I Can Build Every Side Project I Want With AI. Should I?

My opinion of AI right now is a bit of a mixed bag. On one hand, I'm sick of reading AI slop. On the other hand, I've had great success building developer tools with Claude, Codex, and even open models. These AI-built tools greatly accelerate my development process. Having large suites of custom tools and experiments on Hypothesis.sh really speeds up testing of my other work, but even using agents to build me smaller one-off dev tools has been valuable. However, one thing I hadn't been using AI for until recently was building personal/side projects.

My side projects have historically always been built by me, by hand. One of my first AI-assisted projects, Thirsty Bot, was still written back when the best coding tool I had access to was Copilot with auto-complete that occasionally could write whole methods (and not always correctly).

I was pretty happy working like this, but I think every software developer can relate to my predicament: I have too many projects I want to build, and not enough time. What if I could delegate some of these projects to Claude or Codex, and finally have something to show for all those domains collecting dust? Would I feel sad if I didn't implement every idea I had by hand? Most of what I enjoy about software development is the actual coding, right? Am I wasting years of experience by deferring to a machine? Are my skills going to get rusty?

Okay, that's a lot of questions, and there's clearly a lot on my mind when it comes to AI coding. When I'm considering throwing an agent at a new project, here are the questions I ask myself:

Question 1: Why Am I Building This?

The first thing I ask myself is: "Why am I building this?" Usually a project falls into one or more of these buckets:

1. I want it to be a showcase of my skills as a developer.
2. I want to showcase my design/product/other skills (bascially anything software-adjacent, but not coding).
3. I want to learn a new technology or explore a new tool/platform.
4. I want the end result - it will be useful to me.

If the goal is explicitly to demonstrate my coding chops, I'd better be doing the coding - I won't be handing this off to AI. On the flipside, if I just want the end result, the implementation is just going to be a blocker to getting the tool I need or want. In this case, AI can deliver the end result much faster, freeing me up to work on things in buckets 1-3.

My blog (this site) is an example of goal #1. This is not a means to an end - the code and content is the whole point. Every bit of CSS and Javascript is hand-crafted, and every piece of text is written, proofread, and re-written by me.

What about the other buckets? If my goal is #3 - learning - AI can be useful to show me the way, but I'd better be closely reading everything that comes out, and better yet, do a lot of implementation by myself. Otherwise, what have I really learned? (No, how to prompt ChatGPT doesn't count). If I answer #2 (other skills development), it's usually something where I don't care much about the implementation, but I spend a lot of time on the user experience, design, and product side of things. These are well-suited to be AI-assisted on the coding side, with me operating as a product owner and designer.

Question 2: Who Will Use This?

The next question I ask is who my audience is. It's usually:

1. Just me.
2. Me and a small trusted group.
3. The public.

I use this question as not just an AI/no-AI gate, but also to determine the level of effort I put in. If it's just for me or a small group of friends or colleagues, and I decide to go AI-generated or heavily AI-assisted, I may spend less time reviewing the code. Maybe I review diffs from Claude but skip a formal code review/PR and just continue to iterate in production. If I'm deploying this for wider use, and choose to lean on my agents a lot, I'll follow a more rigorous code-review and branching process, like building real production software should, human or not.

Question 3: Am I Ever Going to Do It Myself?

This is the one that got me "un-stuck" on some old ideas I had, and convinced me to (reluctantly at first) let AI work on some of my side projects for me. I'm busy. Between work at Kizen, personal projects, family, cycling, and other hobbies, something is bound to get dropped. A lot of projects I want to build fall into this "dropped" designation. Mostly things that would take a long time for me to code, but have minimal utility or wouldn't be used often.

Here's a good example: My BMW Drive Recorder Player. My built-in dashcam records from 4 cameras and also carries GPS data and speed in a JSON file, but there's no good way to look at these outside the car or view the map. So, I had Claude build me a webapp for storing and viewing them. This is a complicated app - taking the video footage, splitting the 4 camera feeds, playing them back in sync, replaying the GPS data overlaid on a map, etc. Too much work for something I only use once or twice a year if I need to watch back an incident. This idea sat on the shelf for years before I had Claude build it. And I didn't miss out on anything - it would have been boring to build and I'd already done most of the learning during the planning phases. Instead, I get a useful app with minimal time spent on it. That's a win!

Question 4: What Are the Risks When Something Breaks?

Another way to generalize this question would be "How important is this?"

When I am working on something like a core library or an application with billing and paying customers like Thirsty Bot, I'm much less likely to delegate to AI with carte blanche control. I like to carefully and meticulously build these parts. But a developer tool that consumes said library, a simple site or app for testing something, or a personal tool where bugs and downtime only affect me? This is where I'm happy to hand off some of the work, so I can focus on what's a better use of my time.

A real-world example: I've built myself a platform for coordinating coding agents, sharing knowledge across machines and coding harnesses, and tracking usage. This platform is heavily built by Claude - Opus does most of the work, Fable does some optimizations on the big data side of things, and I give direction throughout my work day as I need new tools or MCP servers built. I give a cursory glance at what Claude is doing to make sure it's safe and reasonable, but relistically the risk is low here - if something breaks, my workflow just degrades a bit until I have Claude fix it.

The Last Question

AI has finally allowed me to build all the side projects I've been dreaming of. Every idea can be prioritized, planned, and built. The last question I need to ask myself: Should I Build This?

Coding has never been the only bottleneck to getting an idea out there. For a personal tool or something I desperately need, sure, coding was the main gate between idea and execution. Often formulating the idea, building a plan, designing the product in front of the code, and getting the completed product to market takes just as long or longer than implementation. Also, not every idea is worth building. Sometimes, the effort a project would demand was just enough of a filter to stop me from building bad ideas, or wasting time on useless things.

That's not to say useless things can't be fun, and even worth building! I spent a good chunk of time building my Dead Internet Feed, and also writing about it. I did most of the design on that project by hand, but I wouldn't have had the time to dedicate to such a silly idea if I needed to build all of the agent flows that drive the actual content.

Conclusion

So, do I build every idea now? Definitely not.

Whether I build something myself or AI builds it autonomously for me, it still takes resources. Tokens, hosting, the time I spend designing the solution, chatting with the agent, iterating and reviewing. None of this is free, both in terms of hard cost and cognitive load. It's pretty clear to me that in the days of AI, the path to burnout is shorter and faster. I could build every idea I've ever had, waste time and money on terrible ones, just for the sake of building. Or, I can use my time wisely, focus on refining my best ideas, delegate the ones I don't want to build, and spend time coding the things that are actually valuable to me - things that teach me a new skill, or that I can be proud of my own craftsmanship when they're finished.

I may even throw away more ideas now than ever before, but I'm content in knowing I no longer throw one away because I don't have time to build it.