It's been amazing to vibe code prototypes in any stack, but when it comes to building something reliable/scalable, I couldn't effectively guide the agent unless I knew the technology. And the scariest part is that I'm seeing a lot of my technical skills decreasing due to AI coding.
Reflecting on my journey, I also worry about how the new "AI native" generation of software developer is going to acquire technical depth.
So I built fata.dev: short daily spaced-repetition sessions for programming skills (Rust, CSS, React, Python, TypeScript, Architecture).
You can try it in the browser with no signup: https://fata.app/courses
It's an offline-first mobile app built with Capacitor, RxDB and Firebase. The first courses were painfully written by hand, but most content is now AI-generated. It takes about 3000 LLM calls to generate a course, and every code samples goes through compilation, linting, unit testing, AI and a final manual review.
Would very much appreciate any feedback on the product & website, what works and what could be better. Thanks!
Computer scientists do science. The end result is a new algorithm or theorem, and they work with lots of math, proof systems, formal languages, and so on. This is a profoundly creative and collaborative process.
Programmers build products for end-users. The end result is the user experience, the bottom line, stars on Github, or whatever. This is also a process that constantly needs feedback, from teammates and from clients.
These are broad categories and sometimes they overlap: an operating system or database engineer might do a bit of both, depending on the hardware targeted or on various constraints. Sometimes you need to be a bit clever to get to your product goal.
This cleverness has nothing to do with memorizing keywords or solving interview gotcha questions. This cleverness can't be augmented by AI, and it can't be rotted by AI: it's something that's innate to people. Their internal curiosity just happens to manifest in the context of engineering because they love the craft.
In many ways, it's like reading a book. You can't "make someone" enjoy reading, some people are just boring and uncurious. In many ways, I feel that people that are threatened by AI are like this. AI is an incredible tool, but it feels like an autistic junior engineer. Sure, it might have memorized more library names or more syntax or can reference some little-known fact, but it fails to understand basic data flow. It doesn't grasp ergonomics, ease of use, etc., etc.
A tool like fata is great if you want to be an AI, but it won't turn you into a good engineer.
That being said, people are also inspired by other people or by products, and there are ways to make the learning more effective.
No tools is going to turn you into a good engineer, fata makes it easier to practice regularly and it's that practice that will make you improve your skills
I hope you factor in the time you spend doing that spaced repetition thing when you evaluate your AI productivity boost...
I stopped writing C++ professionally in the early 2010's. I came back to it in the early 2020's and it was extremely easy to pick up. Less like riding a bike, more like driving a car. The car got better over the last 10 years, but it still takes a bit of time to fully meld with the machine. The skills didn't rot, they just fell behind, and catching up wasn't a big deal.
I think the only "skill rot" people are facing today when coding by hand vs by agent is that you know you're doing something the hard way when you know there is another path of least resistance available - and that creates internal resistance to doing it the hard way. It's a mental block, not skill rot.
The one area I do find myself questioning is that my low level tech skills aren't growing - I'm building bigger projects that do more stuff across more areas, and as a result I'm touching lots of new technologies from a birds eye view and directing the agents to do stuff with them. I'm not personally gaining that experience of using those technologies, whereas if I were doing things the old way I would be deep in the weeds and have to know them in depth. But if I were doing things the old way I probably wouldn't be building such ambitious projects as a solo dev either.
IMO it's more like digging out the foundation with shovels vs using an excavator. You don't lose the shoveling skill by using the excavator, but you sure feel like a chump using a shovel when there's an excavator parked and ready next to the job site.
As a non-programmer who is now doing a lot of vibe coding for myself these days (and feeling very scared about which of my vibe coded things will just randomly break one day haha...) I feel like what I'm missing isn't necessary language syntax skills, but rather the higher-level planning/organization/architecture capabilities that a seasoned software engineer acquires over years of seeing what scales vs. breaks in the real world.
Is there anything you would recommend I learn to start acquiring that knowledge (aka. the higher level knowledge a "vibe-coder" who doesn't manually code by hand would need)? I'll try out your clean architecture course, but would love to see more along those lines :)
Obviously I know nothing will replace real world SWE experience, but anything that could plug that gap even a tiny bit would be helpful and is something I'm willing to invest in.
Cool app though! I love the idea of coding Duolingo. But I think the price is going to turn most people off. Another problem I see is that it's not really clear how this is that much different from other similar apps out there.
I do think you have a cool niche here that isn't being covered, maybe your landing page should include more information about the Duolingo-esque features? I just know that pulls me in a lot.
> fata builds the full-stack fundamentals to lead a coding agent through frontend, backend, and the technical design that holds up in production.
"Fata builds" sounds like it does the work, not like it trains you how to do the work. With how many "AI does the work for you" promises out there, it's confusing terminology
Anyways best of luck I suppose. I would rather run a drill press through my hand than use AI agents to write code for me, so I'm not your target audience
The main point is that these things are implementation details like assembler. Very interesting in and of themselves especially if you are on the spectrum and occassionally need to be looked at, but the knowledge is instrumental at best and useless baggage at worst. Validation is what matters. Code is useless cruft. If you can replace it all with "348348-23439 CALL(X(DDD)D)" somehow, that's great, as long as it works. I don't know why people get sentimental about wrangling syntax. We're in automation, we've been doing it to other people. We're now doing it to ourselves.
You see this point often because it is blatantly obvious for anyone that didn't build their identity around proficient wrangling of specific families of syntax.
I suspect the era of LLMs coding in "our languages" is a temporary abberation anyway. It's not necessary at all and in fact actively hampering their effectiveness. I'd say enjoy it while it lasts.
Internet at its worst.