Quiet workshop desk with laptop and tools — small careful projects, not an enterprise dashboard

Niche projects as future prospect

by rf + AI · · 8 min read · Originally written in English

I earn my living as an agency employee on Magento and related platforms. That work is real craft: large catalogs, Cloud ops, integrations, upgrades that must not hurt. It is also not the whole story of what software can be for in the next few years.

AI-assisted delivery — Cursor, gated tests, reviewable diffs — compresses how long it takes to stand up and maintain a careful custom application. The same force that makes some “employee as coding machine” roles fragile (I wrote about the labour and ethics side in Reading Obsolete) also makes niche projects thinkable again: work that died on budget in the classic custom era, and that SaaS never quite fit.

No project is too small — when “small” is honest

“Niche” here does not mean toy demos. It means:

  • a co-op that needs a booking + member area that WordPress plugins keep breaking;
  • an NGO that needs a case tracker that respects privacy and can be self-hosted;
  • a small manufacturer that needs a thin mobile companion for warehouse moves without buying an ERP module they will never understand;
  • a club or collective that needs web + mobile from one shared logic, not two forever-divergent apps.

Those organisations rarely had the budget for a multi-year custom programme or for the maintenance tax that followed. With AI inside a disciplined process, delivery can move in weeks or a few months, and maintenance can stay a fraction of the old burn — if the stack is owned, typed, tested and readable by humans who stay accountable.

Agency day job and niche prospect

These are not enemies. Agency Magento work keeps judgment sharp on hard systems. Niche custom work is a hedge and a vocation: a place where freelancers and small agencies can offer something SaaS will not customise, and where NGOs and small companies can fund a tool that improves how they organise without pretending they are “digital transformation” theatre.

The risk to ordinary coding employment is real. Pretending otherwise is not ethics; it is marketing. The constructive use of the productivity boost is to open doors that budget closed — not to undercut colleagues by racing to the bottom on quality.

Choosing a stack: criteria, not a pitch

When a customer project wants web + mobile from shared logic, I sometimes use Umbraculum (brochure, docs) as an illustrative example. I do not run it as my own daily product. I first encountered it in early Cursor chats. Its licences and posture (open, forkable, serious about governance) make it a sensible reference for the class of niche builds I mean.

That is not an endorsement, not an affiliation, and not “you should buy this.” There are many stacks. What I care about when pointing at any of them:

  • Forkability — can the organisation leave without begging a vendor?
  • Licence — terms a board can understand (AGPL / copyleft / OSI-clear); no surprise SaaS lock-in dressed as open source.
  • Right tools and languages — TypeScript, PHP, or whatever your maintainers can actually hire for.
  • Ethics — privacy by default, labour honesty about AI assistance, and humans who keep the judgment when models help write code.

Choose with caution. A fashionable repo with the wrong licence or a dead maintainer is not a niche strategy; it is a future hostage situation.

Possible applications (sketches, not a catalogue)

  • Member / volunteer ops — roles, shifts, simple approvals, exportable data the org owns.
  • Field + office — mobile capture with a thin web admin; offline-friendly where the network is bad.
  • Internal tools — inventory of assets, small CRM for donors or students, document workflows that should never live in a chat thread.
  • Quality applications — instructions, procedures, reports, audits, non-conformities and the rest of the ISO-ish paperwork loop: the class of tools many companies once ran on Microsoft Access, and that you can now build with a mobile native-first philosophy.
  • Local commerce-adjacent — not “another Magento,” but a focused companion (appointments, subscriptions, community pickup) that talks to a larger shop when needed.

Each of these used to require either a brittle plugin zoo or a budget that never appeared. The new possibility is not magic: it is shorter loops, stricter gates, and stacks you can still fork when the world changes.

What I will and will not claim

I will not claim AI removes the need for craft, review or domain knowledge. I will not claim one framework is the answer. I will claim that freelancers and small agencies who treat AI as an orchestration layer under tests — and who pick tools for forkability, licence and ethics — can now propose niche projects that were unthinkable on yesterday’s budget. NGOs and small organisations can fund small applications that improve how they work, without renting a permanent hostage SaaS.

If that prospect interests you, the service page is Niche projects — no project is too small. The Magento and Docker work stays; this sits beside it, not instead of it.