if you have already been through this process, then ai becomes an incredibly powerful tool. it accelerates your execution because you already have the mental model to guide it, verify its work, and spot its subtle errors. you can drive fast because you have the lane discipline i described in the guardrails i actually use with ai agents. but if you use ai to skip the struggle entirely, you never build those mental models in the first place.
this is why i believe we need to be comfortable being uncomfortable. the awkward, frustrating hours where you are stuck on a problem are not wasted time. they are the exact moments where the learning actually happens.
tension or counterpoint
some people argue that ai can be a personal tutor, showing you the right way to solve a problem so you can learn faster. they believe that by reading clean, correct code generated by an ai, you can bypass the messy trial-and-error phase.
there is some truth to this, but reading a solution is not the same as finding it. if you do not experience the friction of making mistakes and correcting them, your brain does not build the deep neural pathways required to retain that knowledge. you might understand the solution in the moment, but you will not remember it when you have to solve a similar problem under pressure without help. the struggle is not an obstacle to learning, but rather it is the learning itself.
closing
i do not use ai to replace my thinking, but rather to accelerate the execution of what i already understand. the next time you are tempted to ask an ai to solve a difficult problem for you, try to sit with the discomfort a little longer. struggle with the code, read the documentation, and let your brain do the heavy lifting first.
once you have been through that fire, then you can bring in the ai to speed you up. the tool is only as good as the human driving it, and that driver is only as good as the struggles they have survived.
further reading
related on this site