Lessons from the AI Trenches: Building, Breaking, and Learning Weekly
What I am Learning from Building Fast and Failing Even Faster
Hey everyone, it’s been a month since I last wrote here on The Novice. I’ve been deep in the world of AI coding, and let me tell you, it’s been a wild ride. Today, I want to share what I’ve been up to and the lessons I’ve learned so far. If you’ve ever felt overwhelmed by the pace of learning something new or wondered how to make the most of AI tools, this one’s for you.
AI is moving incredibly fast, and if you are reading this newsletter, it means you are still ahead of the curve. This is the golden time to discover new and untapped opportunities, businesses and products. Let’s dive in.
The Weekly Shipping Challenge with Kerry
A few months ago, my friend Kerry and I decided to challenge ourselves: we’d ship a new product every week. Kerry’s a marketer with a knack for explaining things in a way that just clicks with people, and I’ve gotten pretty quick at building things with AI, using Grok and Cursor. Together, we figured we could combine our strengths to turn ideas into reality at lightning speed.
To make this possible, one of the things we did, we built a toolkit called tinylandingpage.com. It’s a set of tools that lets us spin up functional websites in hours—complete with a marketing page and an application layer. The best part? I can tell AI to update the template with new content, and it handles it efficiently. It’s like having a cheat code for rapid prototyping.
Tiny was our way to scale our production without completely burning out by spending endless hours in front of the screen. You will see why it matters as you read the rest.
By the way, let me offer you something — If you are a subscriber of The Novice and you buy a full Tiny bundle from me, I will get on a call with you to explain how everything works, and will help you set it up.
Scoutzie.com and the AI-on-AI Magic
One of the first things Kerry and I did together was rebuild Scoutzie, a company I started years ago but shut down. I never got over killing a perfectly functional company, even if it wasn’t going to become a billion dollar success. Today, Scoutzie’s promise is simple: we help clients build ideas quickly, just like we do for ourselves. We turn your ideas into MVPs, and if need be, we help you launch and market them.
There’s nothing conceptually new about this idea, except that now we can go from an idea to a launched prototype in literally days, if not hours.
To make this happen, we also created an AI intern that helps us do hours of unsupervised work —think of it as AI on top of AI—that takes client conversations and turns them into detailed project plans, complete with requirements, build steps, and logic to make sure we don’t miss anything.
Here’s how it works: instead of clients just telling us what they want, our bot asks them smart questions about design, functionality, and user experience. It’s like having a super-smart assistant that says, “What should the homepage feel like?” or “How do you want users to navigate this?” It’s been a game-changer for turning vague ideas into concrete plans.
I am gonna brag a little here, as I am very proud of the system. You might know how Cursor helps developers to write code, by essentially translating developer’s needs into executable commands to the LLMs. Well, Scoutzie MVP bot helps us write project plans in much the same way. When a client talks to the bot, I usually hang around on FaceTime and watch, and it makes me feel really secure to know that while I would’ve missed all the details, the bot does, it asks all the right questions to get us a hyper detailed plan.
This is 100% the future of AI, btw. Whether you replace people or augment them, systems like our bot will be everywhere, making sure that 100% of the effort is translated to the output.
You can try the Scoutzie MVP bot on your own to get a feel for what it does. I’ll keep making it better. Your money will be well spent!
The Game MVP Rollercoaster
Not everything has been smooth sailing, though. A friend of mine went through our MVP chat process and described a game he wanted to build. Using our toolkit, I had a functional prototype up in an hour—pretty cool, right? But then I spent the next week trying to turn that prototype into a fully functional game worth publishing, and it was a disaster.
The harder I tried, the further we went off the rails.
A good card game should have a couple of components: game logic, game engine, and ability to sync the state between users. I’ve never built a card game before, so I was inventing all these parts to the best of my knowledge and ability. All I could think about was my previous post — I was trying to build in a day what would normally have taken months to complete. While theoretically I had all the tools and the knowledge to do it, I was hitting one wall after another.
AI kept breaking my code, deleting chunks, and generally making a mess of things. I’d add new features, and suddenly, old ones would stop working. It was frustrating, and honestly, I had to step back and put the project on hold.
