The Shift

There's a category of software that's quietly becoming obsolete: boutique, single-user tools. The kind of app that does one thing well for a narrow audience — a stats dashboard, a niche data formatter, a personal news aggregator, a custom report generator. Software that maybe a few hundred people use, built by a solo dev or a tiny team, sold for $10/month or offered as a side project.

These tools are being replaced by something simpler: a single prompt to Claude.

What Changed

The economics of niche software never really worked. Building even a simple web app means choosing a framework, setting up hosting, handling auth, writing a frontend, maintaining dependencies, and dealing with the long tail of edge cases. For a tool that serves a small audience, this overhead is brutal.

A one-shot Claude prompt skips all of that. You describe what you want, you get it. No deployment. No maintenance. No dependency updates at 2am.

A Concrete Example: The Modern Newspaper

We built The Modern Newspaper as a daily briefing tool. It pulls news across politics (left and right perspectives side by side), tech, local Bay Area news, Stanford, robotics, Parkinson's research, and schizophrenia clinical trials. Each section is curated, summarized, and rendered into a static HTML newspaper with AI-generated hero images.

Before LLMs, building this would have meant: - Subscribing to 6+ RSS feeds or news APIs - Writing custom scrapers for each source - Building a summarization pipeline - Designing and maintaining a frontend - Handling rate limits, API changes, and source rotations

Instead, each section is essentially a well-crafted prompt. The entire system is a static site generator that reads prompt outputs and renders them. The complexity moved from code to language.

When This Works

One-shot prompts replace software best when:

1. The user is the developer. You know exactly what you want. There's no product-market fit to find — you ARE the market.
2. The output is text, data, or a simple artifact. Reports, summaries, analyses, formatted data, even HTML pages.
3. It doesn't need to be real-time. Daily, weekly, or on-demand is fine. You're not building a trading system.
4. The "app" is really just a transformation. Input goes in, output comes out. No complex state, no multi-user collaboration.

When It Doesn't

This doesn't replace everything. You still need real software when: - Multiple users need shared state - Real-time performance matters - You need hardware integration - The task requires a persistent UI that people interact with throughout the day

Our other projects — Memex, Buddy, Screen Self-Driving — are real software because they need persistent processes, hardware access, and real-time interaction. A prompt can't replace a face-recognition system running on a Jetson.

The Implication

If you're a solo developer building a tool for yourself or a small audience, ask: "Could this be a prompt?" If the answer is yes, you just saved yourself months of maintenance. The best code is the code you don't write.

The Modern Newspaper generates a full daily edition with 11 sections, hero images, and a signup page. The core logic is ~200 lines of Python gluing prompts to a Jinja2 template. Most of the value is in the prompts themselves — not the code.

That's the shift. The hard part of building niche software used to be the software. Now it's the prompt.