From Vibe Coding to Wish Coding: AI Programming Hits a Consumer Turning Point

Recent AI‑coding tools such as Cursor and Claude Code have dramatically increased developer productivity, allowing faster code creation in a near‑conversational manner. However, these tools still require users to manage IDE setup, dependency handling, and cloud deployment, so the barrier to turning code into usable software remains high.

The industry now diverges into two paths: one continues to accelerate code writing, while the other asks whether it is possible to skip code entirely and deliver software directly from natural‑language intent.

Ant Group’s upgraded Ant Ling Guang app, released on April 20, exemplifies the latter approach with its “Wish Coding” feature. Wish Coding reduces the software creation starting point from writing logic to describing intent, hiding the compiler, environment, and deployment steps. Users simply state the desired functionality, and the system generates, packages, and launches a ready‑to‑use app.

AI‑coding products can be grouped into three delivery levels: (1) code snippets or project files (e.g., Cursor); (2) editable, previewable project environments (e.g., Bolt.new, Lovable); and (3) integrated development‑run‑deploy capabilities (e.g., Replit Agent). The first two still assume users are willing to enter a development workflow, understand project structure, resolve dependencies, debug, and decide how to publish. For non‑technical users, any step in this chain can become a blocking point.

Ling Guang’s answer is “Zero DevOps”: compilation, packaging, and deployment are almost invisible, and users receive a fully functional application without ever seeing code. To achieve this, the platform relies on a structured intent representation layer that parses natural‑language input into a rigorous function‑module tree and interaction flow, defines data dictionaries and module coupling in a high‑dimensional semantic space, and then dynamically assembles code based on this intermediate structure. This modular architecture ensures a stable skeleton that can survive multiple natural‑language‑driven revisions.

A notable breakthrough is the integration of a native runtime on the device side. Unlike many AI‑code generators confined to browser sandboxes, Ling Guang’s “flash apps” run in native mobile containers, gaining access to real‑time location, gyroscope, accelerometer, and vibration APIs.

In a practical test, the system was asked to create a “My Soul Animal” quiz with 30 quirky questions, output a result such as “melancholy elephant” or “joking capybara”, and render a six‑axis radar chart. Within two minutes, a standalone mini‑program was generated, built, and launched. An initial misinterpretation of the title was corrected with a single follow‑up instruction, and the resulting interaction flow was coherent.

Limitations emerged: when tasked with building an AI visual‑recognition tool, the system produced only a front‑end UI mock‑up without actual image processing. Ambiguous or contradictory user instructions sometimes caused logical confusion and bugs, and the generated UI often appeared overtly AI‑generated.

The article argues that comparing such consumer‑grade agents to industrial‑grade software is a mismatch. The core question for a mass‑market coding agent is how to deliver a functional, runnable system from completely unstructured, non‑technical input. This requires the system to act simultaneously as product manager, architect, designer, and programmer, prioritizing functional closure and immediate usability over visual polish or deep logical rigor.

From a broader perspective, this shift mirrors early internet evolution: rough web pages first achieved accessibility, then mobile apps prioritized availability over stability. For the 99 % of users without coding experience, a minimally functional app that runs instantly delivers far more productivity than a perfect code snippet that never executes.

Consequently, the metric for creativity moves from “code implementation ability” to “intent expression ability.” As the cost of trial‑and‑error approaches zero, ordinary people can turn ideas into usable applications, opening a new layer of software production that previously did not exist.

Ling Guang also launched the “Ling Guang Circle,” a collaborative community where users can share flash apps and iteratively modify each other’s creations via natural language, turning software into a plug‑and‑play digital content that blurs the line between producer and consumer.

While professional engineers will still be needed for complex, high‑reliability systems, the article concludes that the consumer‑grade “Coding Agent” path marks the beginning of a mass‑adoption era for AI‑driven software creation.