DEV Community: Unmarked-Kyuubi

I Use AI to Code. I Still Want to Understand It.

Unmarked-Kyuubi — Mon, 19 Oct 2026 09:45:18 +0000

AI makes it easier to build software and other cool stuff. But I don't want to lose the ability to understand what I'm building.

I didn't start coding with AI.

I spent a long time writing code, reading documentation and figuring out why things broke you know the old reliable, belt to ass way . That process taught me more than how to get software working. It taught me how to reason through problems.

Then AI became part of my workflow.

I could build faster and explore a lot more unfamiliar tools, and try ideas that would have taken much longer to implement. I wasn't interested in giving that up. I mean why would I I basically 100x my workflow. AI had become a useful part of how I worked making things more efficient and giving me more hands to deal with stuff since I'm an indie dev.

But once in a while I'd look at code an AI assistant had written and realise I wasn't entirely confident I understood it.

I could ask for an explanation. I could request changes and definitely If something broke I could ask the model to debug it.

But what happened when I needed to make a decision the model hadn't anticipated?

That question bothered me. I knew how to build software, yet I sometimes wondered how much of what I was building I could independently reason about. It brew up a new strain of imposter syndrome.

Working on code isn't the same as understanding what you are doing genuinely.

There's nothing new about developers using tools and abstractions they don't fully understand. Nobody knows every detail of every library they use and they shouldn't, most of the time it's not efficient you just have to know what you are doing

What feels different with AI is how easily we can move past the parts we haven't worked through ourselves. It give a dangerous sense of false mastery.

A feature works so we move on and forget about it. The implementation looks reasonable like senior dev code reasonable, so it's easy to accept it. We might not notice the gaps in our understanding until we need to change something or investigate a problem the model hasn't seen before.

I don't think using AI makes someone a worse developer. But we're still responsible for the software we ship, regardless of who wrote the code.

For me, the question isn't whether I typed every line myself. It's whether I understand the decisions that matter and can take the work further when I need to.

The uncomfortable part of vibe coding

There's a particular kind of frustration that comes with building something successfully and still feeling unsure about your own understanding. 😂

You know you have experience. You've put time into learning your craft. Yet an AI assistant can produce a solution before you've had the chance to work through the problem yourself.

You might start wondering genuinely, Could I have figured this out without the model?

I've felt that tension, and I know other developers are thinking about it too.

I don't believe the answer is to stop using AI or return to doing everything manually. I want to keep building with AI. I just don't want my role to become accepting whatever it produces.

Some of the best parts of development come from questioning an approach or experimenting with an idea, even discovering or outsourcing a better solution could give some sort of internal satisfaction. I want AI to help me do more of that and not replace the thinking that makes it possible. Becuase creativity and ideas makes us human.

Sometimes another explanation isn't what you need

Think about learning asynchronous JavaScript.

You might follow a tutorial and understand the example perfectly. Then you encounter a different situation involving several promises, and suddenly you're unsure what happens next.

Perhaps you understood the explanation. But could you apply the concept?

Those aren't necessarily the same skill.

Sometimes the magical missing piece is some concept you haven't understood properly. Sometimes you understand the individual parts but struggle to use them in a new situation. Watching another tutorial might help, but it might also leave the actual problem untouched.

I've become interested in what happens when we stop treating every difficulty as a request for another explanation.

What if, instead, we could identify where our understanding breaks down and work on that specific gap?

That question became part of why I'm building Vectra.

Why I'm building Vectra

Vectra is my first project and its an adaptive learning engine designed to help people move beyond understanding an explanation to demonstrating what they've learned.

The idea is to check understanding and identify gaps in what people know Not just spam with explanations. Vectra's main philosophy resolves around adapting the learning process when someone struggles, rather than simply repeating an explanation and hoping it eventually clicks.

I'm still developing it but it's well enough for peoeple to try, and I don't want to pretend the problem is solved just because I believe in the approach. I want to keep exploring what genuinely helps people become more confident in what they know.

