Before LLMs, the minimum useful audience for custom software was maybe 1,000 people. Now it might be 10.
Startup software needed a market. SaaS needed many customers. Internal tools needed enough employees to justify the build.
A friend group, household, meetup, or single recurring workflow can be enough reason to build.
The real gain is not “30% faster functions.” It is that tiny friction becomes worth solving.
Setup, UI, backend, database, deployment, and maintenance often cost more than the problem.
A rough but useful version can be built fast enough that small tools become rational.
Vibe coding scales down better than it scales up.
Small apps have nearby users, reversible mistakes, simple data, and one person who can still understand the whole system.
Large systems turn generated code into coordination debt, vocabulary drift, hidden coupling, and unreadable issue queues.
Fun demos, throwaway prototypes, and experiments where the point is learning or taste.
Useful, low-risk software for a small group with fast feedback and manual recovery.
Dangerous unless the agent is constrained by architecture, domain language, and review.
For a few friends, a mini CRUD app can now be easier than maintaining a fragile sheet.
Current LLMs are extremely good at basic software for real people you know.
Few screens, few users, few edge cases. You can still explain the app in one conversation.
The users are nearby. You can ask what happened, fix it manually, and delete bad ideas.
If it breaks, nobody loses money, medical data, legal rights, or production operations.
A motivated person can read the code, understand the data, and know where complexity lives.
A meetup series, trip, game night, habit sprint, or temporary household routine can deserve software.
No roadmap, investors, migration plan, or brand promise. If the group changes, the app can disappear.
Disposable does not mean useless.
A few hours of work created admin controls, game management, and a screen that fits the moment.
In a big greenfield PMS build, agent-written code can grow faster than anyone's shared understanding.
A better LLM helps, but it does not scale to infinity without architecture, domain judgment, and programming experience behind it.
You probably do not need React. You might not need a database. Object storage, a JSON file, or a tiny SQLite schema may be enough.
RSVPs, topic ideas, questions, and links are scattered across chat.
The organizer sees attendance, topic queue, and anonymous questions.
The group gets notes, links, recap, and lightweight history.
Score sheets, timers, round history, and printable summaries for board games.
Agenda, decisions, owners, links, and follow-ups for a small recurring group.
Rotations, swaps, overdue items, and small household accountability.
Shared goals, check-ins, streaks, and weekly reflection for friends.
A clean group knowledge page linked to a bot that can save answers from chat.
Role notes, game setup, and scoring around the existing Resistance / Avalon project.
It is static HTML, published as a Pages, built in 2h.
Wrangler uploads the built static folder directly to the existing Pages project.
The deck is shareable at software-for-10-people.pages.dev, with preview URLs for each deployment.
Look for a spreadsheet, pinned message, or repeated chat ritual that already acts like a database.
One workflow, one URL, one owner, and a data model you can explain without opening the code.
Use the first version to learn the real workflow. Polish only after the group actually uses it.
Not every useful app needs a market. The future is also tiny apps.