12:37Nate Herk | AI Automation · Review
Fable 5.1 vs Fable 5: Same Prompt, Same App, $700 Apart
One orchestrator prompt, two agentic coding models, and a blind-judged scorecard that says the cheaper build won anyway.
September 3rdThe argument in one line.
Learning to code in 2026 is still worth it, but only if you anchor every concept to one real project through a 'knowledge tree' and force yourself to predict what AI-written code will do before running it, rather than letting the agent finish tasks unsupervised.
Read if. Skip if.
READ IF YOU ARE…
- A total beginner starting from zero who isn't sure whether AI has made learning to code pointless.
- Someone who has bounced between disconnected tutorials and never shipped a finished project.
- A beginner who already pays for Claude Code, Cursor, or Codex but has no structured curriculum for learning with it.
- Someone who wants a concrete first-project idea, an MVP scope, and a realistic week-by-week time estimate.
SKIP IF…
- You're already a working engineer past the fundamentals stage — this is a first-90-days beginner curriculum, not an advanced workflow.
- You want an in-depth tool comparison — the video picks a three-tool short list in about a minute and moves straight to the learning method.
The full version, fast.
The video argues coding is still worth learning in 2026, but the method has to change: pick a paid coding agent — Claude Code, Cursor, or Codex — and skip consumer vibe-coding tools like Lovable, Base44, and Replit, which hide the complexity you're there to learn. Then anchor every concept to one real project using a 'knowledge tree': trunk (core pieces), branches (features), leaves (individual tasks). Define an MVP first so you ship something working fast, then move leaf by leaf while an AI-maintained knowledge graph tracks which concepts you've mastered and quizzes you before running code. The failure mode to avoid is letting AI 'just fix it' without understanding why. Realistic timeline: an evening or two to set up, then 1–3 months at roughly 10 hours a week to finish a real project.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:49
01 · Am I wasting my time learning to code?
Opens with the beginner's real question — is coding still worth learning with AI writing more code every year — and answers it directly: yes, from a 15-year Meta engineer who ships AI-assisted code daily.
00:49 – 02:08
02 · The plan for this video
Previews the seven-step plan, the tools, the knowledge tree method, and the one AI mistake beginners keep making.
02:08 – 02:55
03 · Step 1: Pick a coding agent (Claude Code vs Cursor vs Codex)
Narrows the choice to three paid coding agents used by working engineers, with Claude Code as the personal pick for its interactivity.
02:55 – 04:05
04 · The vibe coding tools I'd avoid
Names Lovable, Base44, and Replit as tools to skip for learning because they abstract away the complexity a beginner needs to see.
04:05 – 04:56
05 · What to learn first: context, memory, plan mode
Recommends one evening learning agent fundamentals — context, memory, model selection, plan mode — before touching a project.
04:56 – 08:24
06 · Step 2: Pick a project + the knowledge tree method
Introduces the knowledge tree metaphor and the rules for picking a first project: solve a real problem, size it to your level, avoid over-scoping.
08:24 – 09:17
07 · The prompts and skills pack
Shows the copy-paste prompt used to have AI interview you into a right-sized project idea, part of a free companion skills pack.
09:17 – 12:01
08 · Step 3: Define your MVP
Defines MVP as the smallest end-to-end shippable slice, worked through with a fitness-tracker example reduced to log-workout plus history.
12:01 – 13:53
09 · Step 4: Build the trunk
Defines the trunk as the shared core components of any project — source control, backend, database, frontend, API, deployment.
13:53 – 14:54
10 · Step 5: Build the branches
Defines branches as the ordered horizontal feature slices a project ships in, each building on the last.
14:54 – 15:40
11 · Step 6: Build the leaves
Defines leaves as the individual concrete tasks inside a branch, where concept-level learning actually happens.
15:40 – 17:13
12 · The knowledge graph explained
Explains the knowledge graph that tracks concept mastery status over time and adjusts how much re-teaching a concept needs.
17:13 – 19:11
13 · Step 7: One task at a time (where the learning happens)
Introduces the core forcing function — predicting code behavior before running it — and names the 'just fix it' failure mode to avoid.