If you've ever understood a tutorial but struggled to use the concept independently, I'd love for you to try Vectra for free

I want us to build more, not understand less

AI has made a lot of things possible for me as a developer. I want to see where these tools take us.

Unfortunately speed isn't the only thing I care about.

I want to build things I can explain and most especially come back to and improve. I want to use AI without feeling that I've handed over the thinking along with the implementation.

And when I discover a gap in my understanding, I want to see it as something I can work on and not a reason to question whether I belong in software development.

That's the idea behind Vectra, and it's something I want to keep exploring as AI changes how we build.

How about you guys when you use AI to code how do you tell whether you truly understand what it has built? I want know what the rest of the community does your workflow might help out others that are struggling so let me know in the comments.

Everyone should have cool approaches to this problem mine was Vectra let me know yours down below

I wanted an AI tutor that would notice when I didn't actually get something, so I built one

Unmarked-Kyuubi — Fri, 25 Sep 2026 06:28:11 +0000

I've been working on something called Vectra, and I've been thinking about this problem for a while:

What if getting stuck while learning something isn't a problem with the explanation, but a problem with the route or pathway you're being taught?

A lot of AI tutoring right now is basically:

"I don't understand this."
"Okay, here's another explanation."

They answer questions. Not teach. Even most of the big shot AI's with learning mode. Just ground you in source material.
And sometimes that works.

But sometimes the thing you're struggling with isn't actually the thing you asked about.

You might be trying to understand calculus but realize you don't have strong fundamentals in algebra. You might be trying to understand a programming concept but be missing something much more basic underneath it.

And if that's the case, giving you another 1,000-word explanation of the original thing probably isn't going to magically fix it and Grounding yourself in source material is only the beginning for an ideal learning experience.

That's the idea behind Vectra.

I wanted to build something that treats learning as less of a straight line and more like a map. Most AI execute tasks linearly.

From your specific problem or prompt models output answers with the solution. But human learning has never been linear or straight one shot answer. It's dynamic more like a curve or a graph. If you have a problem. the solution may be a combination shaky fundamentals and concepts that need resolution before or a certain level of comfortability before even beginning to grasp the main issue.

Vectra tries to figure out where the gap in your understanding is, trace the concepts that lead up to it, and then change the learning route based on what you actually need.

And importantly, I don't expect you to understand something perfectly the first time.

The system can change the way it explains something, try a different teaching approach, move further back into the prerequisite concepts, and then actually check whether you can use what you learned rather than just asking if you understood it.

Because there's a pretty big difference between:

"Yeah, that makes sense."

and

"I can actually do this myself."

Understanding is just the beginning. Application and retention are the ideal ways a learner can truly said they have mastered a concept.

I also built a smaller tool called What Am I Missing? based around the prerequisite system. It's probably the easiest place to poke at the idea without committing to the whole thing.

I also made a slightly ridiculous cinematic teaser for Vectra because apparently building an educational tool wasn't enough, so I had to add a bit of lore and build hype around it. You should check it out it's kinda cool 😂.

And here's the actual thing:

It's still very early. There are bugs, things I'm still figuring out, and probably things that make perfect sense in my head but absolutely do not make sense to another human being 👁️👄👁️

That's actually why I'm posting it here.

I'm looking for people to use it properly and tell me where my thinking falls apart.

In particular:

Did it actually identify something you were missing?
Did the learning route make sense?
Did you understand why it sent you down that route?
Did the teaching approach feel like it was adapting to you?
Where did you get confused, bored, or annoyed?
And did it actually help you understand/apply the thing you were trying to learn?

You don't need to be nice about it. If something feels stupid or broken, tell me. That's much more useful to me right now.

If something feels pointless, confusing, slow, broken, or just plain stupid, please tell me. That's honestly much more useful to me right now than compliments.

I'm actively building and pushing updates, so I'll be around the comments and fixing things as I find them.

If you have 5–10 minutes to try it, I'd genuinely love to hear what you think.

Thanks for giving it a shot.

TL;DR: I built an AI learning engine that tries to figure out what you're actually missing instead of just explaining the same thing again. It's early, it's probably a little broken, and I'd really like people to test it.