Just over a year ago I started my vibe-coding journey.

I used Replit a little bit before it went full AI, so I stuck to this platform and, to date, haven’t tried anything else. Some will say that I need to experiment with more platforms to find out what I like, but I honestly liked Replit from the start, and it’s wonderful to watch them improve month by month. I have witnessed their evolution from AI as a small pre-Agent helper tool to its current Agent 3 capabilities.

Before AI, I learned some coding in Python and Ruby as a hobby but I never turned the skill into a job. Naturally, an inexperienced programmer is a shitty programmer even if he knows the theory. The reason why I wanted to learn coding was to at least understand the basics of how software worked under the hood. I’m more business-inclined but I really hate it when people bullshit me assuming I don’t know anything from their field. (Yeah, car mechanics can still bullshit me freely at this time.) So I simply wanted to elevate myself somewhat so that I’m not treated like an idiot and taken advantage of.

But AI has changed the game completely. Now, instead of spending countless days writing code by hand and even more time googling how and what to write, I delegate this tedious manual task to Replit’s Agent 3. And over the past year it got better and better at it before my eyes. Stupid mistakes, silly bugs, hallucinations that were prevalent just a year ago are now practically gone. A more complex task can be broken down into smaller, digestible parts and taken care of by sub-agents and then, upon completion, reviewed by the Architect.

I don’t mean to say that it does a perfect job all the time. AI, even though it aspires to be one, is not the brain of my operation. I direct it all the time, step by step, task by task. It’s more akin to a manual laborer who has access to all the tools in the world and can instantly learn how to use them. You tell him what to do, he chooses the right tool to do it and does his best to produce a result. Like a regular worker, he can misunderstand and make mistakes. But nothing is unfixable. A year ago, I could be stuck for a few days, very frustrated at the agent, because he couldn’t solve a problem that seemed easy for me in theory, but was difficult for him in practice. Today, this almost never happens. At most, I may spend a couple of hours battling a specific issue, but surely not days and definitely with much less frustration.

As far back as I remember myself, I’ve always been lazy to learn theory and preferred to learn by doing. It has got me in trouble a few times (“Yuri, the instructions to this Ikea drawer are in the box, and now you broke it!”), but most of the time action taught me well and added color to my life. I’ve tried and done so many things that I don’t need to imagine what they are like: I have experience! And so in this case, I’m not an academically trained computer scientist, but through sheer action (and thousands of dollars in AI credits) I have learned how to write full-stack platforms. They may not be ideal and orthodox and are likely to have some slop that shouldn’t be there, but the mere fact that I can now do this without asking for anybody’s permission or paying tens of thousands of dollars for shitty prototypes is massively liberating to me. To an introvert like myself, it’s akin to acquiring a superpower.

There is not stopping this train, it will only accelerate. Get on board or you’ll be left behind. Embrace the change, live in it, be it.