Something that has fascinated me about AI adoption in software development is the way that the labs have gamified it with their subscriptions and rolling rate limits. Once you sign up for that $200 plan, your head naturally goes to figuring out how to maximize its value. How do you keep your agents working while you sleep? How do you make sure work is shaped and ready to throw at Claude the moment your limit resets? It can be addictive and, eventually, stress-inducing.
I recognized pretty quickly that I wanted out of that anxious cycle, and I gave myself a challenge. Find a way to be okay with walking away from the keyboard (and phone) for an entire weekend or cancel the subscription. It didn't take me long to figure out the right shape of the solution to the problem.
I created a list of the things I wish I didn't have to spend my time on every day and I created a way for my subscription to do as much of this work as it can. By simply offloading this work, the subscription feels exceedingly valuable even if I don't open Claude myself all week. Here's what my suscription does for me every day:
GitHub Issue Handling - When a new issue is opened, a workflow is triggered that categorizes and shapes the issue. Agents determine whether or not there's enough context to proceed. If not, they tag me for additional information in a comment. If so, they write tests, write the implementation, review, validate, verify, and refine. In some low-stakes projects, the work just gets merged without me ever knowing the issue was opened. If they ever get stuck, I get tagged in a comment and can jump in to help when I get back to my desk.
Ongoing Technical Debt Analysis and Repair - We're moving very fast these days. We're not always reviewing things like we used to. We're letting agents make some decisions we probably should be making ourselves. This leads to a bloated, overly complex app if left unchecked. To a large degree, this can be solved by simply having other agents review recent work and find opportunities to simplify things. Identifying places where new patterns were introduced that don't match with how we've always done things turns out to be something pretty easy and high value for these agents to do. They present me a list of their findings and with a couple of clicks, the ones that matter to me become GitHub Issues, and they sail through that GitHub handling workflow I described above.
Automatic Exception Handling - Once I had the GitHub Issue handling workflow working to my liking, making my app self-healing was as simple as turning on the GitHub integration in Honeybadger and installing the MCP. Now exceptions open issues, my agents pick them up, write tests, write the fixes, and (often) simply merge those fixes for me when all the validation passes.
Slow Route / Query Handling - The same is true for slow queries and routes. By simply having my monitoring software open issues for anything over a threshold, I toss everything into my automated GitHub issue workflow and most of these problems go away.
If you're staying awake at night trying to figure out how to get Claude to keep building while you're away from your computer, maybe it's time to ask a different question. Would you be able to build faster during the day if you had a reliable janitor following you around and cleaning up after you while you're away? It's working well for me so far.