I Made a Web App with Replit, and I Can Barely Code (pt. 3)
- Lindey Hoak
- Oct 7
- 5 min read
I've almost made a web app with Replit. Almost.
Every time I open it, progress feels like two steps forward, one step back. To see where this started, check out part 1, or here's a quick summary:
TL;DR: A recap of parts 1-2
Despite no software development skills, I was able to design what looked like a functional framework of a website, from scratch.
All the heavy lifting was performed by Replit's AI Agent, an overly confident optimist who can see tiny lines of code in screen captures of the console, but can't see the giant gaps of weird space it designs on the page.
The idea: A web app to connect business professionals for lunch.
The outcome: So far, it LOOKS like it works. Until you touch it.
The next hiccup came as I first deployed the site. Over a video call, my boss John and I connected the Replit site to a custom domain, and we talked excitedly about how close the project was to beta testing.
Little did I know that I was far, far away from that stage. But back to the journey and the newest issue I was encountering: My newly published site required any user to log in with their Replit account before using the site.
Seriously? If I share my finished site, people can’t even view it without making a Replit account first? That’s not just unprofessional; that’s a dealbreaker.
I asked Replit's Agent to work around this. Find out how we could make users only log in with their LinkedIn profile (a requirement I'd assigned to the site as a safety feature. We're making an app that connects strangers in real life to meet up in person. I want to ensure that the users are truthful about their identities, but I'm certainly not asking them to create a Replit account. That'd be like asking someone visiting your online store to create a Shopify account first! Not going to work).
To remove the awkward login, Agent suggested I swap to a Reserved VM configuration (a different type of deployment). I followed its advice, which required another large change: Swapping from demo mode to a permanent database. What I thought was fast acting in excitement for a brand new working website had instead messed up the whole site's functionality.
Trying to log in? Sorry, no luck. Trying to log out? You're trapped. The buttons didn't do anything! Problem after problem, I reported issues to Agent... only to feel déjà vu. Bugs I’d celebrated fixing weeks ago were suddenly back, and we were having the same conversations all over again.
The beautiful colored dots I'd customized with Agent to display how the user's availability on a particular day? Gone like magic. Poof. Remember, I'm a graphic designer; this one still hurts.
I have never troubleshooted more in my life. Sure, I've built websites, then proceeded to test-click every button imaginable, but they only needed to be pressed once to know that they worked.
To be fair, my prompts got lazy. After all these conversations, waiting for Agent to finish a step, refreshing, testing, then seeing the same issue was still occurring, I was treating Agent how I would treat a co-worker who couldn't deliver. But Agent wasn't going to feel bad and work harder to be better. My written attitude wasn't going to do any good.
Too many features?
All this troubleshooting was occurring on the site as I had other projects to work on. But despite Agent's ability to work in the background, it wasn't something I could leave unattended for long. I tried time and again to get Agent to troubleshoot itself; to double-check that new features were actually working before handing the reins back to me with a celebratory: "I fixed everything!" but Agent autonomy was a tool that wouldn't come until a few months later.
Another month went by, and instead of a stable product, Superior Network had accumulated a growing pile of half-working features:
On my boss's request, I added a Smart Scheduler. This mode lets AI match the user with a new connection each week and schedules the meeting for you.
Then I started thinking like a marketer. The site was boring! We wanted to keep users coming back to book more meetings. So Agent and I added achievement badges.
Users could fill out their profile, interests, and availability. But every part of this kept breaking. One minute they couldn't update their profile, then whichever age group they wanted to meet with wouldn't be taken into consideration, then their calendar would be overlooked when a meeting was scheduled.
On one call, John observed that jumping into this new web app was overwhelming. How was the user supposed to know what to do or what skills it had? So I added a Tutorial, but troubleshot with Agent to get it to appear on the dashboard page as it guided users through their available tools. After hours of failed attempts, I finally gave up and moved the tutorial to its own page. It looked clean, but it wasn’t what I’d envisioned.
With all these features and little fraying ends, updating John about the site each week was overwhelming. He'd ask, "When can you have those issues resolved?" and every time, I gave an optimistic estimate that ended up being vastly wrong. It was a flaw I'd never had like this before.
The work I do is almost always solo. So if I promised to deliver a finished product by Friday, it was on me and me alone to make that happen. I've pulled late nights when I overestimated my abilities, but they've been rare. After years of design work, I have a good understanding of how quickly I can complete my tasks.
I say all this in my defense because, as it turns out, when you add an AI, those timeline expectations need to be completely reassessed. I was wrong about Replit Agent's timeline over and over and over, to the point where John considered canceling the project entirely, and rightfully so. I’d never entered so many weekly calls empty-handed after making promises I couldn’t keep. But Agent could only work so fast and correctly solve problems on so many first tries, I could only cross my fingers and hope we'd make it in time.
With a big jumbled project that was starting to look like a real web app, I brought it to John, and he hit me with a new development:
"I don't know what this project is about anymore. This web app doesn't seem to know what it is or does."
The app has lost its identity. So what now? Tune in next week for part 4.
Want to skip ahead to a spoiler by exploring the final product? You can check out Superior Network now!
Comments