HarmonyOS 7's AI Ambitions Signal a New Political Economy of Code

It's easy to dismiss operating system updates as just another set of features and benchmarks. But sometimes, an OS release quietly reorders the political economy of software development. That's what's happening with HarmonyOS 7, which debuted at HDC 2026 two months ago. While most coverage has focused on speed and new functions, senior engineer and HarmonyOS advocate Liu Guangzhi argues that the real shift is deeper: the entire system is now being reorganized around AI agents. That's not just a technical pivot. It's a political statement about who controls the app ecosystem—and how.

The Agent-First Operating System

Traditional operating systems are passive hosts. You pick an app, open it, and navigate through menus to get something done. HarmonyOS 7 flips that script. You state your intent—say, "sign me up for a marathon"—and the system's AI assistant, Xiaoyi, breaks that request into subtasks, coordinates multiple sub-agents (health, calendar, search), and executes them in parallel. This is powered by HMAF 2.0, the upgraded HarmonyOS Multi-Agent Framework, which sits at the center of a six-layer architecture: Xiaoyi as the entry point, HMAF for task decomposition, the openPangu 2.0 AI models, system-level engines (Ark, StarShield security, HarmonyLink), developer tools (DevEco Code and CLI), and scenario layers like spatial computing.

For developers, the implication is profound. Your app is no longer a standalone silo. It becomes a service that the OS can discover and invoke. To make that happen, you register an agent extension that exposes capabilities with a structured schema. The system matches your app's declared abilities to user intent and dispatches structured tasks to it. This isn't a voice assistant calling one API. It's a multi-agent negotiation where your app is a node in a larger system.

Developer Tools as Political Instruments

Huawei's toolchain strategy is a two-track approach that mirrors broader geopolitical divides. DevEco Code is the "brain"—an AI-powered IDE that plans, writes, compiles, and fixes code on its own. It's built on Huawei's proprietary Bifang engine plus the open-source OpenCode framework. DevEco CLI, on the other hand, is a command-line interface that exposes HarmonyOS capabilities as commands, letting any AI agent (Claude, Cursor, or your own) plug in without adopting a specific IDE.

This dual-track is a political choice. It serves different developer demographics: DevEco Code for newcomers and greenfield projects, DevEco CLI for established teams with existing AI pipelines. But it also signals Huawei's attempt to balance self-reliance with ecosystem openness. The proprietary engine ensures deep integration with the HarmonyOS toolchain, while the open-source layer invites third-party tools via MCP protocols. It's a classic strategy of courting developers while retaining control.

Three Philosophies, Three Systems

Comparing HarmonyOS with Apple and Google reveals distinct political philosophies. Apple's approach is "open access": Xcode 27 bridges MCP to its internal XPC, letting third-party agents like Claude and Gemini integrate. Google, meanwhile, went the opposite direction—shutting down its open-source Gemini CLI and replacing it with a closed, cloud-integrated Antigravity, which caused a developer uproar. Huawei's "dual-track" sits in between: one tool with an integrated brain, another that's pure capability exposure.

Cost is another political lever. Apple charges for third-party models (Claude Pro starts around $20/month). Google locks you into its Gemini ecosystem, with enterprise plans at $45 per user per month and a $100 tier. Huawei, aiming to grow its developer base, gives access for free—built-in GLM-5.1, 50 calls per minute, and the option to plug in DeepSeek or OpenAI models. In the short term, free access is a strategic subsidy to win developers, a classic move in platform wars.

Skills, Standards, and the New Battleground

Beyond cost, the skill ecosystem is where the real fight is happening. Huawei offers 70+ curated "Skills" for specific scenarios like multi-device adaptation and concurrency safety. Apple and Google have their own versions, but notably, Apple and Huawei both adopt the SKILL.md open format. This "Skill-as-Code" standard is emerging as a de facto industry norm, even across rival platforms. It's a reminder that even in a fragmented world, some technical standards become shared political ground.

The Real Test: Bridging Development and Runtime

The key metric, Liu argues, isn't code generation rates but whether development-time and runtime agents can truly work together. Huawei is betting on a unified architecture where the AI that helps you write code is the same AI that runs the system. That's a bold political statement—it suggests that the OS is not just a platform but an active mediator of all software activity.

Case in point: Kuaishou, a major Chinese video app, reported that after integrating HarmonyOS AI tools, its AI code generation hit 80%, test case adoption was 84%, and overall team efficiency rose 1.7x. But the more interesting detail is that they built a custom Skill for concurrency safety, which cut a two-person-week task to half a day. This shows that the real value isn't raw AI power—it's the ability to embed domain-specific intelligence into the development workflow.

What This Means for Developers

For developers, the takeaway is practical and strategic. First, decide whether your team belongs on DevEco Code or CLI. Second, leverage existing Skills to avoid reinventing the wheel. Third, use community knowledge packs like harmonyos-ai-skill to fill the gaps in ArkTS training data. But beyond the tactical advice, the bigger picture is that HarmonyOS 7 is a political project. It aims to create a self-contained ecosystem that can rival Apple's and Google's, not just in features but in governance—who gets to define how apps are built, distributed, and invoked.

As operating systems become more agent-centric, the battle lines are shifting from hardware specs to the very fabric of software development. The question isn't just which OS is faster or smarter. It's which OS will own the relationship between human intent and machine action. HarmonyOS 7 is a bid for that ownership, and whether it succeeds will depend on its ability to convince developers—and by extension, the global tech community—that its vision is not just technically sound, but politically acceptable.