How Should Tech Organizations Restructure for the Deepening AI‑Native Era?

DRIVE Model and Five Key Practices

The opening keynote introduced the updated “DRIVE” model, a five‑dimensional framework for building AI‑native organizations by elevating cognition from the R&D layer to the business layer. The model defines five strategic practices:

Technology Strategy Decoding : Reposition the technology committee as the primary driver of business growth, responsible for forecasting inflection points and defining three‑to‑five‑year technology strategies.

Team Cognition Reshaping : Align senior leadership and middle management on AI strategy through intensive “training camps” that set standards, benchmark against peers, and ensure end‑to‑end execution.

Safe‑to‑Fail Environment : Establish innovation funds, honor systems, and regular Hackathons to nurture cross‑team collaboration and organic idea emergence.

From Tech Management to Tech Business : Shift from a pure R&D mindset to a business‑oriented view, converting technical capabilities into SaaS‑style outcomes integrated with overall company strategy.

API‑Based Capability Packaging : Encapsulate departmental abilities as reusable APIs or MCP patterns for internal consumption and external ecosystem integration.

Google Cloud Agentic AI

Google Cloud presented its transition to an AI‑first organization, delivering a full‑stack AI solution that combines custom chips, large models, and cloud services. The platform offers data, copyright, and compliance guarantees via digital watermarks and provides a unified stack that optimizes efficiency for global customers.

Kuaishou AIDevOps – Three‑Layer Reconstruction

Li Si reported that 89% of enterprises have deployed AI, yet average productivity gains are only 0.29%. He identified a gap between tool usage and organizational redesign and proposed a three‑layer reconstruction:

Knowledge Layer : Make AI‑accessible data available to models.

Process Layer : Replace serial workflows with Agentic (AI‑driven) workflows.

Organization Layer : Separate delivery from guardianship, dissolving rigid product‑research boundaries.

Applying this reconstruction to a live‑gift feature reduced the release cycle from 20 days to under four days. The L2 (collaborative) paradigm now accounts for the majority of workflows, achieving a 20‑30% cycle‑time reduction compared with L1 (assistive) workflows.

MoonBit AI‑Friendly Language

Zhang Hongbo argued that AI‑native programming languages are feasible despite limited corpora. MoonBit’s design addresses four challenges of AI‑augmented software engineering:

Maintainability – AI‑generated code is hard to treat as long‑term assets.

Toolchain fragmentation – Disparate tools limit AI context.

Reliability – 30‑40% of Copilot‑generated snippets contain defects.

Deployment – Cross‑platform distribution remains a bottleneck.

MoonBit reduces historical baggage with a lean type system, unifies compilation, testing, and diagnostics, and supports multi‑backend output (Wasm/JS/Native) to simplify deployment. An Italian study cited in the talk showed that, with only one‑seventh of Gleam’s corpus, MoonBit achieved better AI‑generated code quality, confirming the advantage of AI‑centric language design.

Kuaidi100 AI‑Native Practice

Li Chaoming described the evolution of Kuaidi100’s logistics intelligence graph, expanding carrier coverage from 10 to nearly 15 major carriers. The “Intelligent Capacity Selection” feature recommends the optimal carrier for each address, serving over a thousand enterprises with more than 5 billion calls.

The AI‑native methodology combines edge‑cloud collaboration, CLI‑based agents, Skills, and MCP protocols. It also introduced the “Fruit‑Bot” agent, a digital courier assistant that operates as an AI‑powered personal assistant, enterprise manager, and courier aide.

Round‑Table Highlights

Tim Yang (GIAC), AfterShip CTO Hong Xiaojun, Lovstudio.ai founder Shougong Chuan, and EverMind COO Han Yunyun discussed breakthroughs in their domains, reinforcing the emphasis on practical AI‑driven engineering, cross‑layer redesign, and robust tooling as essential for true AI‑native transformation.