What's this about?
Mollick says prose becomes less natural during long-running tasks even when tool use remains effective. The post supplies a useful hypothesis, not a measured rate or causal explanation.
What exactly is output drift?
Style or behaviour gradually moves away from the requested standard as context accumulates. Checkpoints, fresh review passes and explicit rubrics can expose drift before final delivery.
Why does this matter?
X post had 213 likes at fetch time; no controlled study attached.
- Long task success should include communication quality, not only completion.
- Context compaction and repeated self-generated text may be contributing factors.
- Teams can sample intermediate outputs instead of trusting only the final result.