Multi-coding agent workflows
summary
A background coding agent runs inside Android Studio using isolated git work tree sandboxes to build, lint and autofix merges while the main editor stays focused; visual diffs from the Compose preview and an ADB-driven emulator test harness validate UI and flows automatically. A prototype converts Figma to Compose by iterating on previews, and a Cotlin bench shows model progress (Gemini 2.5 Pro ~15% to GPT5 high ~30%). The system reports teams committing over 10 million AI-written lines.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 3:12Coding agent adoption
- 7:03Parallelization challenges
- 10:36Valley of death in productivity
- 13:50Demo: Background coding agents
- 16:20Accuracy with background agents
- 23:23UI testing with sub-agents
- 26:53Writing evals for agents
- 29:02Benchmarking progress
- 31:11Future of coding agents
- 32:57Call to action
related talks
40mDevelop faster with Gemini Agents in Android…
37mDevelop faster with Android Studio, AI and skills- 37mDevelop faster with Android Studio, AI and skills
41mHow AI Will Change Mobile Development
35mFirebender's AI Coding Agent - Reverse engineered
35mShip - Hype: Rolling Out AI at Scale to 60 Android Engineers
41mAccelerating the Developer Loop: Exploring a…
38mDesign and Implementation of Next‑Gen App Development Tools Using Compose Multiplatform × AI