Modern Creator
Robin Ebers · YouTube

5 Things I Wish I Knew Before I Started AI Coding

A nine-minute field report from a founder who shipped real apps without writing a line of code for 18 months.

Posted
today
Duration
Format
Listicle
sincere
Views
416
27 likes
Big Idea

The argument in one line.

AI coding tools are interchangeable commodities, so the builder who ships is not the one with the best model but the one who stays long enough with a single tool to understand what they are actually building.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You have started one or more AI coding projects but keep stalling before shipping.
  • You spend more time switching between Cursor, Claude Code, or Windsurf than you spend building.
  • You have shipped something with AI but feel secretly unsure whether it is production-safe.
  • You are a non-technical founder who wants to manage an AI like a developer without writing code.
SKIP IF…
  • You are a working software engineer looking for technical depth — this is a mindset video, not a tutorial.
  • You have already internalized the context-first, stay-with-one-tool discipline and are past the beginner stage.
TL;DR

The full version, fast.

After 18 months of no-code AI building, the five lessons that matter are: stop switching tools because they are all the same; give your AI rich context instead of lazy prompts; use a small number of purpose-built MCP servers to make AI domain-competent; you decide what and why while AI suggests how; and you must understand your app's user-journey blueprint even if you never read the code. Confidence in shipping comes from blueprint literacy, not from the quality of your model.

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 →
Chapters

Where the time goes.

00:0002:42

01 · Stop obsessing over which AI tool to use

Tool-switching is the shiny object trap. After testing 100+ tools, the verdict: Cursor, Claude Code, and Codex are functionally identical. Pick one and stay long enough to feel confident.

02:4203:42

02 · Never just say fix this

Vague prompts produce band-aid fixes that compound into a house of cards. Treat AI like a developer: provide the symptom, the context, and the expected behavior.

03:4205:31

03 · MCP servers and AI skills — selectively

Most MCP tools are trash, but domain-specific ones transform AI from tourist to local guide. Built a macOS recording app with zero prior Mac experience using the right tools.

05:3107:34

04 · You decide what and why — AI suggests how

AI defaults to complexity. A client built a 47-page rocket-ship plan for a bicycle-scale app. Ask for three options and pick what you understand. If you do not understand it, do not build it.

07:3409:31

05 · Coding is dead — thinking like a developer is not

Prompting and hoping builds black boxes. You must be able to draw the user journey from sign-up to payment. Blueprint literacy is the minimum viable developer mindset.

Atomic Insights

Lines worth screenshotting.

  • Every hour spent evaluating a new AI coding tool is an hour stolen from the product that earns money.
  • Switching tools constantly keeps you a permanent beginner — you never stay long enough to feel confident.
  • Cursor, Claude Code, and Codex all look the same because they are the same for 95% of what non-technical founders build.
  • Confidence in AI coding comes from time with one tool, not from choosing the objectively best one.
  • A vague prompt like 'fix this' is a band-aid on a broken bone — the real problem hides and compounds.
  • Treat your AI like a developer: explain the symptom, the expected behavior, and what broke.
  • Without enough context, AI guesses — and each guess builds a house of cards that collapses at launch.
  • Most MCP servers are trash, but the right domain-specific ones make your AI feel dramatically smarter.
  • An AI model without tools is a tourist with a stale map; an agent with the right tools is a local guide.
  • AI defaults to complexity because it has no concept of your time, skill level, or shipping deadline.
  • A 47-page AI-generated architecture plan for a thousand-user app is a rocket ship when you need a bicycle.
  • You must decide the what and the why — AI can only suggest the how.
  • Ask AI for three options on every architectural decision, then pick the one you actually understand.
  • The rule that prevents most project failures: if you do not understand what you are building, do not build it.
  • Coding is dead; thinking like a developer is not. You need blueprint literacy, not syntax knowledge.
  • People who never ship are scared to hit publish because deep down they know they built a black box.
  • Prompting and hoping is not a development strategy — it is deferred fear.
  • If you cannot draw the user journey from sign-up to payment without code, no AI can save the project.
Takeaway

Five rules for building apps without writing code.

