Hey there,

This quote caught my eye today:

And I've been thinking a lot about this lately.

The more we use AI tools, the greater the risk that we lose our core skills. It's also exhausting. And it strips out the fun that writing code used to be.

Nobody has a clear answer yet. Some (even well-known) programmers now say they don't read the code anymore; others have dropped AI entirely and gone back to writing code by hand. Here's where I've landed.

I still believe in a balanced approach where I try to use the tools in a disciplined way, but it's not easy, because they are slot machines.

If anything, I should build some tooling to help find the balance I think.

I could link a pile of talks and articles I've been consuming, but I'd rather have a back-and-forth with you.

How are you balancing this right now? Hit reply and let me know. I'm eager to learn from you, and to share my own thoughts and experience using these tools now for almost 4 years (with agents this year being the greatest shift).

At the end of the day, the tools are the tools. The fundamentals of software dev are to solve customer and user problems.

That hasn't changed and it's inherently a human craft, where we need to communicate and iterate a lot.

The tools can help with features and make us produce code faster, but they can't own the architecture or assert the real intent.

So the main thing for me is to always stay in the driver's seat and use AI as an accelerator.

The other thing I keep in mind at all times: these are nondeterministic systems, so what looks plausible might be (very) wrong. Someone put it well in a comment recently:

"the actual work shrinks but the review load doesn't — you still have to read every diff closely enough to catch the plausible-looking mistake. The pressure just moves from typing speed to judgment speed."

Related article: Judgment: The Skill AI Can't Give You

That's it exactly. Maybe we'll get to more sophisticated eval loops and test harnesses, but for now I think the best bet is to stay close to the code.

Maybe not handwriting it (although I loved that part), but at least reviewing it closely and really challenging anything AI gives back, because ultimately you own what you produce with it.

Looking forward to hearing from you, and till the next one.

— Bob

P.S. AI poses another problem: the paradox of choice. If you're not sure what to learn next, what about flipping the question?

What to build next?

I've made a tool for this that will help you decide + scope it out -> https://scopefirst.dev/

Let me know if it helped you scope out a cool idea. 🚀