19:11 – 21:32
14 · Example: learning variables with AI quizzes
Walks through a real transcript of the method in action, teaching the concept of a variable via a hard-coded list turned into an array.
21:32 – 22:04
15 · Recap of all 7 steps
Recaps the full seven-step plan in sequence.
22:04 – 23:33
16 · How long this actually takes
Gives a realistic timeline: an evening or two for setup, then 1–3 months at ~10 hours a week for step seven.
23:33 – 25:18
17 · The AI coding harness I'm building
Previews a companion product/waitlist — a UI that visualizes the knowledge graph and tracks progress through the plan.
25:18 – 26:15
18 · Join the free Workshop community
Closes with a free-community CTA for beginners going through the same process.
Lines worth screenshotting.
- The three coding agents working engineers actually use — Claude Code, Cursor, and Codex — all cost around $20/month, and that base subscription is treated as non-optional.
- Vibe-coding tools like Lovable, Base44, and Replit are worth avoiding specifically for learning, because they abstract away the exact complexity a beginner needs exposure to.
- A knowledge tree treats one real project as the trunk, its features as branches, and individual concepts as leaves, so nothing you learn is a disconnected floating fact.
- An MVP is the smallest end-to-end slice of a project that actually ships — for a fitness tracker example, that's logging a workout and viewing history, nothing else.
- A knowledge graph tracks each concept's mastery status — introduced, practicing, understood, mastered — so an AI tutor knows how much to re-explain a concept the next time it resurfaces.
- The single biggest mistake beginners make with AI is saying 'I don't care what you do, just fix the problem,' which produces working code you can't explain and a codebase that snowballs out of your understanding.
- Predicting what a piece of AI-generated code will do before running it, not after reading the output, is where the actual learning happens.
- A beginner following this plan can expect an evening or two to set up tools and scope a project, then roughly 1–3 months at 10 hours a week to ship it.
- Even a 15-year Meta engineer admits he still makes the 'just fix it, I don't care how' mistake with AI when he's tired — the discipline doesn't get automatic with experience, it stays a choice.
- Claude Code is favored for its interactivity and feedback loop, while Codex is positioned as stronger at autonomous planning over longer stretches — presented as a tradeoff, not a strict ranking.
Anchor every concept to one real project, not a pile of tutorials.
LEARNING METHOD
Pick a coding agent, scope a real project's MVP, then learn every concept as a leaf attached to that project's trunk and branches while a forcing-function quiz habit keeps you from letting AI finish tasks you don't understand.
01Am I wasting my time learning to code?
- Coding remains worth learning in 2026 even as AI writes more of the code, according to someone shipping code daily at a large tech company.
- The skill that matters now is working with an AI coding agent effectively, not just knowing syntax by hand.
03Step 1: Pick a coding agent
- Claude Code, Cursor, and Codex are the three coding agents actually used by working engineers, each costing roughly $20/month.
- Claude Code is chosen for tighter feedback and interactivity, which suits a beginner still learning to read and question AI output.
- Codex trades that interactivity for stronger autonomous planning and longer unsupervised task completion — useful once you don't need to check every step.
04The vibe coding tools I'd avoid
- Consumer vibe-coding tools like Lovable, Base44, and Replit are worth avoiding for learning specifically because they hide the complexity a beginner needs exposure to.
- The test for picking a learning tool: use what working engineers actually use on the job, so the skills transfer directly.
05What to learn first: context, memory, plan mode
- Before writing any code, spend an evening on agent fundamentals — context management, memory, model selection, plan mode — rather than trying to absorb everything at once.
- One evening of setup plus a tutorial or two on the coding agent itself is enough to get moving; the rest is learned by doing.
06Step 2: Pick a project + the knowledge tree method
- A single real project should act as the trunk that every new concept attaches to, rather than learning languages or frameworks as isolated topics.
- Pick a project that solves a problem you actually have — personal stakes create the motivation to keep coming back to it.
- If you can't think of a project, have the AI interview you about your routines, hobbies, and work to pitch ideas sized to your actual experience level.
- Avoid over-scoping a first project — a two-sided marketplace with payments and multiple user roles is too complex for a first build.
07The prompts and skills pack
- A reusable prompt for finding a project: tell the agent your current experience level and ask it to interview you about your life so it can pitch an idea that's challenging but not overwhelming.