WHAT TO LEARN

The fastest path to shipping an AI-built app is staying disciplined about tool choice, context quality, and blueprint understanding — not finding the best model.

  • Switching AI coding tools constantly resets your confidence to zero — Cursor, Claude Code, and Codex are interchangeable for most use cases, so pick one and stick with it.
  • A vague prompt like 'fix this' produces a surface repair that hides the real problem and makes the next bug harder to find; describe the symptom and the expected behavior every time.
  • Most MCP servers and AI skills are not worth installing, but the handful that match your specific domain can make your AI behave like a subject-matter expert instead of a generalist guesser.
  • AI defaults to complexity because it has no awareness of your time or skill level — always ask for three options on any architectural decision and pick the one you actually understand.
  • You do not need to read or write code, but you must be able to draw the user journey from sign-up to payment; if you cannot, you will be too scared to publish because you have built a black box.
  • Confidence does not come from using the most powerful model — it comes from understanding what your app does at every step, which is blueprint literacy, not coding.
Glossary

Terms worth knowing.

MCP server
A plugin-style tool that gives an AI agent access to domain-specific knowledge or capabilities so it can act more like a specialist than a generalist.
AI skill
A pre-built prompt or workflow module that equips an AI agent to handle a specific recurring task with expert-level accuracy.
Black box
An app built entirely by AI prompting where the builder has no understanding of the internal logic — reliable in calm conditions but impossible to debug when something breaks.
Blueprint literacy
The ability to trace a user's journey through an application from entry point to key action without reading code; the minimum viable developer mindset.
Prompting and hoping
Firing vague requests at an AI and accepting whatever it produces without verifying correctness or understanding the implementation — common among beginners, dangerous at scale.
Resources

Things they pointed at.

02:00toolCursor
02:04toolCodex
Quotables

Lines you could clip.

01:10
No one pays you for the plumbing. They pay you for the product.
standalone, punchy, no setup neededTikTok hook↗ Tweet quote
06:20
The AI had designed a rocket ship when all he needed was a bicycle.
vivid metaphor, lands hard without contextIG reel cold open↗ Tweet quote
08:35
Coding is definitely dead, but thinking like a developer is not.
contrarian, quotable, polarizing in a good wayTikTok hook↗ Tweet quote
The Script

