Review: I Built a Side Project with an LLM Over the Weekend
At work I only touch backend code, but I’d been sitting on an idea for a while, so I grabbed a weekend and built it. To get straight to the result, it made me think, “So one person can get this far on their own.”
The thing I built is a newsletter summarization + tagging service. There’s nothing especially difficult: it’s basically a collect → summarize → classify → send pipeline.
The biggest help was, unsurprisingly, code generation. I’d barely touched frontend before, so I asked about almost everything while setting up the Next.js page structure and components. Of course, I couldn’t just use the output as-is and had to keep fixing it, but it was a huge help in removing the fear of starting from a blank page.
On the flip side, there were things I found lacking.
- When library versions change, it sometimes confidently spits out old code
- Security-sensitive parts like auth/payments ultimately need direct review
- It writes test code well when asked, but it can’t decide what should be tested
The conclusion is that an LLM definitely replaces about “one team member” for someone working alone. That said, the lead role—design and judgment—still feels like it’s on the human.
Next, I’m thinking of adding deployment automation. If anyone else has done a similar weekend project, I’d be curious what stack you used.
6 answers
Agreed—I’m a backend dev from a government-funded bootcamp, and I don’t think I could’ve even gotten started on frontend without LLMs lol. I really think the biggest benefit is that it gets rid of that “blank-page fear.”
I feel like concluding it’s “one team member” is a bit of an overstatement. Team members attend meetings and take responsibility, whereas an LLM is just an assistant that spits out code. I’ve actually run into the library version issue myself—it confidently produced code using an old API, and I didn’t catch it and deployed it, which caused an outage. In the end, you need to know the stack to some degree yourself to filter that out.
I 100% agree that it can’t choose what to test. If you ask it to, it’ll neatly write out given-when-then, but I’m still the one who has to find the actually important boundary values lol