08Step 3: Define your MVP
- An MVP is the smallest end-to-end version of the project that ships and works, not a stripped feature list — for a fitness tracker, that's log-a-workout plus workout history, full stop.
- Shipping an MVP first creates the feedback loop of seeing an idea become a real, usable thing, which is what keeps beginners motivated past the early plateau.
- Every other feature you can imagine goes into a backlog to layer on after the MVP ships, not before.
09Step 4: Build the trunk
- The trunk is the small set of core pieces every project shares — source control, backend, database, frontend, API, deployment — defined before any feature work starts.
- Naming the trunk up front gives a beginner a system-level mental model from day one, instead of only seeing isolated pieces of code.
10Step 5: Build the branches
- Branches are the horizontal slices you ship in order — for example a static HTML page, then interactivity with JavaScript/React, then memory and a database — each building on the last.
11Step 6: Build the leaves
- Leaves are the individual tasks inside a branch — install Git, install Node.js, and so on — which is where concept-level learning actually happens.
12The knowledge graph explained
- A knowledge graph is a running record of every concept touched, each tagged with a mastery status — introduced, practicing, understood, mastered — plus evidence of when it was last reviewed.
- The mastery status determines how much re-teaching a concept gets the next time it resurfaces in a task, so review time isn't wasted re-explaining something already known.
13Step 7: One task at a time
- Step seven — working one task at a time with the agent — is where most of the actual learning time gets spent, and it's deliberately slower than letting AI batch-complete work.
- The core forcing function: before running any AI-generated code, explain in your own words what you expect it to do — the friction of prediction is where retention comes from.
- The failure mode to avoid is telling the agent 'I don't care what you do, just fix the problem' — it ships working code you can't explain and the codebase snowballs past your understanding.
14Example: learning variables with AI quizzes
- In a worked example, turning a hard-coded list of items into a 'books' array and looping to render it teaches the concept of a variable inside a real task, not a standalone lesson.
- After a quiz on a concept, the agent updates its status in the knowledge graph — from 'introduced' to 'practicing' — so repetition compounds instead of resetting each session.
16How long this actually takes
- Steps two through six — picking a project, MVP, trunk, branches, and leaves — take about one evening or a few hours total.
- Step seven, actually building leaf by leaf, takes roughly 1–3 months at about 10 hours a week depending on project complexity.
- The payoff at the end is being able to explain every line of code in a finished portfolio project, not just having shipped it.
Terms worth knowing.
- MVP (minimum viable product)
- The smallest version of a project that ships end-to-end and actually works, used as the first target instead of a long feature list.
- Knowledge tree
- A learning method that anchors every new concept to one real project, structured as a trunk (core pieces), branches (features), and leaves (individual tasks).
- Knowledge graph
- A running record of every concept a learner has touched, each tagged with a mastery status and evidence of when it was last reviewed.
- Coding agent
- An AI tool like Claude Code, Cursor, or Codex that writes and edits code inside a project on a developer's behalf, as opposed to a simple chat-only assistant.
- Vibe coding
- Building software mostly by describing what you want to an AI tool and accepting its output, often through consumer app builders that hide the underlying code.
- Plan mode
- A mode in a coding agent where it drafts a step-by-step approach for review before making any changes to the codebase.
- Trunk / branches / leaves
- The three layers of a knowledge tree: the trunk is a project's core components, branches are its major features shipped in order, and leaves are the individual tasks inside each branch.
Things they pointed at.
02:35toolClaude Code
02:37toolCursor
02:37toolCodex
02:41toolLovable
02:41toolBase44
02:42toolReplit
Lines you could clip.
00:29
“Yes, I think coding is definitely still worth it in 2026. I don't see software engineers going anywhere in the near future.”
02:42
“I would avoid some of the more popular vibe-coding tools like Lovable, Base44, Replit — because they abstract away some of the complexity that makes it very user-friendly.”
09:40
“One of the best things about being a software engineer is being able to see a project come to life.”
17:58
“I don't care what you do. Just fix the problem.”
