Just keep getting better at programming, and have a deeper understanding of a topic like how a computer works low level. That takes time, focus and effort but it will be worth it. It's knowlodge that will remain, who knows MCP how long will be there? Also, even if we reach 99% of the code generated by machines(and I don't believe it if not for trivial code) you'd still need deeper skills to understand it, not only semantically but looking at the big picture in terms of architecture and business/design implications. My suggestion is the contrary of many tech influencers, do not deep dive into prompt engineering or similar stuff, that's the trivial part, if you fail at prompting don't let them convince you you have a skill issue, you're not paid to chat, you're paid to solve problems, prompting is trivial, you must understand problems and requirements deeply. I actually refreshed my high school math and with it I've been able to do so much, from AI basics inner workings to computer graphics, there is so much in core knowledge that is underrated these days. I think I'll soon start the Computer Enhance course from Casey Muratori for low level stuff and performance. Since the advent of LLMs I actually wanted to learn more than before, it has been beneficial to me

Just regular software skills:

* Transmission engineering - reverse engineering, extending, and creating transmission schemes.

* Test automation - Automate the shit out of your applications and job with home grown test automation capabilities (not some big third party test automation tool)

* Performance - Actually measure things with actual numbers to show employers just how shitty/awesome their applications are with actual evidence.

* Knowing your software platform at the lowest level, not some vanity or unnecessary abstraction bullshit

* Knowing people. If you are bad at the soft skills, or are too neurodivergent to actively listen, AI probably should take your job.

Learn prompt engineering and how to effectively use AI coding assistants… that's immediately useful and will save you hours daily.

Vector databases (Pinecone, Weaviate) and building RAG systems. Tons of companies need this now and most devs don't know it yet.

Understanding model fine-tuning and when it's worth it vs just better prompting. Also get comfortable with AI ops - monitoring model performance, dealing with hallucinations, cost optimization. The boring stuff that actually matters in production.

And yeah, just stay curious and adaptive. Half the tools we use today didn't exist 18 months ago.

Theory is our best device for cultivating good judgement. My advise is to deeply invest in understand computer science and mathematics. Those are the foundations which will make it most likely to understand new application landscapes based on them.