Design and Implementation of Next‑Gen App Development Tools Using Compose Multiplatform × AI
summary
A desktop visual builder named ComposeFlow desktop app is built with Compose Multiplatform and models UI as Kotlin data classes, storing project state in a YAML intermediate representation. The system exposes editor operations as callable tools for an LLM-driven agent that runs a loop of tool calls; the pipeline uses KSP to emit JSON function specs and JSON Schema from Kotlin types, caches prompts to save tokens, and includes a serializer that reports exact parse errors and applies fallback defaults to improve retries. The demo shows live recomposition and agent-driven edits, such as adding a navigation drawer.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 1:03Compose Flow overview
- 7:58Visual editor to Kotlin code
- 16:39AI agent architecture
- 28:40Reliable AI agent workflows
- 36:52Summary
related talks
46mPowering Compose Multiplatform Apps with Koog & MCP
34mMulti-coding agent workflows
50mCompose the Future: Building Agentic Applications, Made Simple
38mCross-platform Strategies for Porting and LLM-Driven Development
41mHow AI Will Change Mobile Development
28mLiquid User Interfaces for the Next Generation of AI Apps
29mFrom Copy-Paste to Composition: Building Agents Like Real Software
37mDevelop faster with Android Studio, AI and skills