I’m about to tear this project apart and return it to its core features. Please join me in crossing your fingers in hope that a beautiful, functional MVP will be born from the ashes. Welcome to part 5, the finale of my journey building the web app Superior Network with Replit's AI Agent. To see where this started, check out part 1 , or here's a quick summary: TL;DR: A recap of parts 1-4 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: With an over-featured app falling apart at the loose seams, a Project Manager recommended I return the web app to its core. Spring Cleaning / Trimming Down to Core Value I first learned how to back up the entire site (so far) on GitHub, then got to work playing Operation: Prompt by prompt, I asked Agent to remove features like the Messaging platform, matching based on age and gender preferences, and achievement badges. The beginning of my site bug checklist, which became 115 checkboxes long. I’m not entirely sure who to thank, myself for writing thorough prompts, Agent for understanding them correctly, or Replit for enhancing their AI product since I started this project a few months ago, but this time, I successfully reduced the web app while maintaining functionality. I was finally able to take a full look at the site and write out a checklist of every action that needed to work, so I could check each off as I double-check they worked. In this screenshot from my checklist document, I highlighted tasks in yellow if I tried the task and it didn’t work. I highlighted in green when I got a broken yellow task to work again. In other words, all the green highlighted tasks were my trickiest hurdles, but they were fixable. Inbox Headaches | Is Agent Actually Amelia Bedelia in Disguise? One of Replit Agent’s biggest challenges was email. I was able to set up a Gmail integration easily enough, but once it started sending automated emails on my behalf, it was inconsistent and unprofessional. Pfft, 'bring business cards.' I swear, I never asked it to say that! And visually, this email was looking less than ideal. But even after the emails looked good, Agent kept adding its own flair: How are '12 active professionals' going to attend this meeting if the group size limit is 4 people? Why is the email promising 'collaborative project opportunities' when the Superior Network site doesn't guarantee this anywhere? Most frustrating of all, I could request these types of emails be triggered by certain events, but they weren’t being sent out (at least not consistently). I requested that Agent send a follow-up email one hour after every meeting took place, inviting users to revisit the app and leave a private note on how the meeting went. This was also an opportunity to notify Superior Network that the match was a no-show or, more urgently, to flag/report inappropriate behavior. But despite asking for these emails to go out, they weren’t popping into my inbox. I could ask Agent for a sample of the email template to be sent for review, but it’d generate a new one to send as a sample, not the one we’d already designed. There were moments when I was shocked how specific I needed to be, like I was making my wish with a genie or giving a task to Amelia Bedelia. Big, Bold Moves Towards a Working Replit App I decided to make one more big addition to the app, even though its core features were in place and almost 100% of features were working. I knew I needed a way to better visualize and track the emails being sent out. So I asked for an employee access page: Now THAT was smooth building! With just a few minutes communication and a few days of testing the results, this dashboard Agent built accurately reflected real events and automated emails. There was such a good feeling seeing emails pop into my inbox right on schedule, looking professional and mildly obnoxious, just like all the other businesses out there. This was a business. It was finally happening. I also successfully set up error notifications. Now anytime the app generates an error, I get a detailed email about what happened. Superior Network finally feels well taken care of and regulated. Did I Build a Web App with Replit? Did I build a web app with no coding experience? Yes, yes I did. The caveat? This project took way longer and hit many more hurdles than anticipated. I started in May, expecting to finish by June, and instead I’m wrapping up the MVP in September. It’s built and it works. And that’s what I set out to do. Next step: Beta testing. It's time to see if people will actually use the app, which is perhaps the scariest stage of all. Want to explore the final product? Check out Superior Network ! The above content was written by Lindey Hoak. BearPeak Technology Group is a software studio based in Boulder, CO, offering studio, startup, strategy, and staffing services. Schedule a free consultation at bearpeak.io/contact .