18:23
“It happens in the friction, it happens in the recall, and the quizzing yourself.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
So if you're trying to learn to code today, you've probably asked yourself some version of this question which is, am I wasting my time? Or will software engineers even be needed in a couple of years? Or is AI about to make everything that I'm learning completely irrelevant?
And here's my honest answer. So I've been a software engineer for over fifteen years, currently working at Meta, and I continue to ship code every single day with AI both at work and for my side projects.
And the way that I'm looking at things right now, yes, I think coding is definitely still worth it in 2026. I don't see software engineers going anywhere in the near future, but the way that I would learn to code today looks very, very different from how I learned it before.
And that's because the way that I code today is fundamentally different from how I coded even just a year ago. So in this video, I'm gonna show you exactly how I would learn to code if I were starting from zero today.
I'm gonna give you the step by step plan, the same plan that I would give to a beginner in my community, um, along with an approximate timeline of how long it would take to go from a complete beginner to a decently competent engineer.
Um, I'm gonna tell you which tools I would start with and some of the popular ones that I would avoid. And then I'm gonna introduce a method that I call , which will make learning faster and a lot more durable.
And finally, I'm gonna talk about how to work with AI in a way that actually teaches you instead of having it do everything for you. Because there's one mistake that I see beginners make almost every single day. It's a mistake that I still make sometimes that makes working with AI actually worse than some tutorials.
So if you're able to avoid this, you'll be ahead of almost everyone starting out right now. Okay. So one of the first things that we're gonna need to talk about is of course AI.
And the industry as a whole is moving more and more towards AI generated code. So any learning today with code needs to take that into account.
So if I were starting out today, I would try to leverage AI as much as possible from my learning. That way you get to work on two very essential and valuable skills at the same time, which is coding and working with AI. So step one, um, the first thing I would do is to get a subscription for a coding harness or a coding agent.
And, um, you know, if you wanna stay competitive to in today's world, you will have to learn how to use a coding agent. So my personal preference is ClaudeCode, but you can also use Cursor or Codex.
Now I prefer ClaudeCode because it feels in general more interactive than codex. So codex is, um, generally better at taking a task and working it through to completion.
So it's very good at the planning and it's able to work autonomously on a task for a longer period of time. And that's good for certain situations, but I think for a beginner, the sort of feedback and interactivity is, uh, is more valuable than just having an agent do everything for you. Now I would avoid some of the more popular vibe putting tools like Lovable, Base 44, Replit Because they abstract away some of the complexity that makes it very user friendly.
But if you are learning to code today, you really wanna pick a tool that you would use on the job. So most engineers in the industry actually will use one of the three that I just mentioned which is Cloud Code, Codex, or Cursor.
I think Cloud Code having the majority of the market share. So that way you're developing these AI skills that are actually transferable. Now for the tools that I mentioned, you will have to pay for a base subscription in order to use the coding agent.
So normally they're around $20 a month, but I think the value that you get from it will be really really worth it. I would avoid some of the free ones that are available out there just because they're just not quite up to par with what's available.
And you really wanna be working with tools that that the industry professionals use. So again, yeah, I would choose one of those three.
And after you've picked a coding agent, um, I would just start learning some of the basics with working with an agent. So don't try to learn everything at once. A lot of it will just come as you work with it more.
But I would just learn some of the important fundamentals like, uh, context, maybe context management, some things about memory, model selection, plan mode. Now that's a lot.
Um, I have a couple of tutorials, but there's also a ton of free tutorials online. I would spend about roughly just like an evening, you know, install the the coding harness, the coding agent, and maybe just watch, like, one or two tutorials about it.
And that's basically all you need to get started. So, yeah, again, don't try to learn everything at once.
Just start small. Um, and really just an evening or two spent on this would be enough to get you started. But, yes, like I said, I do have a couple tutorials on my channel for both Claude code and some of the basics around context management skills.
So please feel free to check those out. Okay. So that's step one, picking a coding harness and a coding agent.
So step two is where you wanna pick a starter project. Now this is where I wanna introduce the idea of a knowledge tree. And your knowledge tree is going to be the anchor that you attach all of your learnings to.
