Alibaba's AI Assistant That Builds Apps in 30 Second

Blog image

Author: Boxu Li

Unlike traditional chatbots limited to text, Lingguang delivers code-driven outputs – it can respond with charts, 3D models, images, and even instantly generated apps. This industry-first capability empowers even non-programmers to create functional software tools just by asking, marking a significant leap in consumer AI technology[2]. Ant Group’s vision for Lingguang is to “make difficult topics easy to understand” and to transform how users access information by making it interactive and visual at every step[3].

Flash Apps: From Natural Language to Mini-App in Seconds

At the heart of Lingguang is the “Flash App” feature – a breakthrough that lets users describe an app they need and receive a working version on the fly. Simply by typing or saying a request (for example, “Create a budget tracker with weekly and monthly totals”), Lingguang will generate a personalized, interactive mini-app complete with user interface, logic, and data handling. All of this happens directly within the chat conversation in as little as 30 seconds[2]. This means you can go from idea to a runnable app faster than it takes to boil a cup of coffee, without writing a single line of code.

Blog image

Lingguang can generate functional mini-apps (such as a “New Year Countdown” timer) entirely from a user’s prompt. The assistant designs the interface and writes the code, then runs the app within the chat.

Under the hood, Lingguang’s AI writes the full code for the app (hence “code-driven” outputs) and instantly executes it. The result isn’t a mockup or static graphic – it’s a live, interactive application you can tap and use immediately. In one example, a user asked “溏心蛋要煮多久？” (“How long to boil a soft-boiled egg?”) and Lingguang generated a “soft-boiled egg timer” app that allowed the user to input egg size and desired yolk firmness, then calculated the optimal boiling time[4]. In another case, asking about car maintenance costs yielded a “car cost calculator” app where the user could adjust mileage and fuel price sliders to personalize the result[5]. These examples illustrate how Flash Apps go beyond Q&A – they provide a tool that the user can interact with to get a personalized answer.

Notably, Lingguang’s Flash Apps are fully functional, not just demo front-ends. The system can incorporate back-end capabilities like AI models or data services so that the mini-app can perform real computations or fetch information in real-time[6]. In other words, the apps can “call” backend AI services as needed, which hugely expands what they can do. For instance, a Flash App can generate on-the-fly charts from user data, or pull live map information for a travel planner. This makes Lingguang’s mini-apps far more powerful than a typical template-based app builder – they are dynamic and intelligent by design.

Each Flash App also comes with built-in options to refine and share. After Lingguang generates an app, the user can tweak it through follow-up instructions (e.g. “add a reset button” or “make the background green”). In fact, Ant Group observed that most users iterate about six times per Flash App, and some power users made over 100 tweaks within two hours as they fine-tuned their generated apps[7]. Once satisfied, you can save the mini-app or share it with others, making it easy to distribute your newly minted tool.

How Lingguang Builds Apps: Code-Driven Multimodal Magic

Lingguang’s ability to “think” in code and visuals at the same time is enabled by Ant Group’s advanced AI architecture. Internally, the assistant breaks down your request into multiple parallel tasks – understanding the language, planning the app’s logic, designing the interface, and so on[8]. He Zhengyu, Ant Group’s CTO, explains that Lingguang will “automatically decompose user queries into task frameworks” and process them across different domains simultaneously[8]. For example, if you ask for a travel itinerary app, the system in parallel might generate text descriptions, a map interface, a 3D landmark visualization, and the application code that ties these together[9]. All these pieces[1]: are then merged into[1]: one cohesive response: you receive, say, a travel guide card with[1]: an interactive map and a working trip planner app in one go.

[1]: Introducing apps in ChatGPT1

[2]: Introducing apps in ChatGPT2

[3]: Introducing apps in ChatGPT3

[1] Introducing apps in ChatGPT and the new Apps SDK | OpenAI

https://openai.com/index/introducing-apps-in-chatgpt/

[2] [3] [4] [5] Expedia in ChatGPT

https://www.expedia.com/product/expedia-in-chatgpt/

[6] [7] [8] ChatGPT Integrates Expedia and Booking.com to Transform the Future of Trip Planning - Travel And Tour World

https://www.travelandtourworld.com/news/article/chatgpt-integrates-expedia-and-booking-com-to-transform-the-future-of-trip-planning/

[9] [10] [11] Zillow debuts the only real estate app in ChatGPT - Zillow Group

https://www.zillowgroup.com/news/zillow-becomes-first-real-estate-app-in-chatgpt/

[12] [13] [14] Making Coursera learning more accessible worldwide through an app in ChatGPT - Coursera Blog