It's time for a web app intervention. To see where this started, check out part 1 , or here's a quick summary: TL;DR: A recap of parts 1-3 Despite no software development skills, I was able to design a functional website from scratch. All the heavy lifting was performed by Replit's AI Agent, an overly confident optimist. The idea: A web app to connect business professionals for lunch called Superior Network The outcome: A project full of so many features that it lost its core purpose. When we left off, Superior Network was finally 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." Product Management John was right. In our excitement to push the limits of AI web building, we’d flown too close to the sun, and abandoned any real plan in the process. I needed a product manager, someone to help Superior Network find its footing again. Enter Trevor, a BearPeak PM. You can find more about the great work he does here . Trevor and I jumped on a call and we quickly found the issue: Superior Network had never really had product management at all. "What does the app actually need in order to work?" Trevor tested me. "Well," I thought about this. "It needs to connect people within a close vicinity of each other. And it needs to be able to book a meeting between them. That's it." "My thoughts exactly. That's the core. My advice would be to take all of the extra features away, but just for now. Focus on the product's core and get it out there to test users. That matters more than whether a chat page works or meetings can be auto-scheduled for users. When you build a core product, you need to learn to block out the requests for more content. Block out your users, builders, and even managers, and just build the features that have earned their place; just the ones that deliver core value. Then, as users test your MVP, you can start to re-integrate those cool but unnecessary features." It was a simple and humbling lesson that I think everyone needs to revisit sometimes. As a designer, I can very quickly get caught up in details, and as a designer with the new freedom to build a unique site with just myself and AI, I had thought, "why not see what it can do?" And I certainly had tested its capabilities. But if we were going to take Superior Network seriously, publish and market it as a real web app, it needed to return to the basics. So Will I Build a Web App with Replit? Can this designer actually bring the MVP across the finish line? Can she strip away all the shiny-but-broken features without mourning the hours spent building them? And once it’s reduced to bare bones, will Superior Network even work? Tune in next week for part 5, the finale of this app-building story. Want to skip ahead to a spoiler by exploring the final product? You can check out Superior Network now! The above content was written by Lindey Hoak. BearPeak Technology Group is a software studio based in Boulder, CO, offering studio, startup, strategy, and staffing services. Schedule a free consultation at bearpeak.io/contact .

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! The above content was written by Lindey Hoak. BearPeak Technology Group is a software studio based in Boulder, CO, offering studio, startup, strategy, and staffing services. Schedule a free consultation at bearpeak.io/contact .

Excelling at multiple roles as product, software, and site reliability engineer, John gained a 360° view. Now he manages his own team of hand-selected professionals. BY ENGINEERS, FOR BUSINESSES Meet your strategic advantage in technology. Work at BearPeak → Meet Our Founder & CEO John Eckhardt has been part of 4 startups, all of them successful. Excelling at multiple roles as product, software, and site reliability engineer, John gained a 360° view of the development process. Now, he manages a team of hand-selected professionals with one goal in mind: creating amazing experiences. Talk with John Partnerships Professional Associations Sponsorships Careers Want to Join the Team? Join our mission to help startups start up! - Work remotely - Choose your hours - Connect with excellent software developers, dedicated and supportive teammates, and leaders with top-tier resources and guidance. We're a team that likes to get outdoors, sport BearPeak merch, and solve problems with purpose and drive. If you think you'd be a good fit, let's talk! Check Open Positions

TURNING IDEAS INTO ACTION Great People + Great Process = Amazing Product A preview of our work with: Client: NDTCO A Secure & Unified Investment App BearPeak became NDTCO's outsourced product and engineering team, and also embedded a Fractional CTO with their leadership. In less than 18 months, we reimagined their engineering team and brought their first mobile app to market. Client: NDTCO Legacy CRM -> HubSpot Migration BearPeak became NDTCO's outsourced product and engineering team, and also embeeded a Fractional CTO with their leadership. In less than 18 months, we reimagined their engineering team and brought their first mobile app to market. Client: Ad Fontes Media Scaling Without Compromise When Ad Fontes Media set out to grow their software capabilities, they needed more than extra developers: they needed a strong, solid foundation for long-term growth. Client: TrainingPeaks Beating the Clock We identified risk, optimized schemas, and migrated an athletic platform's database to AWS RDS with zero downtime. Best part? We doubled their delivery velocity. 😎

It's all about the people. Will your business be next? See a preview of our work with New Direction Trust Company and Ad Fontes Media. IT'S ALL ABOUT THE PEOPLE Will your business be next? A preview of our work with: New Direction Trust Company Deeply Embedded BearPeak became the outsourced product and eng. team, and dedicated a Fractional CTO to embed with NDTCO's leadership. The boundary between where NDTCO ends and BearPeak begins isn't even that clear. We're a natural extension of their team physically present to them on a regular basis. In less than 18 months, we've reimagined their engineering team and brought their first mobile app to market. Check out some of our accomplisments: App Build CRM Migration Ad Fontes Media The Natural Extension BearPeak brought in hand-picked specialists to reinforce the Ad Fontes Media engineering team. We helped build front- and back-end apps, move systems from legacy hosting to GCP, and drove the development of the Media Bias Chart 2.0! As they work with more predictive tools based on human-truthed datasets, we're also helping advance their AI story.