Why I Skipped Building a Custom Login System (and Chose the Fast Path Instead)

The Login Wall That Almost Stopped Me

I’ve never built a full-featured authentication system from scratch in Next.js before.

At first I thought: “Easy, just follow a tutorial — magic auth in a weekend.” Then I opened the docs, saw the full checklist (sign-up, login, password reset, email verification, OAuth, sessions, security headers, rate limiting…), and felt the enthusiasm drain away.

Even asking various AI tools to generate it felt overwhelming — the output was correct but massive, and maintaining or customizing it later looked like a nightmare.

I realized I was about to spend days (maybe weeks) on infrastructure that has almost nothing to do with whether people actually want and use the core grammar tool.

A Quick Reality Check

While watching a documentary to clear my head, it hit me:

This project has cost me ~20 CHF (domain + email) so far. The AI usage? Pennies.

Financially I’m only 20 francs in the red. Time-wise, though, every hour spent on non-differentiating work is expensive.

From a pure business perspective: if you don’t yet know whether people will use — and eventually pay for — the product, complexity is the enemy.

Speed without unnecessary features wins early.

The No-Brainer Decision: Use Clerk

Clerk was one of the first major infrastructure decisions—I explained why we generally opt for “good enough” infrastructure instead of building everything ourselves here: Why Speed Beats Perfection When Validating Your SaaS Idea.

After a very constructive chat with Perplexity.ai, I landed on Clerk.

Here’s why it became an instant yes:

• Free tier covers 50,000 monthly active users — way more than I need at launch
• Built-in Google, Facebook, email/password login
• Handles register, login, password reset, session management — everything out of the box
• Clean Next.js integration with almost zero boilerplate
• Secure by default (no need to become a security expert overnight)

For context: there are indie SaaS products making $40k+/month MRR with only ~8,000 active logged-in users. Even if AI Grammar Mentor grows fast, 50k MAUs is nowhere near the free-tier ceiling.

Decision made. Clerk it is.

Future-Proofing (But Not Yet)

Long-term, yes — if the product scales massively, high per-subscription fees from payment processors or auth providers could become painful.

In that future I’d likely migrate auth to a custom Stripe + database setup or bring someone on to optimize.

But right now? We launched two weeks ago. We’re not even close to that stage.

Premature optimization at this point would be the real waste.

Key Lesson for Early-Stage Builders

In the first 30–90 days of any potential product:

• Speed of validation > everything else
• Every day spent on non-core work delays real user feedback
• “Good enough” infrastructure that lets you ship fast is almost always the right choice
• You don’t know yet if the idea will work — don’t invest heavily until data says yes

The faster you get something live and in people’s hands, the faster you learn whether it’s worth doubling down.

Cost savings and architectural purity matter later. Momentum matters now.

Where We Stand Today

AI Grammar Mentor is still very young — but already showing promising signs of real usage.

If you’re tired of wrestling with writing, grammar, tone, or clarity (and you want a tool that actually respects your voice while making suggestions), give it a try.

No complicated setup. No long onboarding.

Just paste your text and see how much clearer (and still completely yours) it can become.

Jump in — I’d love to hear what you think.