Agentic iOS Development with RocketSim and RocketTrace
Article Summary
Antoine van der Lee built his Video Compressor app with AI agents that can actually see and measure what they're building. His accessibility score jumped from 0/100 to 99/100 without manual testing.
Van der Lee shares his workflow for agentic iOS development using Cursor with RocketSim (for UI navigation and accessibility audits) and RocketTrace (for performance monitoring). He provides four production prompts he uses daily, all running in Plan mode so he can review findings before code changes.
Key Takeaways
- Accessibility audit found 31 issues across 141 elements, fixed to 99/100 score
- Analytics showed 61% conversion rate, guiding TipKit implementation priorities
- Performance trace caught 178 hitches and 19 hangs in 51 seconds
- Plan mode lets you review evidence before agents edit code
- Agents capture App Store screenshots with bezels across 18 locales automatically
AI agents become dramatically more effective when they can run your app, audit accessibility, measure performance, and show evidence before making changes.
About This Article
Antoine van der Lee's Video Compressor app experienced 178 hitches and 19 hang/stall findings during a 51.8-second performance trace. Core Animation layout work on the main thread caused 85% of these issues.
Van der Lee used RocketTrace to find the root causes. AI agents then fixed three specific problems: a 198ms stall from PhotoLibraryService copies, PhotoKit work running on the main actor for each video, and 319 duplicate thumbnail ID writes that happened during scrolling.
The agent's fixes targeted measurable bottlenecks based on actual evidence. It also recognized that 102 hitches didn't have enough context to fix and suggested adding OSSignposter instrumentation first.