I spent a year building a PWA to track 150 cafe workspaces. Did AI actually help, or did it just give me a "Tinder for Coffee"?

Kevin G — Wed, 28 Jan 2026 15:47:31 +0000

The Premise I’m a remote worker.

I spent the last year working from 150+ cafes in Denver because I was tired of "Google Maps Roulette"—where a 5-star review means great lattes, but zero outlets and a "no laptops" policy.

To solve this, I built

caferoamer.com

a Vanilla JS/PHP/MySQL PWA. Along the way, I leaned on AI to "speed things up." Looking back at the repo, I’m not actually sure if it saved me time or just gave me new ways to procrastinate.

The "Cool(ish)" UI Trap One afternoon, I decided the site needed a "discovery mode." Instead of a boring list, I thought: “What if it was like Tinder, but for cafes?”

With an AI coding assistant, I knocked out the basic swiping logic in minutes. It felt like magic. But then came the "jank":

Handling touch-start vs. mouse-down events across different mobile browsers.

Getting the "swipe-away" animation to feel fluid and not like a PowerPoint transition.

Managing the state so that a "swiped" cafe didn't immediately reappear in the main feed.

I spent three days polishing a feature that took 30 minutes to draft. It’s "cool," but does anyone actually want to swipe right on a coffee shop? Probably not. AI lowered the barrier to entry for a "fun" feature, which lured me into a week of UI polishing I never planned for.

When AI Became a Time-Sink: I used AI to help build my "Accidental Backend"—a custom firewall, an email server from scratch, and a sophisticated weighted-randomization algorithm for the feed.

The AI was great at generating the "boilerplate" for things like:

PHP rate-limiting logic.
SQL cleanup queries for my 150+ messy field notes.
PWA manifest configurations.

But the friction started when the "boring" stack (Vanilla PHP) met "complex" AI-generated logic. Because the AI didn't have the full context of my custom-built, dependency-free architecture, it kept suggesting solutions that required libraries I didn't want to use. I spent a significant amount of time "un-hallucinating" the code back into raw PHP.

The Verdict: Net Gain or Net Loss? If I'm being honest:

The Wins: AI was a beast at data transformation. Moving my "old person" screenshots and messy notes into a structured MySQL schema was 10x faster.

The Losses: It encouraged "Feature Creep." Because I could generate a merchant-claiming dashboard or a payment-integrated ad engine in an afternoon, I did.

I built a solution for a 1,000-city platform before I even had 50 users in Denver.

The Real Validation now that the "AI-fueled" build phase is over, I’m back to the hard part: human validation. I’m looking for 500 beta users to tell me if the "vibe" metrics I collected (like lighting and food variety) are actually useful, or if I should have spent less time on the "Tinder-swipe" UI and more time on the basics.

To my fellow devs:

Have you found that AI actually shortens your time-to-launch, or does it just make it easier to build "janky" features you don't actually need?

Leave a comment on what you'd like to see in the evolving remote worker ecosystem over the next 10 years.

Forem: Kevin G

I spent a year building a PWA to track 150 cafe workspaces. Did AI actually help, or did it just give me a "Tinder for Coffee"?

The Premise I’m a remote worker.

Have you found that AI actually shortens your time-to-launch, or does it just make it easier to build "janky" features you don't actually need?