My leverage is in not depending on "AI", and in possessing the experience, judgment, and discernment no LLM can emulate. That, and my willingness to maintain complete confidentiality; no LLM can be trusted to do that. Also, I can cope with Indian English and not get pedantic if somebody comes to me with doubts instead of questions.
Nor am I worried about being laid off, because I'm not a W2 employee. I'm a 1099 mercenary. I do the job, and then I get paid. If my contract doesn't get renewed, I take a week or two off and get another one. Every company is replaceable to me, and I'm replaceable to every company until they figure out I'm not, and then try to make me a FTE.
At which point I laugh at them, because I have no interest in being their idea of a culture fit, and I'm not passionate about tech. I'm passionate about getting paid. If the next gig wants me to babysit a clanker or give it instructions, that's what I'll do. If it requires reading reams of COBOL and JCL that nobody has touched in decades because it's a mind-shredding post-Lovecraftian horror, I'll do that. Hell, I'll even deal with VB6 if the hourly rate's high enough.
I refuse on principle to give so much as a damn. If there's a paycheck in it, and I don't regard the principal as unethical, I'll take the contract.
#RoninLife
Finally, I still don't think you've explained what it means for people to be agentic-native. I'm not convinced it's even possible to be that. Nor am I convinced that agents can do anything that a sufficiently skilled UNIX graybeard can't do with a makefile, a shell script, some awk, or some Perl. Except pass a culture fit interview at $STARTUP.
Honestly? I'm looking forward to somebody writing "Command-line tools can be 235x faster than your AI agent". Time will tell if LLMs are the new Hadoop.
https://adamdrake.com/command-line-tools-can-be-235x-faster-...
After all, unlike LLMs and agentic AI, command-line tools are deterministic and don't hallucinate more than Timothy Leary ever did.