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
00:00Did you think that building an app with AI and without a developer is going to be easy? I did. It's been eighteen months since I stopped writing and reading code, but getting here was really hard, and I wish I had known how to do it smarter.
00:11To make building without developers 10 times easier for you, here are five things I wish I had known about AI coding before I got started. So let me start with the mistake that cost me the most amount of time, so you never have to make it yourself, obsessing over which AI tool to use.
00:26There are way too many AI coding tools and literally hundreds of AI models, and people keep obsessing over which one to choose. But every hour that you spend picking and switching is an hour that you could have spent building. Over the past eighteen months, I tested over 100 AI coding tools and models.
00:43And every time a new one dropped, I got distracted. I switched over. I configured it.
00:48I tested the model. Major shiny objects had dropped. And while I was busy setting up Cursor yet again or reconfiguring Cloud Code, you know what was happening to my actual business?
00:58Nothing. I wasn't building anything. Meanwhile, the TikTok kid down the road with a less optimized setup and just a good enough tool shipped plenty of features that actually built his business.
01:10I know because I see this on Twitter every day. It took me way too long to understand that the tech is just the plumbing. No one pays you for the plumbing.
01:19They pay you for the product. And it gets worse because switching not only costs you time, it keeps you feeling dumb.
01:27Every time I jump to a new tool and hit a new bug or something wasn't working, I would hold on a second and ask, is this me or is this the app? Did I set it up wrong? I would just spiral blaming myself, blaming the model, blaming the tool, but never actually learning how to fix the thing.
01:42And now I see this in founders I work with all the time. When you constantly switch, you stay a permanent beginner. You never stick with one tool long enough to truly understand it.
01:51So all the AI coding tools are basically the same. This has never been more true than today. Cursor, Claude code, and Codex quite literally look the same with chats on the left, the actual chat in the middle, and then there's always something on the right, like a preview or a browser.
02:09And you can see this in Cursor, you can see this in Claude code, and you can see this in Codex. They all look the same and do the same. Whether it's GPT, Claude, or even the Chinese ones, all of them are good enough for what we do today.
02:22So pick one and stick with it. Stay long enough to feel confident because that confidence never comes from the best tool. It comes from understanding the one that you have used the longest.
02:33Next, I will show you why you should never ask your AI to fix this because it always leads to disaster. The biggest mistake beginners make is underestimating how much context the AI actually needs.
02:44They think they can just type fix this, and the AI is smart enough to do the rest. I haven't written a line of code in about eighteen months, like I said, but I still act like I manage a team of developers because a manager would never walk up to their developers and just say, make it work or fix this. They provide context.
03:01What is actually broken and what is this supposed to look like or what is it supposed to do? But when I first got into AI coding, I didn't do that. I thought I should could just get away with saying, hey.
03:11Do this, and the AI would be smart enough to figure out the rest. I learned the very hard way that this isn't always true because I never gave it enough context. The fix was like putting a Band Aid on a broken bone.
03:22It looked solved for a second, but the real problem was still hiding in there. And the next time something broke, it was even harder to fix.
03:31So if you truly want to give your AI the best chance of fixing it the right way, you need to provide as much context as possible. Treat your AI like a developer. Explain the problem and the symptom.
03:44Because if you don't, the AI guesses, and it will build you a house of cards that collapses the moment you publish your app. Next, I will show you the one thing that makes any AI agent 10 times faster than it is today. MCP servers and AI skills get massively overhyped.
04:00I get it. But that doesn't mean that you should skip them. Used right, they are the closest thing that you have to an expert developer on speed dial.
04:08Because like I said, it all comes back to context, and MCP servers and AI skills are just another way of doing just that. Except this time, it's expert level context for a very specific thing. Here's what I mean.
04:22You are likely stuck because you still think that you need to understand code or that you need to understand how to build a very specific app. But I'm living proof that you don't because I recently built my own YouTube recording software. By the way, this is exactly what I'm using right now.
04:36And when I look at my screen, you cannot see it unless I take out my phone, which through the magic of video editing, Jason, my editor, will show you right now. So I can see myself and my screen, and it runs on my system. I had never built a Mac OS app in my life.
04:51I had no clue how to record audio or video, let alone pausing and resuming mid recording. But here's the thing. When I tried to build it, it didn't really work.
04:59The sound was often delayed. The videos weren't correctly written, and my video editor that I just talked to you about complained that his video editing software didn't read the files correctly.
05:10I am not an expert in any of this, but I knew that something had to change. And so I equipped it with purpose built AI skills and some MCP servers that ended up being incredibly useful.
05:21You can think of it like this. An AI model without these tools equipped is like a tourist in a foreign city with a map that is slightly outdated, but an agent with the right tools is like having a local guide who's lived there all of their life.
05:35So, yes, most MCP servers and AI skills are trash, but there are a few, depending on what you build, that are more than worth it. And without them, your AI will feel a lot dumber than it actually is. Now I'm going to show you the one decision that makes or breaks your whole app so you don't end up with a project that never actually goes live.
05:54Most beginners let AI decide on the big things. What tech stack to use, even what features to build, that is completely backward. The AI is your employee, not your boss.
06:05The moment you flip that around, you are headed for disaster. And before joining my program, a client of mine wrote a 47 page plan with AI that he didn't really understand. And here's a tough one.
06:15He thought that he wasn't smart enough to understand this plan, But when I read it, I didn't understand it either. There were so many features, so many edge cases, so many outside services coming together for an app that at most needed to handle a few 100 or maybe a few thousand users a day. The AI had designed a rocket ship when all he needed was a bicycle.
06:37But why does this actually happen? Well, it's because when you let AI decide, it defaults to complexity. It doesn't care about our time.
06:44It has it doesn't have a context of time. It doesn't care about your skill level or if you actually ever ship this thing. And in the last eighteen months of working with founders, building with AI, here is the biggest mistake I saw them make again and again.
06:58They let the AI pick whatever sounded the most impressive, like a database built for millions of users, all the things that they really didn't need. So you must decide the what and the why, but you can ask the AI to suggest the how.
07:13When you plan your app, ask for three options. For example, where should I host this app?
07:18What is the best database for this app? Then you pick based on what sounds easiest to you. Continue to ask questions until you really understand because here is the one rule that will save you from a lot of pain.
07:31If you don't understand what you're building, don't build it. Now let me show you the one change that separates people who ship from the people who don't so that you never feel scared of your own app ever again because you are being sold a lie.
07:43And that lie is that you don't need to understand how any of this works anymore, that you can just prompt your way to a finished application and never think about what's under the hood. I'm here to tell you that this is not true. You must have heard the horror stories of leaked passwords and websites breaking for no reasons, customers getting charged twice, whole projects being held hostage by hackers trying to extort their owners.
08:07It's all real too, But all of these stories had one thing in common. Nobody understood what they were actually building. They were doing what I'm calling prompting and hoping.
08:16They wish for something magical as if they had, like, a magic genie in a bottle, and then it magically appears. But they never stopped to understand what the app was actually doing under the hood. All they did is build a black box.
08:28And when you build a black box, you will never actually feel sure about what it is supposed to do. That's why so many people never actually ship their app. They're too scared to hit publish because deep down they know that they have no idea if this thing is good enough.
08:43So here is what I need you to understand. Coding, like the actual code, is definitely dead, but thinking like a developer is not. You will never have to read or write a line of code in your life, but you do need to understand your app's blueprint.
08:56Think of it like a map of how a user goes from clicking sign up to actually paying for your services. If you can't draw out this map with all the things that happen in between, then no amount of AI can save you. And that is what I teach.
09:10I turn you into a person that can look at an AI generated feature and go, nope. That doesn't feel right. There's something off about this.
09:17So you ask questions to figure out why because this is true confidence. At the moment that fear disappears is the moment you start feeling like a true independent AI builder.
09:28If this was helpful, hit subscribe, and I'll see you in the next
The Hook

