In recent months, “Vibe Coding” has become a widely circulated buzzword, with many developers eager to learn how to properly do vibe coding
. A new wave of tools represented by Cursor, Claude Code, and advanced models like Claude Opus 4.7 is pushing software development efficiency to new heights.
Developers who are familiar with engineering systems are experiencing a leap in productivity. They can complete more work in less time, even building complex systems in a nearly “conversational” way.
But this wave of efficiency has not yet truly reached the majority of people.
Even if AI can generate tens of thousands of lines of code, ordinary users are still blocked by tedious steps like IDE configuration, dependency management, and cloud deployment. Vibe Coding has increased the speed of writing code by orders of magnitude, yet it has barely lowered the barrier of “turning code into usable software” — it accelerates coding, not delivery.
Faced with this gap, the industry’s technical path is beginning to diverge. One direction continues to focus on “helping programmers write code faster,” while another tries to answer a more fundamental question: can we skip code altogether and directly deliver software?
On April 20, the newly upgraded Ant Lingguang App offers a noteworthy example of the latter path. It attempts to reconstruct the production relationship of software through what it calls “Wish Coding.”
What Wish Coding Means: From Writing Logic to Describing Intent
So-called Wish Coding essentially lowers the starting point of software creation from “writing logic” to “describing intent.” In Lingguang’s architecture, traditional development environments, compilers, and deployment processes are hidden away.
Users do not need to think about implementation paths. They only need to describe the function they want in natural language. The system completes the entire chain in the background—from code generation to packaging and deployment—and ultimately delivers a ready-to-use application entity.
This is worth serious discussion because it tries to fill a long-missing link in AI programming: enabling ordinary people to complete the full loop from idea to a runnable application.
From Code Generation to Software Delivery: What Layer Is Missing?
Over the past two years, AI coding products have emerged one after another, driven by rapid model advancements such as those seen in the evolution of Claude Opus 4.7 vs Opus 4.6. But from the perspective of delivery, they still roughly fall into several layers:
The first type delivers code snippets or project files, typically represented by Cursor.
The second type delivers editable and previewable project environments, such as Bolt.new and Lovable.
The third type begins to offer integrated capabilities for development, execution, and deployment, like Replit Agent.
All of these products are valuable. But they mostly assume one thing: users are willing to enter the “development process,” understand project structures, handle dependencies, debug errors, and decide how to publish.
For developers, this is not a problem—this is where efficiency comes from. But for ordinary users, any step in this chain can become a blocking point. They may be able to generate code with these tools, but that does not mean they can actually run or use the application.
In other words, developers need “faster development,” while ordinary users need “no development, just results.” These are completely different design goals, and they require completely different solutions.
Lingguang’s answer is Zero DevOps. Processes like code compilation, environment packaging, and deployment are almost invisible to users. Users do not see any code—they simply receive a usable final application.
The Core Mechanism of Wish Coding: Structured Intent Representation
To achieve this kind of end-to-end delivery, the system must solve the inherent ambiguity and divergence of natural language. What users say is often imprecise, incomplete, or even contradictory. How can this be converged into executable software?
This leads to Lingguang’s core underlying mechanism: a structured intent representation layer.
When a novice user inputs a vague or logically inconsistent instruction, the model’s internal process is far more complex than simply outputting code, requiring advanced logic parsing similar to Claude Opus 4.7 adaptive thinking. Lingguang’s agent acts as a system architect. It first parses natural language into a rigorous tree of functional modules and interaction flows, and defines underlying data structures and coupling relationships between modules in a high-dimensional semantic space.
Only after ensuring the completeness of the logical loop does the system dynamically assemble code based on this intermediate structure. This modular architecture ensures that the generated application has a stable backbone. It can withstand multiple rounds of modification and refactoring through natural language, avoiding system collapse caused by chaotic code accumulation.
Another meaningful breakthrough lies in the integration of a native runtime environment on the device side. Many AI app-generation platforms are limited by browser sandboxes, meaning their outputs can only perform simple DOM operations and rendering.
Lingguang’s “flash apps,” however, run directly in a native mobile container. With user authorization, they can access real-time LBS location, read gyroscope tilt and acceleration data, and even control vibration feedback frequency and intensity.
Real Test: How Far Can One Sentence Go?
We also conducted some tests.
For example, we asked Lingguang to design something similar to an MBTI-style “My Spirit Animal” test.
