Learning 10x Faster with AI?

Yup, that's a clickbait hook :D If you regularly read my blog, you probably know that I almost never use aggressive hook lines like this. But honestly, this time I had to, because it felt genuinely fitting for what I want to share about my journey of learning AI over the past year.

Starting from Zero

I started tinkering with vibe coding over a year ago. Back then, I didn't even understand what "vibe coding" really meant; I just saw developers sharing about it everywhere, so I defaulted to assuming it was some new coding paradigm for devs (?) At that time, ChatGPT was still an undisputed juggernaut dominating the AI landscape before Claude and Gemini entered the ring.

Back then, I had two immediate problems and honestly had no idea how to solve them:

How to revamp my personal blog so it looked decent and ran smoothly (I was using Wix, and anyone who has used it knows how frustrating designing on Wix can be—my website back then looked terribly clunky, which was pretty embarrassing considering I also do UI/UX design...).
An automated workflow to refine draft content for me (yup, even a year ago I was already finding ways to be lazy with writing :D).

The result? The website was still rough because I tried following what AI instructed and failed—from tweaking code snippets to following SEO optimization checklists. The content refining workflow fared a bit better, though errors still popped up regularly. But for that content workflow, AI pointed me toward using Google Apps Script with Google Sheets to manage it. To me, that felt like a whole new horizon (mind you, I had zero coding background. My academic base is in Banking and Finance, not IT).

After that milestone, I dove deeper into figuring out how to build and automate things with Google Apps Script. Then Gemini and Google AI Studio launched. Well, things started moving very fast—building standalone apps and web prototypes took just a few minutes. Dopamine kept pumping, and I jumped from one mini-project to another, only to realize it wasn't enough. Almost everything I built was solely for my own use; I couldn't share it with anyone (even when AI later suggested Vercel and Supabase), and it certainly couldn't scale...

The Big Turning Point

Around October of last year, I moved back to Hanoi after two years living in Saigon. That was when I made what I consider the second biggest decision of my professional career: going full-time freelance and dedicating almost all my time to learning AI.

By this point, I had gradually moved away from pure chat models (at the time, Claude, Gemini, and ChatGPT were strictly conversational models—they weren't nearly as powerful or equipped with tools and integrations as they are today) and transitioned into Visual Studio Code with Copilot and code-assist extensions. I started learning how to code with AI, beginning with small standalone websites and isolated utilities.

I had one major advantage: I already understood software delivery processes, from architecture and database concepts to UI/UX design. But when it came to infrastructure, I was virtually blind. I knew the terminology, what components did in theory, and where they sat on a diagram. But I had no real understanding of how they tied together across the entire lifecycle, or why they were necessary.

So I decided to "un-learn" what I thought I knew, start piece by piece, and have AI guide me step by step through hands-on execution (with AI instructing and guiding while I performed the actions myself, rather than having AI just do it for me).

Throughout that process, I encountered an overwhelming amount of information where I was completely lost—both in AI and in core software engineering (yes, even software engineering, despite having worked in the tech industry for 13 years at that point). It was genuinely exhausting and discouraging. I was stuck in a maze of disparate concepts without knowing how to assemble them:

I knew the concept of development environments, but what is their actual nature? Why do we separate them, and why are there so many environments?
What exactly is CI/CD? How does automation work under the hood? When engineers say they "build," what are they actually building? What does deployment mean, and how does it happen?
Once a product is built, how do you actually deliver it to end users? What is a server? What is a VPS? And what about Docker? Can't things just run standalone without containers?...

A barrage of questions surfaced as I worked. There were moments when I felt like quitting altogether. But eventually, I found a methodology that kept me motivated to push forward.

Rescuing Myself to Keep Going

My approach was simple: Break things down, and build a lot. "Breaking things down" meant isolating every single step and tackling it in small increments. "Building a lot" meant running multiple projects simultaneously, ranging from the very simple to the complex. Why?

Breaking tasks down keeps the brain from being overwhelmed, allowing you to set aside anxiety about the macro-system and focus on learning the immediate concept first.
Building multiple projects—both large and small—provides steady dopamine hits from finishing small tools and seeing them work, or shipping a simple site and seeing it live online.

Step by step, I made countless mistakes, built relentlessly, and absorbed an enormous amount of knowledge along the way.

