Most "AI built my app" stories are toy demos — a to-do list, a landing page. This one isn't. It's a real, in-production community platform with a verified business directory, an events calendar, a classifieds marketplace, and a library of practical guides — in two languages, one of them right-to-left. And it went from empty repo to live in weeks, not the months we'd originally scoped.
The site is IraniLink, a directory and community hub for Farsi-speaking Iranians living in the UK. We're sharing it here because it's a clean illustration of where AI agents have actually landed: not autocomplete, but doing the bulk of the work.
What got built
IraniLink isn't a single page — it's four interlocking products:
- A business directory of nearly 500 verified Farsi-speaking businesses across a dozen categories — from healthcare and legal to IT & technology.
- An events calendar for cultural and professional gatherings.
- A classifieds marketplace for buying, selling, and renting within the community.
- A growing set of practical guides covering the things that are genuinely hard when you move countries: immigration, the NHS, banking, housing, driving, and work — written in both Farsi and English.
The hard part isn't any one of those features. It's all of them, bilingual, with proper right-to-left layout, SEO that works in two languages, and a database that stays clean as community submissions roll in.
The honest efficiency claim: weeks vs months
We scoped this the traditional way first: a multi-month build for a small team. Bilingual RTL alone usually eats weeks — every component, every form, every date format has to work in both directions. Add a moderated directory, content for dozens of guides, and the SEO plumbing, and "a quarter" felt optimistic.
It shipped in a few weeks. That's the ~5× compression — measured in time-to-launch, not hand-wavy productivity. The work didn't shrink; the agents absorbed most of it.
How the work was actually split
The stack is the boring, proven kind: Next.js (App Router) + Supabase, with next-intl driving the bilingual routing and RTL. What's different is who did the work.
Code & scaffolding. Claude Code — Anthropic's agentic CLI — handled the structural heavy lifting: the route tree, the bilingual i18n setup with RTL, the Supabase schema and row-level security, the directory and events data models, and deployment wiring. Not snippets — whole features, end to end.
Content & guides. The UK-life guides — immigration, NHS, banking, housing — were drafted by AI in both Farsi and English, then reviewed. Writing accurate, useful bilingual content at that volume is normally the slowest part of a content site; here it ran in parallel with the build.
SEO & data. Metadata, hreflang for the two locales, sitemaps, and the directory's category structure were all AI-generated and audited. The same caching and ISR lessons we've written about for this site were applied so crawler traffic is served from cache, not from your wallet.
Design & UI. Layout and component styling were AI-led too, iterated against the brand rather than handed to a separate design pass.
Orchestration spanned a few tools: Claude Code as the primary builder, OpenClaw for running and coordinating agents on remote infrastructure, and Gemini for specific generation tasks. The point isn't the particular brand mix — it's that the workflow was agent-first. A human set direction, reviewed, and corrected; the agents produced.
Why this works now
Two things changed. First, coding agents got good enough to hold a whole feature in their head — schema, UI, i18n, and tests — instead of one function at a time. Second, the surrounding tooling (the kind catalogued across this directory) means an agent can actually act: query a database, run a migration, deploy, check logs. That's the difference between an assistant that suggests and an agent that ships.
The lesson for anyone building: the bottleneck has moved. It's no longer typing speed or boilerplate — it's clarity of direction and the quality of your review loop. Point capable agents at a well-defined problem with the right tools, and a "multi-month" build is increasingly a few-week one.
You can see the result for yourself at iranilink.co.uk — a real platform serving a real community, built mostly by machines, in a fraction of the usual time.