Kotlin 15th Anniversary & Kuikly Eco Day: Cross-Platform AI Engineering Practices

On September 16, Tencent hosted the "Kuikly Eco Day × Kotlin 15th Anniversary" event at its headquarters, focusing on cross-platform development and AI convergence. The nearly four-hour program featured seven deep-dive sessions from Tencent business teams and Kotlin ecosystem experts.

Opening: Fifteen Years from Language to Ecosystem

Tencent Big Frontend Committee President Qiu Jun traced Kotlin's 15-year growth from a new language to a mature, influential cross-platform technology. He reviewed Kuikly's journey from first business adoption to GitHub open source, crediting community feedback for its continuous improvement. Qiu highlighted the keyword "cross-platform and AI fusion" as "efficiency within efficiency," noting Kuikly has expanded beyond a framework into toolchains (Skills, DSL plugins, Agent, Harness) and that business teams are solving real-world challenges.

Fan Shengyou: AI-Era Kotlin Multiplatform Development Experience

Kotlin senior advocate Fan Shengyou presented "Beyond Code Generation," detailing three new language features with code comparisons showing reduced boilerplate, optimized parameter passing, and simplified collection creation. He outlined key ecosystem investments:

Kotlin Language Server Protocol : Independent open-source project supporting code completion, inspection, and refactoring, integrable into VS Code and any editor.

Official Dependency Search Platform : Provides multi-platform support, star ratings, and version info for efficient library selection.

Unified Toolbox (Alpha) : Integrates project creation, build, test, and publish commands in a single workflow.

Compose Multiplatform : Fills multi-platform UI gaps and introduces Hot Reload — save and redraw without recompilation or restart.

Cross-Platform AI Agent Framework : Written in Kotlin with built-in tool calling, memory, and RAG capabilities.

The team emphasized that despite AI-assisted development trends, Kotlin's language evolution and tooling remain core to code maintainability and human-machine understandability.

Sun Tao: Making Coding Agents Reliably Work

JetBrains AI Solutions Expert Sun Tao addressed the gap between strong models and reliable Coding Agents. He decomposed agent evaluation into three pillars: fixed dependency environments, real engineering problems, and deterministic acceptance. The evaluation system centers on "task solve rate + compile pass rate."

Introduced Harness as a core concept, advocating inclusion of custom toolchains, project indexing, and RAG context supply in comprehensive agent capability assessment.

Demonstrated IDE transformation into an "Agent-first development environment" via plugins.

Presented an experimental dev environment supporting agent configuration, harness comparison, and optimization suggestions.

Data shows efficient IDE tool integration significantly reduces response latency and token consumption.

Sun stressed "prefer official tools" and revealed a shift toward a next-gen AI plugin supporting multi-subscription switching and full execution trace retention, laying groundwork for reliable, observable Coding Harness environments.

Li Hanxiong & Chen Zhenhua: AI-Friendly D2C & Inspector for Kuikly

Kuikly client team experts tackled "AI restoring design drafts" and "on-device UI understanding."

D2C (Design to Code)

Directly reads design structure to generate developable code. Restoration quality depends on design input, tool mapping, and engineering integration. The team moved rule checks before transcoding to avoid "garbage in, garbage out," significantly improving generated code's engineering value.

Inspector

Focuses on unified runtime semantics. After components sink to native platforms, identity, accessibility text, and source locations are lost. The solution: "compile-time call site recording + runtime high-level semantics collection," enabling AI precise location, code modification, and automated testing — building foundational capabilities connecting coding, running, and verification.

Wang Zhenyu: Native2Kuikly AI Transcoding Practice

Tencent Stock App client lead Wang Zhenyu shared an engineering case: migrating legacy native code to the cross-platform framework via AI. Facing large codebases, missing docs, and zero-exception release requirements, the team built a phased AI migration pipeline following four principles:

Context capsule isolation

Subtask independent rerun

Breakpoint resume

Sequential dependencies

Combined with a multi-stage scheduling workflow, migration became controllable and maintainable. Notably, first-round transcoding quality jumped dramatically by supplementing structure diagrams, visual alignment, and item-by-item closure; remaining issues were patched manually. Wang underscored that the core difficulty is not "code generation" but "continuous stable delivery" — echoing Sun Tao's reliability focus.

Chen Chaoming: QQ Browser AI Native Practice & Evolution

QQ Browser Tech Lead Chen Chaoming recounted the team's evolution from early cross-platform trials to supporting HarmonyOS. Confronted with massive legacy rewrite, they chose "KMP + KuiklyUI" as the next-gen architecture to reuse existing native capabilities. For AI adoption, they avoided generic workflows, instead building standardized workflows with multiple architectural gates and unified contract declarations, achieving three-end (HarmonyOS, iOS, Android) API reuse and stable iteration.

Three-end code sharing rate increased significantly, accumulating a batch of reusable components.

A substantial portion of routine requirements are AI-generated; simple requirements achieve "one-sentence delivery."

Adding compile verification and context collection as deterministic processes drastically compressed AI interaction rounds, making delivery stable and controllable.

Xie Yicheng: Kuikly H5 in QQ Game Center — Landing & Evolution

QQ Value-Added Frontend Expert Xie Yicheng shared Kuikly H5 practical experience. By refactoring the rendering chain, thousand-node render time dropped significantly. For black-screen on launch, a dynamic resource management scheme enabled independent resource release and preload, greatly optimizing cold and warm starts, nearly eliminating black screens. On engineering efficiency, dependency analysis drove on-demand builds, sharply cutting overall build time; activity page artifact size was also notably compressed. The team plans deeper WebAssembly vs. JS performance comparisons for large-data computation and codec tasks.

Huo Bingqian: Kuiklybase Kotlin — Yesterday, Today, Tomorrow

Tencent Video Cross-Platform Framework Lead Huo Bingqian closed the event, reviewing Kuiklybase Kotlin's version evolution from early HarmonyOS adaptation to today's multi-ecosystem compatibility challenges. The team has completed underlying architecture unification, validated internally. For the "inevitable" upgrade, they adopted pragmatic strategies:

Establish seamless switch, one-click rollback, small-traffic verification for smooth upgrades.

Phased delivery: first ensure business-transparent switch and API compatibility, then progressively optimize performance.

Long-term philosophy: "reduce custom baggage, rely on official optimizations, contribute customizations upstream."

A concrete "performance dividend": bottom-layer compile chain duration dropped from hours to minutes. Huo emphasized that upgrade landing is just the start; building long-term maintenance mechanisms is key to sustainability.

Technical References

Kuikly GitHub repository: https://github.com/Tencent-TDS/KuiklyUI

Kuikly Documentation: https://kuikly.tds.qq.com/Introduction/arch.html