AI autocomplete is the inline-prediction modality popularised by GitHub Copilot and Tabnine in code, and now common in writing tools (Notion AI, Grammarly) and email (Gmail Smart Compose). The interaction model is that the cursor is followed by gray suggested text; pressing Tab accepts it.
Autocomplete is fast, cheap, and unintrusive, which is why it scales: a typical engineer accepts 20-40% of Copilot suggestions and accepts 5-15% of suggested lines in a typical day. The model needs low latency (under 200ms ideally) and good ranking; quality of the autocomplete is mostly about ranking the right next line for the cursor context.