Looks like a solid post with solid learnings. Apologies for hijacking the thread but I’d really love to have a discussion on how these heuristics of software development change with the likes of Cursor/LLM cyborg coding in the mix.
I’ve done an extensive amount of LLM assisted coding and our heuristics need to change. Synthesis of a design still needs to be low cognitive load - e.g. how data flows between multiple modules - because you need to be able to verify the actual system or that the LLM suggestion matches the intended mental model. However, striving for simplicity inside a method/function matters way less. It’s relatively easy to verify that an LLM generated unit test is working as intended and the complexity of the code within the function doesn’t matter if its scope is sufficiently narrow.
IMO identifying the line between locations where “low cognitive load required” vs “low cognitive load is unnecessary” changes the game of software development and is not often discussed.
With LLM generated code (and any code really) the interface between components becomes much more important. It needs to be clearly defined so that it can be tested and avoid implicit features that could go away if it were re-generated.
Only when you know for sure the problem can't be coming through from that component can you stop thinking about it and reduce the cognitive load.
Regarding some of the ‘layered architecture’ discussion from the OP, I’d argue that having many modules that are clearly defined is not as large a detriment to cognitive load when an LLM is interpreting it. This is dependent on two factors, each module being clearly defined enough that you can be confident the problem lies within the interactions between modules/components and not within them AND sharing proper/sufficient context with an LLM so that it is focused on the interactions between components so that it doesn’t try to force fit a solution into one of them or miss the problem space entirely.
The latter is a constant nagging issue but the former is completely doable (types and unit testing helps) but flies in the face of the mo’ files, mo’ problems issue that creates higher cognitive loads for humans.
> I’d really love to have a discussion on how these heuristics of software development change with the likes of Cursor/LLM cyborg coding in the mix
I would also be interested in reading people’s thoughts about how those heuristics might change in the months and years ahead, as reasoning LLMs get more powerful and as context windows continue to increase. Even if it never becomes possible to offload software development completely to AI, it does seem at least possible that human cognitive load will not be an issue in the same way it is now.
I’ve done an extensive amount of LLM assisted coding and our heuristics need to change. Synthesis of a design still needs to be low cognitive load - e.g. how data flows between multiple modules - because you need to be able to verify the actual system or that the LLM suggestion matches the intended mental model. However, striving for simplicity inside a method/function matters way less. It’s relatively easy to verify that an LLM generated unit test is working as intended and the complexity of the code within the function doesn’t matter if its scope is sufficiently narrow.
IMO identifying the line between locations where “low cognitive load required” vs “low cognitive load is unnecessary” changes the game of software development and is not often discussed.