So I think one of the things that happens when you're trying to learn something like coding, for example, that a lot of beginners will try to, you know, pick up a specific language like, you know, I wanna learn JavaScript or if wanna work on front end, I'm gonna learn HTML or CSS. And I view these as just kinda individual concepts or .
Now having a structure or an anchor to attach the leaves on makes it a lot easier to kinda form relationships and make the learning a lot more durable. So if you have a project, which in this step by step process, I recommend step two is to get a project.
It acts as the tree, and this is the anchor that we're gonna be attaching all of these leaves onto. So once you have your project, I'm gonna be walking through this step by step. As you're gonna I'm gonna start defining some of the branches, and you can use AI to help you with this.
Again, I'll walk through this step by step. And once you have the branches, then the individual leaves that you learn become a lot easier because now they there's relationships between them and you know where each leaf or which piece of the puzzle, like, fits where.
So I think this is a much better way to learn something is when you have something to attach it to. Right? So now we're we're learning something like API.
We're like, okay. Now when I try to think about what an API is, I I think about it in the context of my project. And it's no longer just this dangling leaf that, you know, I might pick up, um, in the future somewhere.
So, um, I really wanna kinda use this knowledge tree as we go forward in the next next couple of steps.
So step two is to pick a project. Now I have some suggestions about how to pick your projects.
Uh, first, I would say try to pick something that matters to you. So try to solve a problem that you have.
Now picking up a project like this serves a couple of purposes. So if you pick a project that it's actually meaningful to your life, there's more motivation to continue coming back to it. And one of the the best feelings or one of the reasons that I became a software engineer, and I hope that, you know, this is you get this feeling too, which is, um, you know, I think one of the best things about being a software engineer is being able to see a project come to life.
Now that's building something and having it in your hand or either or someone else's hand and actually seeing the product come to life and being have it being useful.
So if you have trouble coming up with an idea, you can actually just have AI interview you about your life. It can ask it to ask you about your hobbies, your routines, your work, and have it pitch you ideas of what you can build.
So, um, one of the things you also wanna watch out for here is you don't wanna build something that's way too complicated. So, you know, maybe like a two sided marketplace with payments, uh, shop builders, like that might be too hard for a first project.
So when you have an AI interview about a project, you can also tell it about your general experience with building products and software, and it'll pitch you ideas that are sort of appropriate for your current level. Now just one quick note for each of the steps after step two, I'm gonna be providing a skills pack that you can install that will walk you through every single one of these steps to make it easier for you.
So, um, also, if you don't feel like installing a skill, I have I'm also gonna be providing this, um, set of prompts that you can easily just copy and paste into your coding agent that will take kinda take care of each each of these steps for you. So, for example, step two for picking your project, this is the prompt that I would feed into your coding agent.
It just says, hello. I'm looking for ideas for a project we can build together.
Uh, my current experience with software is x y z. Please interview me about my life, my work, routines, hobby, anything I do to repeatedly that software could help streamline. And, of course, this is the sizing that's important.
So something that's challenging but not overwhelming. So, yeah, I will be providing this also, and the link will be in the description.
Okay. So step three, after you define the project, is you want to define an MVP. Now if you don't know what MVP is, it's a very common acronym in engineering.
It just stands for minimum viable product, which just means what is the smallest thing that I can ship end to end. So why is picking an MVP important?
I think one of the most valuable feedback loops for an engineer is to be able to go from an idea and all the way to the finished product. So like I said, one of the most rewarding parts about being an engineer is having that feedback loop of, wow, I actually have this thing built in my hand.
And one of the things that can get people stuck in the beginning is you end up working endlessly on a feature and it just stays on your computer forever. And so defining an MVP allows you to just have your idea, the very minimal set of features that you can ship to have something working end to end.
And then once you have that, then you can start layering on more and more features. And I think this is a much better way to to build things out than just to than to rabbit hole down a complex set of features.
So how this works is you've picked a project. Let's say in this example, we're gonna be building a fitness tracker. So obviously, a fitness tracker, you can have a billion of these features like you can have an AI coach, sleep tracking, calorie counter, dark mode leaderboards, all of that.
But what do you really need for just version one of it? Alright. So fitness tracker, if we just break it down to the bare essentials, you wanna be able to log a workout and have your workout history.
