What Is AI‑Native Thinking and Why It Will Shape the Next Wave of Applications

Introduction

Baidu founder and CEO Robin Li predicted early this year that the biggest opportunity in the era of large models lies in the application layer, where "killer" applications will emerge.

AI‑Native Thinking

AI‑native thinking means designing and building applications with artificial‑intelligence technology as the core driver, re‑imagining product architecture from the ground up.

Data‑driven mindset : use data analysis and mining to discover and solve problems.

Interdisciplinary mindset : integrate knowledge from different domains.

Innovative mindset : continuously improve technology, products, and services.

Collaboration‑win‑win mindset : cooperate with AI itself, AI professionals, and relevant enterprises.

Risk awareness and responsibility : consider potential risks and assume responsibility when deploying AI.

AI‑Native Applications

AI‑native (AI Native) refers to a method of building and running applications that places AI at the core, encompassing data collection, model training, deployment, and management. The concept is analogous to "electric‑native" (products built on electricity) and "cloud‑native" (software built for cloud infrastructure).

One of the first wave AI‑native products is Baidu Wenku, which has evolved into a "one‑stop intelligent document platform". Recent weekly releases include features such as document‑to‑PPT generation, AI‑generated charts and insights, PPT chart creation, intelligent summarization, and Q&A, all delivering industry‑leading accuracy, richness, and speed.

Examples of these capabilities:

Document‑to‑PPT: understands article structure, style, and key points to generate presentation slides.

AI‑generated charts: transforms textual data into visual charts for research or financial analysis.

Long‑form summarization and personalized Q&A: helps users quickly grasp content and interact with the material.

Hackathon Ideas

During the "Embrace the Large‑Model Era" Hackathon, the author participated and observed many imaginative AI‑native concepts, such as an AI "love strategist" and an AI interior‑design assistant. The most intriguing, though not awarded, was "如诉" – an intelligent legal‑dialogue tool for ordinary users and lawyers, offering knowledge Q&A, AI lawyer consultation, and automated legal content creation.

Advantages highlighted:

Massive legal knowledge base gives AI an edge over human experts in matching statutes and cases.

Local regulatory compliance reduces competition from foreign firms.

High demand for affordable, high‑quality legal advice creates a strong market.

Three Conditions for AI‑Native Apps (Li Yanhong)

Natural‑language interaction must be supported.

Apps must fully leverage understanding, generation, logic, and memory capabilities of large models.

Interaction depth should not exceed two menu levels.

Design Principles

When adding AI capabilities, follow the "add one, subtract two" rule: introduce new entry points while removing unnecessary elements to keep the product simple and improve user experience. Also, prioritize scalability and sustainability to accommodate rapid model evolution.

Baidu Cloud Qianfan Platform

The Qianfan large‑model platform provides a one‑stop development and service environment for enterprises, offering Baidu’s own ERNIE‑Bot, third‑party open‑source models, AI development tools, data management, automated model training (SFT), inference services, and cloud deployment.

Technical Foundations for AI‑Native Apps

Frequency‑control services to manage costly large‑model usage.

Content‑moderation (sensitive‑word detection) for user queries and AI‑generated output.

Real‑time SSE communication for streaming responses and reducing wait times.

Intent recognition, multi‑turn dialogue, and implicit prompt engineering to improve generation quality.

Development Opportunities

Large models are entering the second half of their lifecycle; their ability to understand, generate, reason, and remember can be combined with NLP, speech, prompt engineering, SFT, and web‑crawling to create solutions across media, e‑commerce, entertainment, gaming, finance, education, industry, and healthcare.

Compared with blockchain, large‑model applications have broader reach and easier integration, offering tangible user‑experience improvements. Like the evolution of mobile networks (2G→5G), large models serve as a foundational technology that will drive multi‑industry upgrades.

Conclusion

AI‑native applications powered by large models are becoming a hot topic, but high entry barriers mean the field is still exploratory. Understanding large models, continuously learning, and boldly experimenting are essential for creating value‑driven AI‑native products.