Why I Built Manabit.ai Instead of Just Using Anki and Graded Readers

I studied Japanese the "normal" self-study way for a long time before I started building manabit.ai. Anki deck in the morning, a graded reader in the evening, WaniKani in between, occasional guilt about not doing enough output. It worked, in the sense that my numbers went up. But at some point I noticed the gap between "I can pass a reading section" and "I can actually sit down and read something I picked because I wanted to, not because it was leveled for me" wasn't closing. If anything it felt like it was getting more obvious.

This post is basically the origin story of the app — why I started it, what specifically felt broken, and what manabit.ai is (and isn't) trying to fix. I'm the person building it, so take the "graded readers aren't enough" argument with the grain of salt it deserves: I have skin in the game here.

What graded readers actually do well

I don't want to trash graded readers, because they solve a real problem. Going from zero to being able to parse simple sentences without a dictionary in one hand is genuinely hard, and graded readers (Tadoku-style, NHK Easy, the various leveled series) are a solid, proven way through that stretch. Comprehensible input works. Nobody serious disputes that at this point.

The problem is what happens after that stretch. Graded readers are, by design, graded — someone controlled the vocabulary and grammar so you'd stay in your comprehension zone. That's the whole mechanism. But it means the material is written for the learner, not written for a Japanese reader. At some point you want to read the things Japanese people actually read: a blog post, a light novel, a manga without furrowing your brow at every panel, news that wasn't simplified for you. That's a different kind of text, and graded readers don't get you there by themselves — they get you to the point where that jump is survivable, not necessarily comfortable.

The other issue is more mundane: once you're past the beginner graded-reader levels, the pipeline mostly dries up. There's less leveled material at intermediate-plus, so you're thrown into native content earlier than the "graded" method implies, usually with a dictionary app in one hand and a lot of friction.

Where Anki fits, and where it doesn't

Anki (or any SRS) is great at one narrow thing: making sure a word or pattern you've seen doesn't quietly evaporate from memory. That's valuable and I still use spaced repetition. But an SRS deck, on its own, has no relationship to what you're actually trying to read. You can have a deck of 3,000 mature cards and still get stopped cold by a sentence in a novel, because vocabulary in isolation and vocabulary in context are not the same skill. Recognizing 語彙 on a flashcard and recognizing it mid-sentence, doing a specific grammatical job, with three other unfamiliar words around it, are different tasks that happen to share a word.

So what I kept running into was a structural gap: Anki keeps you from forgetting things, graded readers get you comprehension habits early on, but neither one is actually about reading real things and getting unstuck when you hit a wall in real text. There wasn't a tool built around that specific moment — you're reading something you chose, you hit a wall, now what.

What manabit.ai is trying to be

The idea behind manabit.ai is narrow, on purpose: let people read real Japanese — the stuff they'd want to read anyway, not material written down to a level — and lower the friction of getting unstuck without yanking them out of the reading experience every time. Look something up, get grammar context when a sentence structure is doing something unfamiliar, and keep going, instead of the read-stop-dictionary-app-context-switch-forget-what-you-were-reading loop that kills a lot of independent reading sessions before they start.

I want to be honest about where the product actually is right now, because "building in public" is worth nothing if I only post the highlight reel. It's early. The core reading-and-lookup loop works and I use it daily, but there are rough edges — coverage gaps on certain text types, UI decisions I'm still not sure about, features people have asked for that aren't built yet. If you're looking for a polished, feature-complete platform, that's not an honest description of what's here today. If you're curious about an approach to extensive reading built by someone using it to study for their own JLPT prep, that's what this is.

I'm not trying to replace Anki or graded readers — I still think they're the right tool for the first stretch of learning. manabit.ai is aimed at the part after that: when you're ready to read real things and want the tool to get out of your way instead of setting your level for you.

If any of this resonates — the specific frustration of being "graded-reader ready" but not "real-book ready" — I'd genuinely like to hear about it. This project moves based on that kind of feedback more than on a roadmap I wrote in isolation.