We required the system to construct 30 bizarre multiple-choice questions, and at the end generate results like “a melancholic, overthinking elephant” or “a joke-loving capybara,” along with a hexagonal radar chart showing six unusual dimensions such as intelligence, imagination, and theatricality.
In less than two minutes, a standalone mini-program-style application was built and deployed directly within the chat interface.
Although the system initially misinterpreted the intended title “My Spirit Animal” as “Spirit Zoo,” this small flaw could be fixed instantly with a simple follow-up instruction. From smooth page transitions to the final rendering of the radar chart, the underlying interaction logic was quite clear and coherent.
Consumer-Level Coding Agents: Not Betting on Perfection
During the experience of Lingguang, surprise and roughness appeared at the same time.
For instance, if a user asks Lingguang to build an AI vision recognition tool, it will most likely deliver a “simulator” with only front-end UI interactions, without actually processing uploaded or captured images. Also, because natural language is inherently ambiguous, contradictory instructions across multiple turns can sometimes lead to logical confusion and bugs. The generated UI is still stiff—it looks obviously AI-made.
However, evaluating these “flash apps” using the standards of industrial-grade software, or comparing Lingguang with premium AI coding tools designed for professional developers—where teams often have to weigh commercial factors like Claude Opus 4.7 pricing—is itself a mismatch.
As a consumer-level coding agent, Lingguang is solving a completely different problem: how to deliver a functional, runnable system for users with no technical background, within a completely unstructured input space.
Ordinary users’ inputs are often ambiguous and fragmented, and the system must simultaneously act as product manager, architect, designer, and developer. In this high-dimensional semantic convergence process, ensuring a functional closed loop and immediate usability becomes the priority, which inevitably leads to compromises in visual design or deeper logic.
This kind of roughness is, in fact, an inevitable stage as software engineering moves from “elite production” to “mass expression.”
From Elite Engineering to Mass Creativity
This evolution closely resembles the early days of the internet. Early web pages were rough and chaotic, but they achieved one key thing: they could be accessed. Early mobile apps were unstable, but they enabled a new paradigm of “available anytime.”
At every turning point of technological adoption, “usable” always comes before “perfect.”
For professional developers, a flawless piece of algorithmic code may be extremely valuable. But for the 99% of people with no programming experience, a barely passable application that can run instantly on a phone delivers far more value than perfect but unusable code fragments.
This is the underlying logic of consumer-level coding agents.
As long as the applications they produce can run reliably, be used immediately, and be continuously modified and iterated, they have already crossed the critical threshold—from AI generating code to AI delivering software.
The real bet is not whether they can generate refined applications today, but whether they open up a new possibility: when the cost of trial and error approaches zero, ordinary people can turn their intent into usable applications.
Lingguang has at least preliminarily validated that the path from intent to application is feasible.
A New Era of Creativity: The Rise of Wish Coding
Along with this upgrade, Lingguang also launched “Lingguang Circle”—a social layer for AI applications. Users can share the flash apps they create, and others can continue modifying them using natural language. It is essentially an open-source community of intent.
When these applications begin to circulate within a community and are iteratively modified by others, we start to see a new model of software evolution.
In traditional software engineering, iteration relies heavily on development teams and long release cycles. But in a Wish Coding ecosystem, software behaves more like plug-and-play digital content. Users can create tools instantly for specific micro-needs, use them, and leave them behind for others to build upon.
In such an ecosystem, the boundary between production and consumption begins to blur. Users are both creators and consumers. Applications can be generated, used, modified, and redistributed rapidly. Software starts to evolve like content on a platform.
Boundaries Still Exist
Of course, the boundaries remain clear. For professional engineers, building complex systems is still irreplaceable. High-reliability applications, critical infrastructure, and complex systems will continue to depend on rigorous engineering practices.
Determinism, maintainability, testability, and compliance do not become less important—they become even more important in higher-value systems.
But for a much broader population, the barrier to creating digital tools is decreasing. Wish Coding may open up an entirely new layer of software production outside professional development.
Here, the standard of creativity shifts—from “ability to implement code” to “ability to express intent.”
In this sense, Wish Coding reveals a possibility: when describing a requirement becomes harder than implementing it, the bottleneck of software engineering shifts from technical ability to intent expression.
We may be standing at the early stage of this transition.
As code generation capabilities continue to become democratized, Lingguang is exploring a path toward consumer empowerment—at the cost of tolerating early-stage roughness. For the vast majority of people who have never entered the world of code, this new era of raw creativity has only just begun.