The bait, then the rug-pull.

Eighteen months without writing a line of code, and this founder has the scars to prove which mistakes cost the most. The promise is five things he wishes someone had told him before he started.

Frameworks

Named ideas worth stealing.

02:42model

Developer/Manager Context Loop

  1. State what is broken
  2. State expected behavior
  3. Let AI fix with full context

Managers brief developers with context — the same discipline applied to AI prompts prevents compounding bugs.

Steal forany tutorial on writing better AI prompts
07:13concept

Three Options Rule

  1. Ask AI for three options on any architectural decision
  2. Pick the one you understand
  3. Keep asking until you understand your choice

Prevents AI from over-engineering by forcing the builder to make an informed choice rather than accepting the first complex suggestion.

Steal forany architecture or tech-stack decision in an AI coding project
08:35concept

Blueprint Literacy

  1. Draw the user journey from entry to payment
  2. Identify every step in between
  3. Be able to explain what each part does

The minimum viable developer mindset: not code literacy, but flow literacy.

Steal foronboarding non-technical founders to production thinking
CTA Breakdown

How they asked for the click.

VERBAL ASK
09:27subscribe
If this was helpful, hit subscribe, and I'll see you in the next

Minimal — single sentence, no overlay, no urgency. Clean but forgettable.

MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Storyboard

Visual structure at a glance.

open / hook
hookopen / hook00:00
5 Things title card
promise5 Things title card00:17
Cursor logo — lesson 1
valueCursor logo — lesson 100:53
Confidence from mastering one tool
valueConfidence from mastering one tool02:35
Developer/Manager graphic
valueDeveloper/Manager graphic02:42
Recording app B-roll — lesson 3
valueRecording app B-roll — lesson 304:06
Right tools make AI smarter
valueRight tools make AI smarter05:46
Simple vs Complex
valueSimple vs Complex06:54
Do not build if you do not understand it
valueDo not build if you do not understand it07:33
Coding Horror Stories list
valueCoding Horror Stories list08:01
Know how your app actually works
valueKnow how your app actually works08:58
Subscribe CTA
ctaSubscribe CTA09:27
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this