AI IDEs go agentic: Cursor "demos" and Windsurf Cascade
AI IDEs are shifting from code suggestions to autonomous agents that run, test, and showcase changes, led by Cursor’s new demo-first experience and Windsurf’s Cascade engine. Cursor now emphasizes "demos, not diffs," with agents that can run the software they build and send video evidence of their changes ([YouTube](https://www.youtube.com/watch?v=XbZvC4KTH68&pp=ygURQ3Vyc29yIElERSB1cGRhdGU%3D)). Meanwhile, Windsurf’s agentic Cascade engine promises project-aware, multi-file edits on a familiar VS Code foundation with simple onboarding and settings import ([TechCompanyNews guide](https://www.techcompanynews.com/how-to-use-windsurf-step-by-step-guide-for-beginners/)). The direction is clear: AI IDEs are moving from inline suggestions to autonomous, runnable workflows. Operational maturity remains a concern: users report surprise auto-updates ([automatic updater](https://forum.cursor.com/t/cursor-automatic-updater/152697)), Windows update failures ([Windows updates failing](https://forum.cursor.com/t/updates-on-windows-are-failing-still-antivirus/152819)), and visibility issues before approval in a recent build ([v2.5.20 diffs visibility](https://forum.cursor.com/t/modified-code-changes-not-visible-before-approval-cursor-v2-5-20/152760)), alongside UI changes like replacing "Keep All" with auto-approve ([discussion](https://forum.cursor.com/t/the-loss-of-keep-all-the-addition-of-auto-approve/152780)). Community threads also cite rate limits even on paid plans ([Reddit](https://www.reddit.com/r/cursor/comments/1rdfk9p/what_would_make_you_switch_from_cursor_to_another/)) and a practical auth fix for a Windsurf codex plugin by clearing a local token file ([Reddit fix](https://www.reddit.com/r/codex/comments/1rdddu3/windsurf_codex_plugin_issue/)). Teams are sketching an "AI builder stack" that pairs an agentic IDE with project tracking, instant deploy previews, and AI QA to close the loop from change to validation ([HackerNoon](https://hackernoon.com/the-ai-builder-stack-linear-cursor-vercel-and-qatech?source=rss)). New native entrants like macOS-focused G-Rump hint at a widening field and specialization opportunities ([Swift forums](https://forums.swift.org/t/g-rump-a-native-macos-ai-coding-agent-looking-for-early-feedback/84953)).