I built it alone with AI agents, without a software team
Akkademi is an online question-practice platform for students preparing for the ALES and DGS exams in Türkiye. It has a website, a mobile app, memberships and payments, legal texts and ad measurement. I built all of it as one person, working with AI agents, without a software team.
How I worked. I gave agents jobs but never gave up the steering wheel. Each agent had a role: one for site and mobile code, one for releases and servers, one for design and content, one for legal text and measurement. The shared rules lived in one file: no passwords or keys written anywhere, nobody touches another agent's folder, ask before deleting, every job leaves a dated report folder.
The key lesson: a release gate. AI writes fast, and it also makes mistakes fast. So every change went through the same order: backup first, then automated tests and a check gate, then release, then a live check, with a way back to the previous version. An agent saying "done" was never enough; the live site was verified separately by me or by another agent.
Legal and privacy work was not skipped. For personal data, cookie consent, marketing consent, deletion requests, cross-border transfers and payment terms, I had agents draft the texts, compared them line by line with how the site really behaves, and left the final decision to a lawyer. The rule was never to publish anything untrue: no fake reviews, no fake counters, no fake discounts.
Measurement and visibility. Page titles, sitemaps, redirects and structured data for search; event tracking and double-count prevention for ads and purchases; phone and desktop checks for speed and layout. Those same checklists have now become the audit report we use for other people's sites.
Mistakes happened too. Patches collided, a test server timed out, one agent misread another's work. What we learned: split work into small pieces, test every piece, keep a written record of who touched what, and keep final responsibility with a human.
What this story does not claim. We do not say AI does everything on its own, or that anyone will get the same result. This is an account of experience, not a promise of client outcomes.
What Archirenk does with it. We apply the same method (clear tasks, written rules, tests and a release gate, human review) to other businesses' sites: audit first, then fixes, then care.