Hello HN, I'm a long-time iOS developer (since the iPhone 3GS era). Recently, I solo-built and launched a global dating app using Flutter, Supabase, and Next.js – all in about 100 days, heavily relying on Cursor's Pro plan ($20/mo).

While AI tools like Cursor make rapid development possible, I wanted to share some practical realities from the engineering side and the economics of building with them.

1. Market Validation Experiment (Why a dating app?)

Before coding, I tested the market by creating a female profile on an existing major dating app, using lightly AI-edited photos.

- Day 1: 200+ likes, and it continued strongly for a week.

- Rough math: If each like represents ~$1 in potential revenue for the platform, that's over $1,000 in a week from a single profile photo.

- Reality: This convinced me the market was huge, but launching my own app showed the challenge – great functionality isn't enough without network effects and brand. It's a quiet launch so far.

2. Cursor's Economics (The changing value)

In the first few months, the Pro plan delivered tremendous value. Thanks to unlimited Auto Mode, my actual usage was worth over $1,000/month – while the bill was only $20/month.

It felt like working with a team of 10+ developers. Tasks that would normally take me a week were often completed in a minute. It was exhilarating, but over time it became exhausting. As a human, keeping up with AI's pace drained me mentally.

Starting around month 4, I started hitting usage limits much faster (e.g., exhausting monthly compute credits in days on heavier tasks). Cursor has shifted to a compute-based usage pool (equivalent to ~$20 of API credits per month), with overages if you go beyond. It's a reminder that these tools' "unlimited" phases can evolve as models get more expensive.

The analogy to a "drug dealer" model fits in the sense that early generous access builds dependency, then costs adjust – fair for sustainability, but something to budget for.

3. Engineering Challenges (Where AI helps – and where it doesn't)

AI handled a lot of boilerplate, but integration details required manual work.

- SSO (Google Login): Always tricky; issues like browsers not closing or missing callbacks. Supabase Auth made it manageable compared to past experiences – sticking closely to official docs was key.

- Notifications: Firebase/Google Cloud consoles are still confusing. Choosing OneSignal for the backend simplified things a lot and was a solid decision.

- IAP: The 30% cut (15% for smaller devs) is steep with no real alternatives yet. Planning local payment gateways later.

- Translation: I implemented on-demand translation (button-press like LinkedIn). Then saw some competitors (e.g., Chinese apps) with seamless real-time translation across full content – impressive tech gap, and a reminder there are always advanced implementations out there.

4. Conclusion

AI tools enabled a solo 100-day build, which felt impossible before. But building the product is one thing; getting users and traction is another – marketing/brand is the real barrier, like building a great hotel on a deserted island. Curious about others' experiences with Cursor's evolving limits or solo-launching consumer apps. Links:

App Store: https://apps.apple.com/us/app/weconnect-cultural-exchange/id...

Google Play: https://play.google.com/store/apps/details?id=com.abus.wecon...

Web: https://www.wctokyoseoul.com

The Joy of Hex and Brouwer's Fixed Point Theorem (2013)

New Ways to Corrupt LLMs

AI and camera counter human trafficking

AI Browser Extensions Leave Fingerprints Everywhere

Norman Podhoretz, 1930-2025

FTX insider Caroline Ellison has been moved out of prison

The Genius Effects of Old Movies [video]

Show HN: Made a Visionboard Tool

Minimum Viable Benchmark (For Evaluating LLMs)

Fara-7B: An Efficient Agentic Model for Computer Use

Apple TV's new intro was done practical, not CGI or AI [video]

Luminar Technologies, Inc. Initiates Voluntary Chapter 11 Proceedings

The Lost Generation

Video: Lunar impact flash detected on the moon by Armagh Observatory

How I Assess Open Source Libraries

Deaf Crocodile Blu-Rays

The Core Problems of AI Coding

Project Zeros New Website

Show HN: Website to APK Convertor

Building a WASM Runner with Cloud Hypervisor

A Guide to Magnetizing N48 Magnets in Ansys Maxwell

The Bifurcation in the AI Market

OpenAI in Talks to Raise $10B, Adopt Amazon's AI Chips

Show HN: Better Gmail Tabs – turn search queries into tabs for fast email load

Show HN: Obsidenc – a Rust-based paranoid-grade encryption utility

After years of resisting it, SpaceX now plans to go public

What happened to iRobot can happen to anyone

You're 25-35 years old? You're running out of time

Tesla engaged in deceptive marketing for Autopilot and Full Self-Driving, judge

Show HN: 200+ Likes/day as fake profile → built my own dating app in 100 days

Show HN: Video Cards-Turn Articles into Video and Podcast Scripts in Seconds