Frequently Asked Questions About Jason Ku AI-Native Learn-to-Code Framework
21 answers covering everything from basics to advanced usage.
// Basics
What does 'learning happens in the friction' actually mean?
It means real understanding comes from deliberate slowdown, not speed. Before running any AI-generated code you predict what it will do, then you read every line and answer quiz questions on new concepts. That effortful friction — resisting the urge to just accept and run — is where learning forms. Speed is the enemy at the beginner stage.
What is a coding agent and how is it different from a vibe-coding tool?
A coding agent (Claude Code, Codex, Cursor) is an AI tool that writes, runs, and iterates on code interactively while exposing the underlying code and concepts. Vibe-coding tools (Lovable, Base44, Replit) hide complexity to maximize ease of use. The framework uses agents specifically because they show you the code you must understand, and their skills transfer to real jobs.
What are the trunk, branches, and leaves in the Knowledge Tree?
The trunk is the core system components almost every project shares: Git, backend, database, frontend, API, and deployment. Branches are the high-level feature slices that build sequentially (static page → interactivity → persistence). Leaves are the individual tasks and the underlying concepts they introduce, like variables, arrays, or API endpoints. Leaves only matter when attached to branches and a trunk.
What are typical branches for a first web app project?
A common sequence is: (1) serve a plain static HTML page, (2) add interactivity with JavaScript or React, (3) connect to a backend API, (4) persist data to a database, and (5) deploy to a free host. Each branch is a shippable increment that builds on the last, so you always have something that works while learning progressively harder concepts.
// How To
How do I pick a good first project?
Have the agent interview you about your hobbies, routines, work, and recurring frustrations, then pitch ideas calibrated to your level. A good first project solves a real problem you personally have (so motivation is intrinsic) and is complex enough to teach real engineering but small enough to ship. Avoid two-sided marketplaces, payment systems, and multi-role platforms.
How do I set up my coding agent without over-investing time?
Spend just one evening: install your chosen agent and watch one or two tutorials covering context management, memory, model selection, and plan mode. Don't try to learn everything — the deeper agent skills like token budgeting are learned naturally on the $20 plan as you build. Over-preparing on the tool before starting a project is a pitfall.
How do I initialize the Knowledge Graph?
As the agent defines each task (leaf), it must identify every underlying concept the task introduces. For each concept, create an entry with the concept name, current mastery status starting at 'Introduced,' and a list of evidence from completed tasks. The agent then consults and updates this graph before and after each task to calibrate explanation depth and quiz intensity.
How do I run a single task session with the agent?
For each leaf task: (a) the agent explains what the task will do, (b) it generates code, (c) before running, you answer 'in your own words, what do you expect to happen?', (d) run and review the result together, (e) the agent quizzes you on every new concept, then updates the Knowledge Graph with mastery evidence. Never skip the prediction or quiz steps.
// Troubleshooting
What's the number one mistake beginners make with AI coding?
Letting the AI generate and run code without first predicting the output or understanding what it does. This produces fast-moving, zero-understanding codebases that spiral out of control. The precise habit to avoid is handing the agent an error and saying 'just fix it.' Always require explanation and prediction before running any generated code.
What should I do when I get tired and want to skip the quizzes?
Treat fatigue as a signal to stop the session, not to lower the bar. Skipping quizzes to move faster is exactly how you end up with a codebase you can't explain. End the session, rest, and resume when you can engage with the prediction-and-quiz loop properly. Never keep going while cutting the friction that produces learning.
My project keeps growing and I can't ship it — what went wrong?
You likely picked a first project that's too complex or never scoped a true MVP. Strip back to the smallest end-to-end working version — the fewest features to close the full loop — and push everything else to a backlog. If the whole concept is a two-sided marketplace or payment platform, it's too big for a first project; choose something simpler.
The agent keeps over-explaining things I already know. How do I fix it?
This usually means the Knowledge Graph isn't being maintained. Ensure the agent consults the graph before each task and updates mastery status after each one. Once a concept moves from Introduced to Practicing or Understood, the agent should skip basic explanations and jump to harder quizzes. Skipping graph initialization in step 6 causes both over-explaining and gap-skipping.
// Comparisons
How does this compare to just using ChatGPT to write my code?
Using ChatGPT to generate code you copy without understanding produces the exact anti-pattern this framework prevents. Here, a professional agent works inside a structured Knowledge Tree and Knowledge Graph, forcing prediction and quizzing at every step. The goal isn't finished code — it's a deployed app you can explain line-by-line, plus transferable coding and AI-collaboration skills.
How does this compare to a coding bootcamp?
Bootcamps deliver a fixed curriculum on a rigid schedule, often with generic projects. This framework is self-paced, anchored to a project you personally care about, and calibrates explanation depth to your existing knowledge via the Knowledge Graph. It also teaches AI collaboration from day one — a skill most bootcamps still don't structure around — while producing a real deployed portfolio piece.
Why not use Replit or Lovable since they're more beginner-friendly?
Because vibe-coding tools abstract away the complexity you actually need to understand, and the AI skills they build don't transfer to a professional engineering context. Being beginner-friendly by hiding code is the opposite of what produces durable understanding. The framework insists on professional-grade tooling from day one so your skills are transferable to a real job.
How is the Knowledge Graph different from just taking notes?
Notes are static and passive; the Knowledge Graph is a living, agent-maintained record with mastery statuses and evidence logs that actively drives the agent's behavior. Before each task the agent reads it to decide whether to explain or quiz. It turns your progress into an operational input, not a document you rarely revisit.
// Advanced
How many weekly hours do I need to commit?
There's no fixed minimum — available weekly hours are an optional input used to calibrate timeline expectations, not gate entry. Because the project is scoped to an MVP, even a few focused hours per week can ship something working within weeks. What matters more is protecting the friction: never trading understanding for speed to hit an hours target.
What is the 'coding harness' Jason Ku refers to?
The coding harness is the broader system wrapping the coding agent — the interface, step tracker, Knowledge Graph visualization, and session management layer that keeps the learner oriented throughout the project. It's the scaffolding around the agent that maintains structure so you always know where you are in the Knowledge Tree.
How does mastery progression work across a project?
Each concept moves through four statuses: Introduced → Practicing → Understood → Mastered, tracked with evidence from completed tasks. A concept like 'variable' might start Introduced in branch 1, become Practicing by branch 3 — at which point the agent skips basic explanation and delivers a harder quiz. Progression is evidence-based, driven by demonstrated understanding across real tasks.
Can I use this framework to restart after quitting coding before?
Yes — it's explicitly designed for learners who stalled on previous attempts. Prior quitting usually stemmed from collecting isolated tutorials (dangling leaves) with no project to anchor them and no early shipping win. Starting with a project you care about and an MVP you ship in weeks closes the motivational feedback loop that tutorial-hopping never provided.
Why develop coding and AI collaboration skills at the same time?
Because in 2026 both are high-value and complementary, and deferring either wastes the opportunity. Working with an agent from day one means you learn coding fundamentals and how to direct, review, and quiz an AI in parallel. Neither is treated as a prerequisite for the other — the framework builds both simultaneously through the same project.