So just taking those two and having that be your minimum set of features that you can build an app, and that is gonna be your MVP.
So you might have all these features that you wanna build, and you can put those into the backlog. And once that once you've built your your your first shippable product, anytime you layer in one of these features, you can actually see it live on your phone.
And also, I think this is just a good habit to build if you're trying to do any sort of software startup is that, um, this is the same strategies that most startups follow is you build the very bare minimum, the minimum viable product, and you start showing that to users, getting feedback, and iterating before you start working on a ton of features that might not be used.
Okay. So step three is to define an MVP, and again, we have a prompt here that can help with that. I won't read through this entire thing, but, basically, um, it'll ask your coding agent to help you define the minimum set of features that makes this product viable.
Okay. So after we have defined our MVP, step four is to start building out . So if we go back to our project here, this is starting to lay the foundations of, like, what are the core pieces that I'm working with.
So this is to help your mind get a scaffolding when you start building out the branches and the leaves. Again, this will help you anchor your knowledge. Now because the skill of software engineering has has sort of moved up a level, I think it's important for beginners to start thinking about the system as a whole from day one.
And this is a really good way to do that is to clearly define and see at a high level what are the core pieces of shipping a product from start to finish. So I'm gonna show you an example here of a project that has been broken down into the core components.
So this is the trunk. This is a project called the book club organizer. And the core components of and a lot of these core components will be shared between projects, and sometimes they'll differ a little bit depending on, you know, if you're working on just like a pure back analytics thing, you might not have, like, an API or or a client.
But a lot of these core pieces will be shared between projects. So at the base, you have you know, you definitely wanna have source control and usually use Git for that. And some of the other core pieces, of course, you have the back end, the database, the the front end, the API, and the the then the deployment.
So that's the that'll make up the core and the trunk of your tree. So again, there's a prompt to do this. It's also in the skills pack, but, uh, basically, it will take your project and the MVP that you defined, and then break it down into the core pieces.
So it's good to just read through this to to understand at a high level what you're going to be building in your project. Okay. So we have the trunk defined, and now we're going to start defining the branches.
Now the branches are the high level sections or the features, the the horizontal slicing of how you're going to ship your project.
And what and what that looks like is something like this.
So your branches would be, for example, the first section would be getting running with your first page. Now if you're working on a web app, this is literally just serving a plain HTML page.
The next step might be having some interactions on your HTML page.
So inserting things like JavaScript, and in this case, we're gonna be using React. Um, the next horizontal slice there would be giving it some memory and database, and each of these branches and each of these sections start to build on top of each other.
And these will be the branches of your tree. I know it looks different on this tree, but, um, essentially, that's what we're trying to define is the the individual features that build on top of each other.
And so that's step five, and then step six is actually defining the leaves, are the individual tasks. So you start with the high level spec of what you're trying to build, and then you start breaking down the individual features, and then you go into task.
So what these individual leaves look like, so if we're just starting out from the very bare essentials. Right? So getting started with git.
So the first thing that you need to do is to open up a terminal and run a couple of commands, install git, install node dot j s. And there will be learning concepts that are weaved within each of these individual tasks, and that's where we start getting into the weeds of it.
So step six is also where things start to get really interesting because this is where we start actually start leveraging AI to help us learn. And this is where I wanna introduce the idea of this knowledge graph.
So the knowledge graph is basically a graph of everything that you've learned and everything that you're going to learn, and it goes all the way down to very low level concepts like CSS styling, variables, and it can include higher level concepts and ideas such as git and commits.
But this is essentially the graph the knowledge graph that you're gonna be building with. And so in step six, when the AI breaks down your tasks, as you start working through the tasks, it will also construct a knowledge graph.
And it will take a look at the task and figure out, okay, what are some of the core concepts that we're gonna be covering within this task? And for each of those, it will it will first check if there's already an entry for that, and if not, it will create an entry for it.
So the power of this knowledge graph is keeping track of your status, the last time it was reviewed, and then a list of evidence. So as you're mastering a concept, it'll go from, like, for example, introduce to practicing. The next step might be something like understood and then finally mastered.
