I first started experimenting with LLMs for coding sometime in the fall of 2024. My free time had become increasingly scarce since becoming a father of two, and my ability to work on personal projects had shrunk to whatever snippets I could grab here and there.
Before, I'd spend entire weekends diving deep into code, wrestling with bugs until 3 AM, and endlessly refactoring until things were just right. Now? I had maybe an hour here or there, and my brain was operating at maybe 70% capacity on a good day. This new reality meant my long list of hobby projects I wanted to build was likely to remain just that - a list of ideas.
Then I discovered the potential of using LLMs for development assistance. I started with integrating various models into my workflow through the Roo Code plugin in VSCodium. After trying several options, I found Claude Sonnet 3.5 to be particularly effective for coding tasks. I was skeptical at first - I've been coding long enough to remember when Stack Overflow was the height of programming assistance. The idea that an AI could genuinely help me code faster seemed far-fetched.
I was wrong.
My first serious test was with a Flask API I'd been meaning to build for months. I had the general architecture in mind but kept putting off starting because I knew the boilerplate setup alone would eat up the precious little time I had. I described what I wanted to Roo Code, and within minutes, I had a working skeleton with proper project structure, dependency management, and even some basic endpoint implementations.
What would have taken me an entire free weekend was now up and running in a single brief session.
The real game-changer came when I upgraded to using Sonnet 3.7. The improvement in understanding complex programming concepts was immediately noticeable. I could describe architectural patterns I wanted to implement or bugs I was facing in natural language - the same way I might explain them to a senior developer - and get meaningful solutions back. Of all the models I've tried, I've found this one to be the most capable for coding assistance.
One evening, with only about 40 minutes of free time, I sat down to work on a particularly tricky model. Pre-LLM, this would have been just enough time to get frustrated before giving up. Instead, I explained the problem to the LLM model, and we collaboratively worked through a solution. By the end of that short session, I had a working, tested model.
The projects started actually getting finished. My data collection routines evolved from basic scripts to robust systems with proper error handling and scheduling. I was able to develop the models that I needed to analyze my data collected data - something I had wanted to do for ages but never found the time to complete.
It's not just about writing code faster, though that's certainly part of it. It's about maintaining momentum on projects despite fragmented time. Before, if I had only 30 minutes, I'd often spend 20 of those just remembering where I was and what I needed to do next. With AI assistance, I can quickly describe where I left off and get immediate context to jump back in.
I've found some workflows particularly effective. Instead of starting with a blank file, I now begin projects by describing the architecture and components I'm imagining to the model. We iterate on this high-level design first, which saves countless hours of refactoring later. For complex algorithms, I explain the problem and my initial thoughts, then get help refining the approach before I write any code.
With my limited time, the constant starting and stopping used to make coding feel like pushing a boulder uphill. Now it feels more like a collaborative process that can adapt to my fragmented schedule. The interruptions that are inevitable with family life aren't nearly as disruptive to my workflow as they once were.
For other developers out there struggling to find time for their passion projects, I cannot recommend this approach enough. These tools won't replace your skills or creativity, but they will help you maximize whatever limited time you have. They're like having a patient pair-programming partner who's always available, never tired, and doesn't mind if you disappear for a while and come back hours later.
I still haven't found the secret to having more free time as a parent of young children, but I have found a way to keep building things I care about in the margins of my busy life. And for now, that's enough.