Skill issue

Last week I had to extend complicated, unfamiliar Helm chart. While I've discussed approach with my carbon-based coworker, actual editing and extending was done by Claude Code.

Refactors and renamings were tedious, but Claudius did them in no time. During the review we decided to significantly change the way chart is organised. Claude executed the changes perfectly.

If I had to fully understand the working of the chart, it would have taken me a good part of the week. Making the changes – another day or two. Significant rework midway – I'd probably be too demotivated to do it.

With AI assistant, I have finished work in two days. Tested, simplified, cleaned up.

But I have no more skills than I had before. I do not understand this Helm chart much better. I've spotted a thing or two during the review, but in no way I have a fuller comprehension. AI let me borrow expertise without acquiring it.

Next time when I'll have to work on this chart, I will use Claude again. I've got softly locked-in in using AI assistant to work. And I'm not happy with that.

Now, my employer sees this as an acceptable tradeoff. Paid for some tokens, unlocked couple of my pricy senior-level hours to do other stuff. It balances. Subsidising Scam Altman and his ilk is a money well spent from this perspective.

We are at the point where we offload more work to so-called AI. Those are just tools. You know what more primitive tools we had before? Assemblers. Then compilers. No one writes machine code by hand anymore. Maybe someone despairs because of it, but people commonly writing binaries are more likely to be dead by now.

No one treats high level compilers as fad and stuff kids these days do. It's just a tool, expected to work and dissappering into the background.

There's one important difference. We do not have to pay for the compilers, thanks to work of the GNU Project last century (gcc) and Apple more recently (llvm/clang). But at this point we are paying for tokens. I expect that as we cannot fathom working with any codebase without a compiler now, a LLM-driven assistant will be required to tackle bigger projects in near future.

To be free, we need to be able to run assistants at reasonable cost. Free is reasonable. Not paying through the nose for electricity is reasonable. We can hope for good quality chinese models to be available for free. We need a free toolkit revolution again. It was GNU Compiler Collection for the previous generation. Would it be Kimi now?

And we need the proprietary AI bubble to pop and rationality to return.

066/100 of #100DaysToOffload