In February I came back to work after a 6 week sabbatical to Opus. It has completely changed the way I do my job. It allows me to spin up agents which investigate code so I don’t need to keep everything in my head. It helps prototype designs for me when I’m trying to think about architecture. It can automate trivial tasks which I wouldn’t normally have time to do.

I would have considered myself pretty anti-genAI prior to my sabbatical. I still think some use cases are lame or morally wrong. But, I genuinely get some value from it as a tool in the current price point.

That being said, it stresses me the heck out! I became a software engineer because I enjoy coding and I genuinely find it relaxing. I often equate programming in Haskell to assembling puzzles.

My primary issue has been that we’ve moved our entire bottleneck to review. Self-review becomes absurdly important and I feel like most people weren’t spending enough time on it pre-Opus. Code review on a thousand line pull request was already hard and now I probably spend twice the amount of time. Code review is hard, much harder than coding. I often can’t tell if someone is doing something on purpose or the AI just did a drive-by. How are we going to improve our codebase overall if I let anything slip through?

My next concern is around “efficiency-maxing”. I feel obliged to have my agent doing something useful as much as possible. Did I remember to check on my agent before going for lunch? I’ve hit a blocker on my tertiary side project. Okay, agent please spin up a draft design document so I can unblock that. My organization doesn’t force me to do this, but I feel like I need to be more efficient. My damn brain chemistry mixed with this tool is not good.

To be clear, I think efficiency-maxing is not token maxing. I’m not trying to break the top 10 users at work nor has that been an incentive. I worry that I’m not very good with the tool. When I’m not very good at something the path forwards is pretty clear… practice. When I practice programming or my drums, I get better at those things. When I practice genAI, I get better at prompting genAI. I also end up efficiency-maxing my personal projects. I’m literally building two plans for two projects while writing this blog post. My time on this planet is limited and I’d like to spend it learning things which make me happy.

I know a lot of people who are stressed out by this technology. This blog post isn’t ground breaking in any way but it helped me. Heavily subsidized compute can’t last forever. Thanks for reading and feel free to send me a message on BlueSky if you have any questions, corrections, or want to tell me I’m wrong. I’m always happy to learn new things.

One hot take for the road, Erlang is better than your favorite programming language.