AI, Learning, and the Things Worth Doing the Long Way

Last week, we switched our kids from a shared room to separate bedrooms. As part of the effort, I bought a load of new furniture from IKEA—because, you know, we’re fancy like that. At checkout, IKEA gives you the option to hire TaskRabbit to assemble the furniture. Given the 6+ hours it took me to assemble the beds and dresser, that certainly would have been a good deal and an efficient approach.

But even though the task was arduous, it was actually fulfilling—the relaxation of turning off my brain and just following instructions, the satisfaction of seeing tangible results of my effort, and plenty of time to catch up on podcasts and Netflix. Thus, outsourcing the effort to someone else would have been a mistake. It would also have made me less competent at repairing the furniture whenever the kids inevitably break it. 

The assembly outsourcing question was similar to one I’d been asking myself about AI over the last few weeks. What knowledge and skills can people only learn by doing it the hard way, making AI tools counterproductive?

Surely, AI can help you summarize a long text, but how much more do you learn from a slow read? AI can debug your code, but isn’t there something important you learn about a discipline when you have to pull an all-nighter to find your mistake?

Really, it’s a question about experience. Can anyone really “know” something without having seen it up close and for a long period of time? For example, I’d never trust a doctor who’s read an AI summary of anatomy but has never seen the insides of a human body or who hasn’t personally made enough mistakes to know when their technological tools may not be providing the best information. 

The journalist Derek Thompson synthesized this dynamic on Pablo Torre Finds Out when referencing a blog post about the distinction between a job and a gym. “They said, with a job, the point is to get the work done, but at a gym, the point is to lift the weight. You can’t go to the gym and ask somebody else to bench 135 lbs. and tell yourself that you bench 135 lbs because you asked somebody else to do it.” He said that’s just a formula for your muscles to atrophy. 

I agree.

So how do we know where it’s okay to let AI “lift the weights” for us and where it’s important for us to use old-school learning methods?

Thompson suggested that our goals should drive what situations are more like a job and which are more like a gym. “You have to kind of feel it for yourself. When am I leaning on this technology to do my job in a way that’s keeping me from building the kind of cognitive muscle that’s necessary to get better month after month? And when am I using it to just help the work that I’m doing become richer?”

For me, drawing the line starts with asking: On what topics do I want an original perspective? If I want to develop a perspective that’s different from the norm or what can be found on the general internet, it makes sense to engage with the topic differently than others, which inherently means direct contact, offline thinking, and the more arduous paths for learning. 

The arduous path may also be required for competitive reasons. For example, in a world in which all of the recipes and techniques are knowable, why would anyone go to one restaurant over another? The answer is likely tied to whether the chef has taste or a unique perspective on the food or the dining experience. And if we’re the executive chefs of our careers, we shouldn’t outsource the critical factors like building taste to the same tools that everyone can access—AI or otherwise.

Finally, aside from drawing the right professional lines on using AI, there might be an even more important personal lens: How do we avoid accidentally allowing AI to steal truly enjoyable activities from us?

Last weekend, I learned this lesson because I used Claude to help me plan a dinner party—mostly because I was experimenting with an unfamiliar cuisine and was behind on planning. Claude helped me quickly land on recipes that would work, which was definitely useful. However, I realized afterwards that much of the fun I get out of dinner planning comes from rereading cookbooks, searching for the right solutions, and imagining how to combine recipes and techniques. Thus, not only did using AI reduce the potential for learning about the cuisine by getting me to the destination without the journey, but it also eliminated the fun!

This is the same reason why hiring someone to assemble the IKEA furniture didn’t make sense. Sometimes, hard work is genuinely satisfying, and the friction is the point.