agent Two Orders I used to think duplicate delegation was a small prompting problem. If an agent received two instructions in one turn, surely it could infer which one mattered most. That is a comforting assumption. It is also a poor runtime rule. While building APX, I found a more honest default: one turn gets one instruction per target agent. A second…
One turn should only command one action from an agent, according to the author's experience building APX. Duplicate delegation can lead to confusion and make the system harder to audit. When an agent receives two instructions in a single turn, it becomes difficult to determine which one is the correct action. The author implemented a guard in APX that records the target agent and the instruction for each turn.
If a second instruction is sent to the same agent during that turn, the system will refuse it and provide a clear error message. This guard is deliberately small and does not attempt to infer the agent's intent from ambiguous instructions. The author also introduced a follow-up flag that allows callers to explicitly correct or extend an earlier instruction, but this flag should only be used when the new message deliberately modifies the previous one.
The author acknowledges that this rule does not solve all planning issues, such as overlapping work between agents or the need for confirmations and permission boundaries for consequential actions. However, it does prevent a common error of the same turn giving the same agent multiple tasks. By enforcing this rule, APX creates a more transparent and trustworthy workflow, where handoffs are visible and contradictions are resolved through explicit correction rather than hidden precedence.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
