New on KB Cafe: Practice in Every Lesson, Capstones, and a Path Builder

KB Cafe has always said “learn by doing.” For a while, the honest version of that was: read a clear explainer, then go play with a tool. Good, but the reading and the doing lived in different places. This update closes that gap. Now the doing happens inside the lesson, the moment you learn a thing you get to use it, and the whole system remembers your progress. Plus two substantial new guides and a pile of polish. As always: free, no account, nothing leaves your browser.

Practice, right in the lesson

The headline is a new practice engine. Instead of a single quiz bolted onto the end of a page, lessons now interleave short exercises between the ideas: read a concept, do it, read the next, do that. There are three kinds of exercise, and the best way to explain them is to just try them. This is live, it grades right here:

That was a multiple-choice check, a fill-in-the-command with real answer matching (it accepts git add ., git add -A, and the regex in between), and a code exercise that actually ran: your function executed against hidden tests, in your browser, with no server in the loop. Every exercise also has a “Check it with your AI” button that copies the task, your answer, and a tutoring prompt to your clipboard, so you can paste it into Claude or ChatGPT and get a hint. That is our honest take on the “AI tutor” that paid platforms charge a subscription for: use the one you already have.

It is already interleaved into the Git lesson (run the everyday loop, branch and merge, dodge the .gitignore traps), shell scripting (quoting, permissions, redirection), and tool calling, with more lessons landing the same treatment. Your completions are saved on your device, so a page you half-finished remembers where you were.

A streak, XP, and a badge that means something

Doing things now counts. Every exercise you pass, every capstone you finish, every step you check off feeds a running XP total and a day-streak, shown as a small badge in the header once you have earned anything. It is deliberately low-key: no leaderboards, no notifications, no dark patterns. Just a quiet nudge that you have been showing up. And because it is all in local storage, the streak is yours and private, it never touches a server.

Here is the exact badge, from your first exercise to a month of showing up:

Day one, a week in, a month deep. It lives in the top-right of every page, right next to Learn, and quietly climbs.

Capstones: build the thing

Reading a path and building something at the end of it are different skills. Every learning path now ends with a capstone: a real, small project with a checklist that tracks your progress. Not a toy, a thing you actually make:

• Master Prompting → write and optimize a real prompt for a task you need done this week.
• AI Infrastructure → trace one of your own questions through tokens, context, embeddings, and retrieval.
• Build AI Agents → spec a real agent end to end: its job, its tools, and the guardrails you will not skip.
• Automate Work → design one automation that runs itself, deciding where a workflow ends and an agent begins.
• Coding with AI → ship one real change with AI, plan first, read the diff, commit your safety net.

Each links the exact tools for the job, and checking off the last box is worth a satisfying chunk of XP.

The Learning Path Builder

Not sure where to start? The new Learning Path Builder takes a goal in plain language, “build an AI agent that can search the web,” “get better at prompting,” “monetize a channel with AI,” and assembles an ordered route for you: foundations first, then the core, then where to go deeper, with the hands-on tools marked.

Here is the important part, and where it differs from an AI “course generator”: every step is a real page a human wrote and checked. Nothing is generated on the fly, so you never get a confident-sounding lesson that is subtly wrong. It composes a path out of KB Cafe’s existing, vetted guides and tools, following the same prerequisite graph that powers the knowledge map. It runs entirely in your browser, and it is now featured on the Learn hub too.

Two new guides

Alongside the system, two meaty new pages went up, both built the KB Cafe way: mechanism first, real numbers, no guru nonsense.

How to prompt Claude Fable 5

How to prompt Claude Fable 5 covers what actually changes when you move to a model built for long, autonomous work: effort as the main dial, why you can steer it with fewer instructions rather than more, keeping long runs honest, and the migration gotcha that quietly falls back to Opus if you tell it to show its reasoning. It is grounded in Anthropic’s official guidance, with a hands-on match-the-fix exercise, and an honest note on the June-to-July re-release.

Monetize a YouTube channel with AI

Monetize a YouTube channel with AI is the anti-hype version of every “get monetized in 90 days” post. It uses AI as your strategist to plan, package, script, and diversify a channel, but it is honest about what that requires: the real 2026 Partner Program thresholds (with an interactive eligibility checker that tells you exactly what you still need), how the money actually splits, and the 2025 policy that quietly demonetizes mass-produced AI “slop.” Every number is from YouTube’s own docs; the eight planning prompts are rebuilt with the income guarantees stripped out.

Smaller things, worth mentioning

• One clean menu on mobile. The header used to show two identical hamburger icons on phones. It is now a single menu button, in the same place on every page, opening one tidy drawer. No more guessing which icon does what.
• The knowledge map is tappable again. On the constellation view, tapping a star now reliably opens its detail card, a subtle touch-input bug was swallowing taps.
• A live status widget in the wild. Our post on what happens when Anthropic goes down now embeds a live AI-status widget, so you can see the exact thing the article describes, updating in real time, without leaving the page.

The through-line

We spent a while studying the best paid, practice-first learning platforms, the ones that get people to actually do the work. They are good. They are also behind a login and a monthly bill. The question we kept asking was: why does any of this need an account? Grading a code exercise, remembering a streak, assembling a path, none of it requires a server or your email. So we built it the way we build everything here: it runs in your browser, it stores your progress on your device, and it never asks who you are.

Where to start

Open the Git lesson and actually run the exercises, then check off a capstone on the prompting path. If you are not sure what to learn, tell the Path Builder your goal and follow the route it gives you. Watch the streak in the header tick up. Same promise as day one: you keep this stuff by using it, so we keep building things to use.