Every morning when my feet hit the floor, I tell myself “I’m the dumbest mother fucker in my industry”. I spend the rest of the day earning my paycheck and leveling up my skills so I can earn one tomorrow. I’ve spent 30+ years in tech. The tech I use to get paid today didn’t exist five years ago. That’s over three decades of ongoing education to stay relevant enough to be employed.
Side projects are how this worked for me.
But I’m going to tell you what everyone else isn’t: your side project is never going to earn a dime. You’re not going viral, and you’re never going to end up with Zuck-level fuck-you money. (You don’t need it.) Your side project gives you the ability to play with new tools and technologies that you can’t use at work. Later, when your company starts to adopt them (or you are at your next job interview), look who’s standing there with the skills and experience to make it work.
Your Average Day
You log in for the day. You check your Jira queue and find the next ticket assigned to you. After reading it, you know what needs to be done. You open your terminal, navigate to the repo directory and fire up Claude Code. You type We need to add a new endpoint to this API service that returns the current status of the order. Claude does its thing, you verify the code and push up the commit to open a PR. This is still a huge productivity boost. You could have written the code yourself. But AI is pretty good, and as long as you keep an eye on the output, it feels like a cheat code.
Your Average Day Leveled Up
You can push this further. You’ve heard of 10x engineers. With AI, people are talking about 100x engineers. They are real. Let’s take a look at the average day for one of them. You start at the same time. You log in for the day. Instead of checking Jira, you already have Claude connected to Jira via an MCP server. You open your Claude session and type Grab every ticket assigned to me this sprint. Organize them by priority and the areas of code they touch. For each ticket, dispatch a subagent with the ticket details and definition of done. When each subagent completes, review their PR. If it looks good, ping me.
Same desk. Same tools. The difference is one prompt handles a ticket, the other handles a whole sprint.
A Sandbox to Build the Skills
I built ClawesomeSingles.com to develop and sharpen these skills for myself. Every part of the site, from the code, to the clawdbot personas, to the artwork, to code review, to deployment was done using agents managed by an orchestrator. Is it a dumb site? Absolutely. Is it still relevant? Nope. But was it a lever in helping me land my current role?
Ka-ching, mother fucker.
Along the way, I learned lessons that still apply today. Monitoring your agents is mandatory. Agents will blow their context and start hallucinating. Other agents cheat. Costs can rack up, and finding the cheapest model that can get the job done can save tons of cash fast.
Extreme Ownership
Your value as a software engineer comes from knowing the difference between solving the problem right and solving the right problem. On another project, I had a swarm working all weekend to build a hunting terrain app. On Sunday, the orchestrator told me all agents were done and all tickets were closed. I launched the site. Not a single thing worked. After digging in, I discovered one of the subagents decided it would be much more efficient to just write the application code inside the unit test. The other subagents saw this pattern and thought it was brilliant as well, so they all did it. The failure here is not the subagent or the subagents that followed.
The failure is on me:
I failed to specify the instructions, constraints, and definition of done so an agent can succeed.
My plan was weak, and the subagents exploited that to get the job “done”.
The Weekend Challenge
You may be hesitant to learn this workflow in your company. Things are going to go wrong, and those things have consequences. That hesitation is smart. You don’t want to be the person who brought down production because an agent went rogue. That’s where the side project comes into play. Pick a project, any project: that one you’ve been putting off, the personal website you’ve wanted to redesign, that one killer idea you think everyone is overlooking. Build it this weekend... but here’s the challenge: Don’t write a single line of code. Don’t dispatch agents one at a time. Create a plan, give the plan to an orchestrator, make the orchestrator do the work.
When things don’t come out as expected, resist the urge to fix it yourself. Ask your agents why they did what they did. Learn the patterns when they ignore or avoid clear instructions. Update your instructions based on what you learn. Put in the guardrails so it doesn’t happen again. Some lessons will apply to this project. Many will apply to every project. Elevate these so they automatically get applied to all agents, across all projects.
I’m Too Busy For Side Projects
That’s a problem that will solve itself during your next job interview or performance review.
X-posted on X here.



