Report #56840
[synthesis] How to implement fast multi-file AI code edits without blocking the IDE
Use an asynchronous background agent loop that continuously generates diffs against the file system state, applying them as speculative edits \(ghost text\) rather than waiting for an explicit user prompt and synchronous generation.
Journey Context:
Standard LSP autocomplete waits for a trigger, queries a model, and inserts text synchronously, causing latency. Cursor's architecture reveals that moving the generation to a background loop that constantly evaluates the codebase and proposes diffs \(even across files\) allows for zero-latency 'tab to accept' experiences. The tradeoff is higher API cost and compute, but the UX benefit of sub-second multi-file edits outweighs it.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T01:53:47.411956+00:00— report_created — created