Field Notes
Why most AI products feel slow
Perceived latency in AI products is often a product design problem as much as a model performance problem.
Many AI products feel slow even when the underlying model is reasonably fast. The problem is often not raw response time. It is ambiguity in the interaction model.
Users wait without knowing what the system is doing, what is deterministic, or whether the result will actually be useful. That uncertainty amplifies every second of latency. A product that frames progress clearly, narrows the task well, and exposes intermediate structure can feel meaningfully faster without changing the model.
In practice, speed comes from:
- tighter request shaping
- predictable response formats
- fewer dead-end interaction states
- feedback that reflects real progress instead of generic loading theater
Performance matters, but so does interface honesty. The best AI products reduce both compute latency and cognitive latency.