I have a math-Ph.D friend who complains that he hates taking math classes that use “PowerPoint”. It is a frustating phenomena that I can immediately relate with. But I have never been giving too much thought about until now. I have the same feeling with the AI-coding agent tools. The fatigue and frustation feels surprisingly simialr.

Going back a bit, this semester I take a weed-out gradudate-level theoretical computer science course: Algorithms. For those who are not familiar , this is the class where we write proof and arguments for the correctness and efficiency of the algorithms (zero programming involved). I don’t know what exactly “mathematical maturity” means (stated in “recommended” prerequisite) until I know I may not really have it.

Joke aside, I find myself falling behind the lecture in real-time i.e., the 30-minute video requires me to watch like 50 minutes or more. I don’t know why. Yes the class is hard but why there is so much lag.

“Repetition is pursuasive. It is mother of all learning. You never go to church and expect to hear the 11th commandment - but you still go.”

When I learn new things, I always go to Youtube to learn the same materials from different people. In particular I found the lecture series from Tim Roughgarden from Stanford. The lecture is delivered via the traditional chalk and blackboard. I found the information flowing more easily (to my brain). In contrast, the class I have hard time with is delivered with PowerPoint.

Turns out, human cannot absorb the information like sponge. We need time to parse it. I totally understand that professors do not want to “waste time” writing on a board so he/she might prepare it upfront with slides. But that inadvertently puts more cognitive load to students (and thus unpleasant in-class experience). Of course, this “cognitive load” must happen regardless as this “friction” is critical for learning (I guess it makes neoron connection in our brain somehow).

Back to day-to-day work, I think the reason engineers complaint about the AI-coding tools has an overlap with what I just talked about. If you think about it, the AI presents the code it writes as “one-single patch”, and we go review it. It is like the PowerPoint again. One giant text for you to absorb. Compare this to when you write a program (or do pair programming), the code is written line-by-line. It is like learning with the chalk board. We absorb information better, and the comprehension devlops easily. And this feeling of flow is satisfying. I think the same thing can be said on any corporate presentation & meeting that uses slides presentation, especially in complex topics.

As of today, I don’t think there is a way around this. We are the product of the evolutionary process, and somehow nature gives us this much capacity. Yes, we can improve that practice through time (that is called competence & expertise). For the software though, as long as we care about the maintainability of the software we write (at least someone has to got to understand the thing - especially when it breaks), it doesn’t really matter how many agents we have, our understanding (and speed of getting to that) is a bottleneck.

If I put my investor hat on just a moment, I think the small-mid-size, specialized SaaS business will still be around. Think about the software for oil and gas exploration or manufacturing. The companies that license these software will not one day develop their own full-suite software. It is not their competence domain. The long-tail nature of software is also a thing (again, when it breaks)- we can see lots of banks still use IBM computer system.