[falcor84]

The first comment to the post referenced Kent Beck's:

> "The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x. I need to recalibrate" [0]

At the time that felt like an exaggeration, but from my own use of Claude Code over the last month, I now entirely agree. My take on this is that we need to educate future devs from the very start as engineering managers - to be concerned a lot more with the "what", the "why" and the "is it good for our expected needs" rather than the low-level "how".

[0] https://tidyfirst.substack.com/p/90-of-my-skills-are-now-wor...

[zeroCalories]

What have you accomplished with Claude Code? Anything you can share?

[falcor84]

Nothing I can share externally at the moment, but I've been using it to almost speed-run through creating a docker compose stack of apps and deploying it on an cloud server. I've had Claude Code write 99% of it (not exaggerating), including app code (mostly in Python), configs, tests (unit, integration and playwright browser tests), deployment scripts and documentation. For this use-case, it's not yet good enough for working fully independently, and I need to steer and course-correct it every once in a while, but it feels a lot more like managing an eager but inexperienced person, rather than a "tool". And while I do make some manual changes every once in a while (particularly in the plan files I have it generate), in the vast majority of cases, I find it more convenient to tell it what to change and have it change it, until I'm happy.

More recently, as I've gained trust in the process, I've also started more and more asking it to manage "subagents" to do some of the lower-level work, and it's been a very valuable power multiplier too, allowing for additional parallelism, without growing the main agent's context.