To support this learning journey, I invested in a dedicated desktop machine to serve as a home server, bought subscriptions for various AI models, spun up VPS instances, and bought domains just to experiment. Eventually, I was able to configure my own CI/CD pipelines and set up isolated development environments. I finally understood why engineering teams set things up the way they do. Today, I even run a self-hosted Gitea instance for personal development—managing three separate workstations connected through an internal CI/CD pipeline with Gitea and Coolify (for web apps), alongside Woodpecker CI for heavy build workloads. I handle full production deployments: provisioning VPS servers, migrating DNS through Cloudflare, configuring Nginx reverse proxies, and managing live releases.

Why Didn't I Flex Any of This?

Reading this far, you might wonder: "If you learned and built so many things, why didn't you share flashy demos or flex on social media like everyone else?"

Because I was busy. Genuinely, intensely busy. Here's a glimpse of the evidence (this project management tool is something I built just a month ago, and this represents my workload over just the past month alone): tons of tasks, and quite a few active projects. Nowadays, my routine is much like an engineer's—if a client reports a bug in the middle of the night, I'm the one who has to debug and patch it (since I built it myself, I own the responsibility). The products I built are actively generating revenue, so being able to still find the time to write content, teach classes, and support students already feels like a major personal achievement. I simply had no bandwidth left to think about showing off (maybe later down the road, as things settle down :D).

In Summary

I ended up learning and vibe coding the entire software lifecycle from A to Z within a single year, across virtually every operational stage. If we measure how long it would have taken a single individual to acquire this breadth of practical experience before AI existed, I honestly don't think saying "10 years" is an exaggeration (superhumans aside, of course).

Here are a few personal takeaways I've gathered along the way (this reflects my personal journey and might not apply to everyone, but hopefully it offers a useful perspective):

Build first, study theory later: It's similar to learning guitar—most people learn chord shapes and strumming patterns before diving into music theory, right? Some never even study theory at all. Building first gives you immediate, tangible lessons when things fail. And even when you fail, producing something tangible—even if slightly flawed—delivers enough dopamine to keep you motivated to tweak it or build the next iteration: feeling that you're almost there, that with just one more fix it will work...
Use AI to learn, not just to delegate work: AI is a tool, and we have to admit it possesses a wider knowledge base than most of us, synthesized from decades of collective human experience, and it computes at remarkable speed. We should take full advantage of this vast knowledge base to learn. Some people think: "If AI already knows so much, why not just hand everything over to it and skip learning?" But at its core, AI is still a machine. It does whatever you instruct it to do; it rarely pauses to verify whether what it generated is truly right or wrong, whether it aligns with your underlying intent, or whether it misunderstood key requirements... and it relies heavily on unstated assumptions (see this post). If you leave it entirely unattended, it will almost certainly make mistakes—and besides, what would your value be then?
Only delegate to AI what you already understand: This is a natural follow-up to the point above. If you don't understand the underlying principles, you cannot critically challenge AI's output. Even if you don't write every line of code by hand (like me), you must at least be able to evaluate business logic, function breakdown, system architecture, and data flow.
No pain, no gain: Failure is the prerequisite to mastery. I won't claim when "success" arrives—I don't consider myself a success story. But one thing I know for certain: failing is the fastest way to learn. It frustrates you, and your brain is wired to imprint that friction deeply. When you finally resolve it, the satisfaction is far greater than if everything had worked seamlessly on the first attempt (after all, if nothing fails, what did you actually learn? :D).

Returning to the title question from the beginning: Do you really learn 10x faster with AI? After a year of intense, hands-on grinding, my answer is: Both yes and no. AI doesn't make us smarter simply by virtue of using it; it functions as a magnifying glass that amplifies your own capacity for self-learning.

If you are intellectually passive, AI will only help you churn out flawed software 10x faster (garbage in, garbage out). But if you adopt a sound learning method, embrace friction, and stay persistent, you will acquire a tremendous amount of knowledge. This past year has been an extraordinary journey for me—AI moved at breakneck speed, and I was running at full tilt just to keep up. Looking back now, I never imagined I could have learned this much in 12 months.

I hope sharing this experience gives you a bit of encouragement on your own path of personal growth. Investing in yourself is an investment that never loses value.