And depending on the status of a certain concept, the next time that it shows up in your plan or your task, the agent can know exactly how in-depth to go into each one. So again, for example, if if we've already introduced what variables are in task two and you have a pretty good grasp of it, when it comes up again in a different task down here, um, AI doesn't need to, you know, teach you about it or quiz you about it quite as much.
So step seven, which is the final step of our learning process here, this is the step that you're gonna be spending the majority of your time in. So what step seven instructs the agent to do is to follow that checklist that we made, construct the knowledge graphs, and then and then provide quizzes as you're going through each of the task.
So this will actually solve the problem that I brought up earlier in the video and the mistake that most beginners make, which is building things way too fast with AI and not understanding any of it. And, you know, this is this is something that's really, really hard to do on your own.
Like, even I still make this mistake sometimes when, you know, I get tired or I get lazy. I have a tendency to just let the AI do whatever it wants to and just be like, okay. I don't care what you do.
Just just fix the problem. And when you do that, you're not learning anything and that's a way for your code bases to kinda snowball out of control where you don't understand anything anymore.
So, yeah, it's it's a very common problem and it's just because of the the culture that we live in and just how fast AI is able to do things. So having this forcing function, forcing you to slow down and look at every single line of code, quizzing yourself on every single line of code, yes, it does slow you down a lot, but this is where the the learning happens.
It's it happens in the friction, it happens in the recall, and the and the quizzing yourself.
So step seven is where we introduce that friction and using this prompt going through one task at a time and making sure that we understand every single line, every single concept that's introduced within that task. So let me show you a quick example of what a step seven looks like.
So I recorded, um, one of the transcripts as I was testing this.
And for example, let's take a look at this task right here. So in this task, we are introducing for the first time the idea of a variable.
So the example that we're working with is we're rendering an HTML page with initially three line items.
So we're rendering a list of of books. And initially, we just have that hard coded directly into the HTML page. And what we wanna do with variables is instead of having it hard coded, we wanna store all of the books in this array.
So, um, an array called books as a variable. And then so instead of hard coding each of the lines, we're using some kind of loop through the variables to do the rendering.
So what the AI will do is introduce what that task is about. It will generate some code for you, and then it will quiz you on it.
So for example, before it runs a code, it'll ask you, okay. Um, before we run this, uh, in your own words, what do you expect to happen when we run through this code?
So this is what makes the the learning truly interactive and puts you in the seat of being fully involved in in the learning. And as you complete each of these quizzes and you go through each of these tasks, the AI goes back in and updates your knowledge graph.
So for example, this is from task 2.1. The the core learning here was variable. And if we go to our knowledge graph, there is a section here called variable, and we can see that status isn't practicing because we've been quizzed on it once and we're still learning that concept.
So when we come back to it again, it might update into understood and then eventually into mastery. So I think this is just a really powerful way to learn because one of the biggest complaints I hear from beginners is that they get lost and confused about, you know, what to learn, how to learn it, and just trying to navigate everything that you have to learn.
So when you have a knowledge graph that keeps track of everything that you've learned so far and everything that you're going to learn, um, you have this this really simple step by step program that you that you follow. So repeating step seven through all of the your sections will basically take you from a complete beginner all the way to shipping and deploying your app.
And at the end of the project, you will actually have confidence that you've not only do you understand every single line of code, but you've had some repetition and practice with each of the concepts. Okay.
So I know we covered a lot here. So just to do a quick recap of all of the steps.
So step one is to pick a coding agent to work with. Step two is to pick a project. Step three is to define your MVP.
Step four is to start building out the trunk of your knowledge base. Step five is to build out the branches. Step six is to build out the leaves and then define knowledge graph.
And then finally, you repeat step seven until you finish the entire project. So in step two through six, you can probably get through within an evening or two.
Just one evening if you just spend a few hours on it. Step seven, depending on how complex the project is, if you spend maybe roughly ten hours a week working on it, I'd say it would take a minimum of a month up to two or three months.
But if you really take the time to focus and slow down and work through it by the end of it, you will have a really solid foundation going into whatever you do next. So you'll have a portfolio piece and be able to explain every single line of code in it, and you will have an understanding of the system and the whole architecture, and again, just a really really solid foundation as a software engineer.
Now again, you don't have to memorize all these steps yourself. I'll be providing the set of prompts that you can use for each of these steps.
