Choosing learning over autopilot

RAG, llm pipeline industry just continues in the same fashion of throwing even more glue, insanely slow, expensive, but works due to somehow companies having money to waste, perpetually. Not that much different from the whole Apache stack or similar gluey expensive and slow software.

There is similar mindless glue in all tech stacks. LLMs are trained on it, and successfully do more of it.

Even AI companies just wastefully do massive experiments with suboptimal data and compute bandwidth.

I feel like a lot code is pretty sticky actually. I spend two weeks working on a feature and most likely that code will live for a time period measured in years. Even the deprecation period for a piece of software might be measured in years.

It's kind of amazing that the really mainstream jobs create and pitch throwaway code, while a few key niche jobs, with little demand, can really create enduring products.

Kind of like designing a better social media interface probably pays 100x what a toilet designer would be paid, but a better toilet would benefit the world 1000x.

'If I had to guess, I'd say that is the main "skill" being lost'(to endure frustration).

I think this might be true for you, but for less experienced and new developers, well, they actually won't get to that stage because their 'learning' is basically prompting and they have nothing to forget nor remember. And that might be bigger issue.

Yeah I accept it , because most of the new gen are actually trained to be ,copy and paste and that is where the actual capability of the critical thinking is lost and I would accept your words being a new dev, because it feels cool after using ai for our work to complete and then after it completes and If I just come back and see ,what I have really did is,it feels like am I really unproductive? Actually no,but it is a bit like yes, because when we are good at driving the car and then help with the auto pilot is so good ,but when we don't actually know ,how to drive the car and then it feels like ,we are in the control of the autopilot disguised as the automation,what most people won't admit and I wanted to say that,we must actually have some knowledge on what we are trying to produce because, without any prior knowledge on what we are trying to do,we are just using the autopilot not our brain for the brainstorming ideas,even though we can get and we can take the help of the ai to convert that ideas to execution, atleast we must learn what we have been going through

I wouldn't call that cursed but useful tooling usage. Had the same scenario where I wanted to work on a tool for a project written in Go, of which I know next to nothing. Claude code was able to spit out 100's line of code that worked and I (almost) understood and could explain what was happening where and why, but I had no chance of debugging or extending it on my own.

I've limited myself to only use Claude's webchat to do almost exactly as you've mentioned except creating snippets, it can only explain or debug code I enter. I prompt it to link relevant sources for solutions I seek. Plus it assists me subdivide, prioritise and plan my project in chunks so I don't get lost.

It has saved me a lot of time this way while still enjoying working on a project

Following up - I am the most excited about using computers because the barrier from intent to product are being dropped. At this point my children can 'code' software without knowing anything other than intent. Reality is being made manifest. Building physics into a game would take a decade of experience but today we can say "allow for collision between vehicles".

If you have ever gone running the ability to coordinate four limbs, maintain balance, assert trajectory, negotiate uneven terrain, and modify velocity and speed at will is completely unknown to 99.9% of mortals who ever lived and yet is possible because 'biological black box hand wave'.

I can't believe I took a mandatory technical drawing class.

Are you arguing that we are no worse at building bridges than we were 100 years ago?

Imho this AI "revolution" will be the death of non technical middle management first.

Engineers who practice engineering (as in thinking about the pros and cons, impact, cost) will simply get to work closer with relevant stakeholders in smaller teams and the role of the project manager will start to be seen more of a barrier than facilitator.

While I tend to agree, I think it's super easy to think you are using AI and being productive and then hitting a brick wall once all the things start failing because the system is not internally coherent.

This is also what I often wonder.

Imho, AI is a multiplier and it compounds more as seniority grows and you know how to leverage it as a tool.

But in case of juniors, what does it compound exactly?

Sure, I see juniors being more independent and productive. But I also see them being stuck with little growth. Few years ago in an year, they would've tremendously grow at least on the technical side, what do they get better at now? Glueing APIs via prompting while never even getting intimate with the coding aspect?

I'm only a data point, but some years ago I spent a whole year learning a mathematical book above my level at the time. It was painful and I only grasped parts of it.

I did again the same book this year, this time spending much time questioning an llm about concepts that I couldn't grasp, copy pasting sections of the book and ask to rewrite for my understanding, asking for fast visualization scripts for concepts, ask to give me corrected examples, concrete examples, to link several chapters together, etc..

It was still painful, but in 2 months (~8h-10h a day) I covered the book in many more details that what I ever could do some years ago.

Of course I still got some memories of the content from that time, and I'm better prepared as I have studied other things in the meantime. Also the model sometimes give bad explanations and bad links, so you must stay really critic about the output. (same goes for plots code)

But I missed a lot of deep insights years ago, and now, everything is perfectly clear in two months.

The ability to create instant plots for concepts that I try to learn was invaluable, then asking the model to twist the plot, change the data, use some other method, compare methods, etc..

Note: for every part, when I finally grasped it, I rewrited it in my own notes and style, and asked the model often to critic my notes and improve a bit. But all the concepts that I wrote down, I truly understand them deeply.

Of course, this is not coding, but for learning at least, LLMs were extremely helpful for me.

By this experiments I would say at least 6x speedup.

Honestly I feel I have never learned as much as I do now.

LLMs remove quite a lot of fatigue from my job. I am a consultant/freelancer, but even as an employee large parts of my job was not writing the code, but taking notes and jumping from file to file to connect the dots. Or trying to figure out the business logic of some odd feature. Or the endless googling for responses lying deep inside some github issue or figuring out some advances regex or unix tool pattern. Or writing plans around the business logic and implementation changes.

LLMs removed the need for most of it which means that I'm less fatigued when it comes to reading code, focusing on architectural and product stuff. I can experiment more, and I have the mental strength to do some leetcode/codewars exercise where incidentally I'll also learn stuff by comparing my solution to others that can then apply back to my code. I am less bored and fatigued by the details to take some time more focusing on the design.

If I want to learn about some new tool or database I'm less concerned with the details of setting it up or exploring its features or reading outdated poorly written docs, when I can clone the entire project in a git subtree and give the source code to the LLM which can answer me by reading the signature, implementation and tests.

Honestly, LLMs remove so much mental fatigue that I've been learning a lot more than I've ever done. Yet naysayers will conflate LLMs as a tool with some lovable crap vibecoding, I don't get it.

I learned a fair bit about architectural choices while vibecoding because if you don’t spec out how things should work it goes off the rails fast.

Haven’t found a good way to learn programming language basics via AI though