> Just agentic coding is a huge change
I've been programming professionally for > 20 years and I intend to do it for another > 20 years. The tools
available have evolved continually, and will continue to do so. Keeping abreast of that evolution is an important part of the job. But the essential nature of the role has not changed and I don't expect it to do so. Gen AI is a tool, one that so far to me feels very much like IDE tooling (autocomplete, live diagnostics, source navigation): something that's nice to have, that's probably worth the time, and maybe worth the money, to set up, but which I can easily get by without and experience very little disadvantage.
I can't see the future any more than anyone else, but I don't expect the capabilities and limitations of LLMs to change materially and I don't expect to be left in the dust by people who've learned to wrangle wonders from them by dark magics. I certainly don't think they've "torched decades of best practice in my field". I expect them to improve as tools and, as they do, I may find myself using them more as I go about my job, continuing to apply all of the other skills I've learned over the years.
And yes, I do have an eye-wateringly expensive Claude subscription and have beheld the wonders of Opus 4. I've used Claude Code and worked around its shitty error handling [1]. I've seen it one-shot useful programs from brief prompts, programs I've subsequently used for real. It has saved me non-zero amounts of time - actual, measurable time, which I've spent doodling, making tea and thinking. It's extremely impressive, it's genuinely useful, it's something I would have thought impossible a few years ago and it changes none of the above.
[1] https://github.com/anthropics/claude-code/issues/473