if a task is repetitive, routine, mechanic, it should be hard programmed, no llm/genai needed. that will be efficient, stable, and scalable.

if a task is abstract, ambiguous, high level, open ended, one-time, explorative, then llm/genai is needed.

genai can help to build and improve code.

many operational tasks are routine and repetitive, those tasks should be programmed

writing, solving a new problem, improving are not repetitive, they are explorative

program is like cache, data, that are repetitively retrieved (and then executed). it’s data for a certain procedure. potentially highly optimized.

evolving code is explorative, genai can be a good helper or sometimes can do better than human

so genai can be viewed as a good tool to speed up building

but not good at executing

it’s a multiplicator to human intelligence

human + genai can build more things, better things, faster

then it poses challenges on resources

as running more things needs more resources

that’s why power shortage is the bottleneck

as genai is 1) not efficient, 2) human are generating more resource demand (solving more problems) by using genai, 3) more things build need more resources to run.

but some resources are limited

like human attention

so content created by genai could potentially leads to worse overall quality

as genai content is cheap, and more worse quality content are made

code has the same quality risk

time will be shifted to code review, high level design and quality eval

research / science / solving problems has no limit

once a problem is solved, it’s solved, no need to resolve the problem

and there are limitless problems

so they will benefit the most


<
Previous Post
Agent skills
>
Next Post
refactor ai code