The Software-Shaped Problem: Why Instant App Creation Tools Miss the Real Barrier

A recent post by Jasmine Sun crystallizes a persistent issue in the generative AI and low-code tool space. The quote, posted on January 24th, 2026, reads:

"If you tell a friend they can now instantly create any app, they’ll probably say 'Cool! Now I need to think of an idea.' Then they will forget about it, and never build a thing. The problem is not that your friend is horribly uncreative. It’s that most people’s problems are not software-shaped, and most won’t notice even when they are. [...] Programmers are trained to see everything as a software-shaped problem: if you do a task three times, you should probably automate it with a script. Rename every IMG_*.jpg file from the last week to hawaii2025_*.jpg, they tell their terminal, while the rest of us painfully click and copy-paste. We are blind to the solutions we were never taught to see, asking for faster horses and never dreaming of cars."

This observation cuts to the heart of the current hype cycle around AI-powered development tools. The market has been flooded with products promising to let anyone build software—tools like Replit's AI agent, Bolt.new, and the coding capabilities within Claude 3.5 Sonnet and GPT-4o. The pitch is straightforward: describe what you want, and the AI will generate the application.

Yet Sun's quote identifies the missing link. The barrier isn't the creation of software, but the recognition of opportunities for automation. This is a cognitive gap, not a technical one.

The Programmer's Lens vs. The User's Reality

The distinction Sun draws is between two modes of thinking. A programmer, trained in systems logic, sees a repetitive manual task and immediately conceptualizes a script or application as the solution. The mental model is "This is a software-shaped problem."

For the non-programmer, the same task is simply "work." The pain of clicking through 200 photos to rename them is accepted as part of the process. The idea of automating it doesn't surface because the solution space—command-line interfaces, scripting languages, file system operations—is invisible. They are asking for a faster horse (a better manual tool) because they cannot conceive of a car (a fundamentally different approach).

This is the core of what some in the AI community call "vibe-coding"—the ability to generate code from natural language prompts. The recent surge in "coding agents" like Claude Code and Devin demonstrates impressive technical capability. These agents can parse complex requests and generate functional codebases. But they still require a user to articulate the problem in a software-shaped way.

The Real-World Application Gap

The disconnect is evident in the market. While developers use these tools to accelerate their work—generating boilerplate, writing tests, or refactoring code—the broader adoption for non-technical users remains limited. The tools are powerful, but they solve the wrong problem for this audience.

Consider the recent article mentioned in the source material: "Wilson Lin on FastRender: a browser built by thousands of parallel agents." This is a deeply technical project, likely built by and for developers who already think in terms of software architecture and automation. Similarly, "First impressions of Claude Cowork, Anthropic's general agent" discusses a tool designed for users who have already identified tasks suitable for automation.

The true breakthrough would be an AI that doesn't just generate code, but helps users identify software-shaped problems in their daily lives. This requires a different kind of intelligence—one that understands human workflows, pain points, and inefficiencies, and can suggest automation where the user doesn't see it.

Limitations and the Path Forward

Current tools are limited by their input mechanism: the prompt. A user must know what to ask for. The most sophisticated coding agents can handle vague requests like "build me a website for my bakery," but they still rely on the user's initial conception of the problem as a website.

The next evolution in this space may involve AI that observes user behavior (with permission) and identifies automation opportunities proactively. Imagine an agent that watches you perform a repetitive task in your workflow and asks, "I notice you do this every day. Would you like me to automate it?"

This shifts the paradigm from "ask for what you want" to "here's what I can do for you." It bridges the gap between the programmer's lens and the user's reality.

Until then, Sun's quote serves as a necessary critique. The promise of instant app creation is compelling, but it only addresses the final step in a longer chain of reasoning. The real challenge isn't making software easier to build, but helping people see the world through a lens where software is the obvious solution. That's a problem that requires more than just better code generation—it requires a fundamental shift in how we think about tools and work.

The recent articles and discussions on platforms like Hacker News and technical blogs continue to explore these themes, but the core insight remains: technology is only as useful as our ability to imagine its application. For most people, that imagination is still a missing piece.