Since I Started Letting AI Code ④ — Chasing AI, Learning Alongside AI

AI Adoption Journey (a 4-part series)

Trying Every Tool Until I Built My Own

The System AI 'Works' In — Engineering and Verification Ownership

When the AI Server Goes Down, I Turn Into an Idiot Too — Local LLMs and AI Dependence

Chasing AI, Alongside AI — Even How I Learn Has Changed ← you are here

I've pulled this series all the way to this point. I hopped between tools (Part 1), hardened how I work into an actual system (Part 2), and confronted AI dependence (Part 3). This final post is about how that change eventually reached past work itself, into the way I learn. But first, there's a confession I've been putting off.

Paradigms Get Torn Up Every Few Days

In Part 1, I said there's no finish line for tool use — but that's not just about tools. The paradigm of how you work changes on a cycle of days to, at most, a few weeks. Roughly, the progression looks like this.

Prompt engineering → context engineering → harness engineering → loop engineering → Goal prompting.

At first, it was all about "how do I phrase the question." The center of gravity soon shifted to "what do I feed it as context," then to "what execution structure do I put on top of the model" (that's the harness that flopped so badly in Part 1), then to "how do I get this running automatically," and I got swept along, dizzy, all the way to "can I just throw a goal at it and let it figure out how to get there." I don't know what's next. But one thing is clear — staying as close as possible to this shift, without falling behind, is the whole game. There's no tidy right answer. All you can do is grit your teeth, keep chasing so you don't fall behind, and grow your own ability in the process.

A Confession — AI Wrote This Blog, Even This Post

The story in Part 2 about boundaries disappearing is something I'm now living directly. I'm a backend developer, and these days I get AI to churn out frontend code for me, even some of the design. I'm curious myself how far this boundary will keep dissolving.

The proof is right in front of you. AI took the lead in building this blog itself. I only handled the overall direction and orchestration. To be even more honest, AI is the one who organized and wrote the very post you're reading right now. All I do is review it.

But "review" turns out to be a far more active job than it sounds. I read through AI's draft, fill in whatever it left out, and catch the places where the facts drifted. Where my own memory is fuzzy, I have AI dig through old work logs to find the evidence — this is exactly where the records I've been piling up in Obsidian since Part 2 pay off. When what it finds contradicts my memory, I correct my memory; when AI has plausibly written down something wrong, I fix that instead. I don't fill gaps with guesses. I leave a spot blank until it's confirmed, and only write it into the post once I've talked it through with AI and actually understood the reasoning. (Farming research out to a subagent, the way I described in Part 2, is part of this same division of labor.)

So this is nowhere near a "hand it off and you're done" kind of job. Working through notes like "flesh this part out more" or "go find evidence for this" one by one, polishing a single post still eats a fair chunk of time. No post comes out finished in one pass — I take the draft, fix it, read it again, fix it again. In development terms, it's not a single clean build, it's dozens of iterations; in writing terms, it's endless revision. That process of working a rough draft into shape is, in the end, what writing actually is. Still, reading line by line like that, catching every note, I end up picking up details I hadn't known and reviewing my own past work and knowledge all over again. So the time I spend reviewing isn't pure cost. It's a strange feeling, having AI write the whole post for me — but honestly, I'm pretty satisfied with where things stand right now.

The reason I haven't given up on writing, even with review eating this much time, comes down to speed. I used to sink a whole day or several straight days into a single blog post. Just finding the right information and researching the facts before I even started writing took a serious chunk of time on its own, and I'd often burn out partway through and abandon the draft. Now that research time has dropped dramatically. With AI pulling in sources and pointing me to evidence, the whole pace of writing sped up — especially for technical posts. So I've decided to pour everything I know into this blog going forward. This series is the first step.

Even the Way I Read Books Has Changed

The change didn't stop at code and writing. At some point, I started reading books together with AI too. And that changed how I read, period. I started analyzing every sentence and every scene, the way you'd pick apart a passage in a language-arts class back in middle or high school. Sentences I used to just skim past, I now sit with — going back and forth with AI on "why would they write it this way" and turning it over in my head.

Poetry was even more dramatic. Honestly, I used to be someone who genuinely couldn't understand why people read poetry at all. But reading and discussing poems with AI, I finally got why people read poetry — and I started reading it myself. Naturally, I do this the same way, like studying for a language-arts class. I take a line, work out my own reading of it, and then ask AI, "am I reading this right?" When AI says yes, I don't stop there — I push back with "then critique the holes in my interpretation." Go back and forth like that over a single stanza for a while, and an hour slips by on one poem. Watching myself sink that much time into a single poem genuinely surprised me. I'm planning to unpack where this change has taken me, separately, sometime in my hobby notes.

AI turned out to be more than just a tool for efficiency — it finally opened a door I'd kept shut my entire life. That's a change no productivity metric will ever capture.

So, Back to Where the Fear Started

This series started from fear. Part 1's fear of replacement — "will my job disappear" — and Part 3's fear of dependence — "am I an idiot without AI." Neither one has fully gone away, even now. But the place I've reached, after passing through that fear, is clearly not the place I started from.

The fear of replacement turned into a different kind of exhaustion — the sheer amount of work I can now take on, thanks to AI, exploded instead. The fear of dependence is being filled in, bit by bit, by a system built out of verification, records, and understanding the underlying principles. Somewhere in between, I stopped writing code by hand, took ownership of verification, built a skill of my own, got my hands dirty down at the model's foundation, and started reading poetry.

So if I had to compress everything that's happened into a single sentence, right now, it would be this.

In the age of AI, I chase AI, and I work at growing my own ability alongside it.

That's the entire realization I've come to, working with AI up to this point. I still don't know the right answer. The paradigm will shift again within a few days, and I'll hop to another tool and flop all over again. That's fine, though. Because while I'm chasing it, I'm clearly growing. As proof of that growth, I'm leaving this post behind — written together with AI.

The AI Adoption Journey series ends here. If you'd like to start over from the beginning, head back to ① Tool Evolution.