And if you want, I will also be providing the skills pack that will walk walk you through all of these steps without you having to enter in the prompts manually. Okay. So this is how I would start to learn to code if I had to start from scratch today.
I think as you're learning this way, you'll also get experience of what it's like to work with a coding agent. You know, what are some of the capabilities it has, some of the limitations.
You'll naturally start learning things like context management, um, picking the right models for things because, um, on the $20 plan, you will have to try to manage your your token budget.
So those are things that just come with the territory as you're learning more and more. So the final thing that I wanna talk about is I I will be providing all this for free, but I I know that even with these skills and prompts that it's easy to potentially get lost.
And so to make things even easier, I am working on, uh, this is something that I introduced in one of my previous videos, something that I called the AI coding harness. Now just to give you a rough idea, these are some of the prototypes and wireframes that I'm working with.
But the idea is you have a interface that keeps track of all of that for you. So, um, you'll be able to visualize your knowledge graph in something like this, um, in a web interface.
So the idea is that it will plug into your whatever coding agent you do. It'll keep track of your sessions and then give you a nice UI to to interact with.
So you'll have a nice, um, knowledge graph visualization, um, with all of the things that you've learned and the evidence provided for each one.
Um, you'll have the step tracker so that you'll always know exactly where you are in your project. And one of the other things that I think will also be important for this is, um, I wanna introduce this idea of an assessment.
So some of you might be coming in with varying levels of knowledge, um, and coding experience.
So one of the things I wanna do initially is to have an assessment to figure out where you are in your coding experience. So this will sort of seed some of the initial knowledge graph so that, you know, if you're already comfortable with some of the the ideas and concepts, it won't try to requiz you on on those things.
So if this is something that you're interested, I do have a wait list that I will be putting up, and the link will also be in the description. And finally, I just wanna do one final plug for the community.
So if you are just starting to learn coding by yourself, even with all these tools, it can be somewhat intimidating. And having a community of people who are starting out just like you and having a place where you can go and and share knowledge, talk to people, ask questions, and just be in a community of people who are going through the same thing can make this journey a lot more fun.
So if that's something you're interested in, the workshop is also completely free. The link will also be in the description. So, yeah, that is it for this video.
If you have any questions or comments, please feel free to drop it in the comment section. And, again, I also have a few other tutorials on in my channel that will help you with some of the basics of starting out.
So feel free to check those out as well. But other than that, thank you for watching, and I will see you in the next video.
The Hook
The bait, then the rug-pull.
The question every beginner is quietly asking — am I wasting my time learning to code while AI writes more of it every month — gets a straight answer up front: no, but the method has to change completely from how anyone learned it even a year ago.
Named ideas worth stealing.
04:56model
The Knowledge Tree
- Trunk — core project components
- Branches — ordered feature slices
- Leaves — individual concrete tasks
A structure for anchoring every new coding concept to one real project instead of learning topics in isolation: trunk (shared core pieces), branches (features shipped in order), leaves (individual tasks where concepts get introduced).
Steal forany self-directed skill curriculum where scattered tutorials fail to stick — content courses, onboarding docs, cohort curricula
00:00list
The Seven-Step Learning Plan
- Pick a coding agent
- Pick a project (knowledge tree)
- Define the MVP
- Build the trunk
- Build the branches
- Build the leaves
- Follow the plan one task at a time
The full beginner curriculum condensed to seven ordered steps, moving from tool selection to a real, understood, shipped project.
Steal forstructuring any beginner onboarding sequence that needs a concrete end state, not just a topic list
09:17concept
The MVP Definition
An MVP is defined strictly as the smallest end-to-end shippable slice — not a feature-reduced product, but the minimum thing a user can actually use — demonstrated with a fitness tracker cut down to log-workout plus history.
Steal forscoping any new product or feature before greenlighting a longer build
How they asked for the click.
VERBAL ASK
25:18link
“Join the free Workshop community”
Soft single mention at the very end, after all seven steps of value have already been delivered. A second, separate plug for a paid-adjacent product waitlist (the 'AI coding harness') runs earlier at 23:33 — a two-tier CTA (free community now, paid tool later) rather than one hard pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
