AI is getting so good now (at the end of 2025) that I now have a new, primary recommendation going into 2026:
Think very carefully about where you get help from AI.
I use a Job vs. Gym analogy.
In the first case we just want the output, and in the second the whole point is to do the work ourselves.
Going forward, and especially as AI improves, it's critical that we don't confuse these two.
Step 1 is figuring out which are which for you.
For me, any sort of:
...are all Gym tasks.
These are core to my how I see myself, and I want to not only maintain my skills with doing these things, but I want to get better at them over time.
That beign true, because my work is largely cognitive, and the whole point of AI is to magnify my ability to do that work, I inevitably will use AI to do many of these Gym tasks in a given day/week/year.
I've started building a system into my customized AI stack that functions not just as a worker, but also as a tutor.
Currently this takes the form of a weekly session where my Digital Assistant, Kai, can look at all the Gym tasks that he performed for me and can interrogate me on how I think it was done, how I think the code was generated, what I think the architecture was, why I think he made those decisions, etc.
Okay, so over this last week, you had me do x, y, and z.
Now it's time for me to show you how I did it and hit you with questions to ensure that you fully understand what was done and why I made the decisions I did.Kai
From there, we can go into an interactive back-and-forth, getting to first principles all the way down to code-level specifics or whatever.
This is currently done via a Claude Code skill, but I'm experimenting with some other interfaces and interaction modes as well.
So here's what I practically recommend going into 2026 and beyond.
Keep the robots out of the gym.