While I was wrestling with the pigs, friends provided suggestions on how to get AIs under control. In short, if you want AI to be very good at its job, you have to be good at your — provide as granular and painstakingly detailed instructions as you can imagine. More on that below.
If you hit the limits and you cannot get through — delete the whole project and start again. I did that, and the following week our process got a lot better.
GA 4 Hell and the Power of Less
After the game project hiccup, we decided to take a different approach with our next idea: GA4 Hell. If you’ve ever used Google Analytics, you know it’s gotten so complicated that most people can’t make sense of it anymore. So, we built a simple analytics tool that plugs into Google Analytics and actually makes the data usable.
We needed this for ourselves, so we could track all the projects we are launching, and it became apparent that other people wanted this too! This time, we focused on building as little as possible—just the core features that make the product shine. We put up a splash page where people could leave their email if they were interested, and within a day, we had sign-ups.
I have an engineering degree and it pains me to create incomplete projects, but launching GA4Hell with no features, with nothing hooked up other than the landing page, and then doing everything semi-manually over email was the best thing we could’ve done at that point.
It’s been a few weeks and we’ve got a couple dozen people exploring the reports, we just need to convert them over to a monthly subscription. A challenge in itself, but a much less time intensive challenge than writing thousands of lines of code.
Sometimes, building less is the secret to building better.
Oh yeah, at some point I wanted to know how much code I write every month, and I created a github code counter. At the time of creation, I had written 300,000 lines of code in 3 months. Not bad.
Automating Everything—Even Writing and Videos
While I’ve been focused on coding, Kerry’s been experimenting with AI in his own domain: writing. He’s figured out how to use AI to write blog posts that sound exactly like him. Before he starts writing, he pitches his ideas to the AI, which then drafts multiple versions. It prompts him to correct things or answer questions, and then refines the post based on his feedback. The result? Blog posts that feel authentic to his voice, but without him having to write every word.
We’ve taken this automation even further. Soon my YouTube channel will features videos by AI — our avatars talking about our content, but produced and released a robot. It’s wild—like we’re cloning our creativity, but it’s the only way to survive. There’s only 24 hours a day, and we can’t possibly do everything, but we could get as close to everything as possible, when computers do most of the work.
What’s next?
After two months of work, we have automated project ideation, product creation, marketing, analytics, and soon sales and distribution as well. The possibilities feel endless, and it’s exciting to see how much we can do with these tools.
We have an unlimited list of ideas to ship, now we just need to make our tools do all the work to a point where it’s a fully automated, repeatable process. Easier said than done, but also super exciting!
When it comes to shipping code or writing posts, I tend to prioritize shipping. But I should write more, you deserve it. Meanwhile, you can:
Follow me on X. // Find our new projects on QED. // Follow us on YouTube.
If you are a VC, send a term sheet. ;)
Bonus: Lessons for Building with AI
Through all of this, I’ve learned a few key things about building with AI that I want to share with you:
Focus on specific areas: If you’re coding frontend in Next.js, get really good at it. If you’re working on backend in Go, master that. Don’t mix your AIs with different tasks—stick to what you know, and what they know. If you are working on a repo that mixes code, then open it in different Cursors and tell each one to work independently.
Structure your work clearly: Whether it’s a project plan or a coding task, break it down into clear sections or modules, and make those small. Define what each section should include and what information goes where. This makes it easier for AI to follow along and produce useful output.
Instead of saying: build an app that does X, think long and hard about the problem and break it all down. Split your instructions into game rules, game UI, authentication, information management, and so forth. Then split those sections into further smaller and smaller details.
KISS — Keep It Simple, StupidBalance exploration with exploitation: Exploring new things is great, but it can slow you down. If you want to move fast, become a hyper-expert in a particular niche. You’ll be able to do in minutes what used to take days. But if you’re constantly learning new things, like I did with the game project, expect some roadblocks.
This section deserves multiple blog posts of its own. Stay tuned for more —
Try It Yourself
If you’re curious about building with AI, I encourage you to give it a shot. Start small—maybe try automating a simple task or building a quick prototype. And if you do, I’d love to hear about your own novice stories in the comments. What did you learn? What surprised you? Let’s keep the conversation going.
Thanks for reading, and I’ll be back soon with more updates from the frontlines of learning!