KM4BI

The Five Kinds of Software You Can Build

A practical map for matching a personal software idea to the right shape, tools, and degree of control.

Published Updated

Personal software does not automatically require a complicated stack. The first decision is to identify the shape of the project before choosing a model, coding agent, or platform.

Five useful shapes

A **local tool** runs on one computer and suits individual work such as organizing documents, sorting photos, or doing private research. A **web app** opens in a browser and can live on a phone home screen; it is the simplest starting point for most personal projects shared with a partner or a small team.

A **native mobile app** is warranted when the product truly needs deep phone capabilities: Bluetooth, NFC, background location, reliable notifications, or app-store distribution. A **background service** may have no screen at all: it wakes on a schedule or event, processes information, and sends a result. A **hardware project** connects software to the physical world through sensors, radio, cameras, or buttons, using tools such as Raspberry Pi, ESP32, or Home Assistant.

Pick tools after the shape

For a first web app, the video recommends a hosted builder such as Lovable to minimize setup and get an interface online quickly. Replit and Bolt offer other integrated routes. For more control, a coding agent such as Codex or Claude Code can work alongside GitHub, Supabase, and a host such as Vercel. That path has more moving parts, but it makes it easier to replace one layer without moving everything.

The important distinction is between layers: a model supplies reasoning, a coding agent operates files and tools, and a hosted builder packages an environment and publishing flow. Owning the code, data, and decisions matters more than stacking several agents.

Stay in control

Choices affecting cost, data, privacy, deployment, or portability should be visible. Four small files can preserve the project goal, decisions, real-world scenarios, and agent-specific instructions. Those scenarios also become practical tests on the devices people will actually use.

The closing advice is to start with the simplest path, keep applications private by default, keep secrets out of code, enforce authorization in the data layer, and validate the result yourself. AI lowers the barrier to building; it does not remove the need for sound architecture and human testing.

Source