Modern Creator Network
Alex Finn · YouTube · 19:56

9 months of Claude Code Lessons in 19 minutes

Alex Finn built a $300K/year app and 50K subscribers talking about one tool — here are the eight lessons nine months of 12-hour days actually taught him.

Posted
5 months ago
Duration
Format
Tutorial
educational
Channel
AF
Alex Finn
§ 01 · The Hook

The bait, then the rug-pull.

Nine months. Twelve hours a day. A $300K/year app and fifty thousand subscribers built almost entirely inside one tool. Alex Finn has probably logged more hours in Claude Code than anyone outside Anthropic — and the lesson he keeps hammering is the one nobody wants to hear: stop adding complexity.

§ · Stated Promise

What the video promised.

stated at 00:32Whether you're a day one beginner or you're an absolute Claude code professional, I promise you, you're going to learn a ton in this video and be a master by the end of it.delivered at 19:00
§ · Chapters

Where the time goes.

00:0000:45

01 · Intro & credibility setup

Alex establishes authority: 9 months daily use, $300K/year app, 50K subscribers, 8 lessons nobody else talks about.

00:4502:41

02 · Lesson 1: Use the VS Code extension

VS Code extension beats CLI-in-Cursor for stability and UX. Cursor is a fork; VS Code gets upstream updates first and has zero layout bugs. Free, no subscription.

02:4105:56

03 · Lesson 2: Ignore the gurus — keep it simple

Sub-agents, MCPs, 40 slash commands = content-farming noise. Sonnet 4.5 is smart enough to find docs on its own. 99.9% of users need nothing beyond a clear prompt.

05:5607:49

04 · Lesson 3: Select Haiku for 10x credit efficiency

With /model Haiku selected, Claude still uses Sonnet for plan-mode planning but drops to Haiku for implementation. Thorough Sonnet plan + Haiku execution = same quality, fraction of the cost.

07:4910:19

05 · Lesson 4: Don't waste downtime — Copilot CEO

80% of vibe-coding time is idle wait. YC's Plinko IDE = worst product ever made. Use idle time with a desktop AI chat as co-pilot CEO: feature roadmap, marketing, life decisions. Doom-scrolling is the alternative.

10:1912:46

06 · Lesson 5: Nine-rule CLAUDE.md

Nine rules that eliminate lazy loops and whack-a-mole bugs: structured todo plan, check-in before execution, step-by-step explanations, simplicity above all, DO NOT BE LAZY x2, no temporary fixes.

12:4614:55

07 · Lesson 6: Prompt engineering is dead

Complex context/UI/database/schema prompt templates are useless. Plan mode handles context discovery. Just tell it what you want. Simple natural language beats structured prompting every time.

14:5516:29

08 · Lesson 7: Always include screenshots for UI

Default Claude UI = blue and purple gradients, always. Fix: drag in a v0 design system screenshot or Pinterest app-design board image alongside any UI prompt. Screenshot = art direction Claude can actually follow.

16:2919:00

09 · Lesson 8: Treat Claude as a creative partner

Before plan mode, run brainstorm mode: ask Claude for options, what to add, what to change. Demo: asks about folder system, Claude suggests hashtag tagging — that idea wasn't Alex's, it came from Claude and made it into the app.

19:0019:56

10 · Recap & CTA

Three-point recap: keep it simple, don't waste downtime, the AI is smarter than you think. Subscribe, newsletter CTA, alexfinn.ai/subscribe.

§ · Storyboard

Visual structure at a glance.

intro / credibility
hookintro / credibility00:00
VS Code extension
valueVS Code extension00:45
ignore the gurus
valueignore the gurus02:41
Haiku efficiency
valueHaiku efficiency05:56
copilot CEO / downtime
valuecopilot CEO / downtime07:49
CLAUDE.md rules
valueCLAUDE.md rules10:19
prompt eng is dead
valueprompt eng is dead12:46
screenshots for UI
valuescreenshots for UI14:55
creative partner
valuecreative partner16:29
recap / CTA
ctarecap / CTA19:00
§ · Frameworks

Named ideas worth stealing.

10:19list

Alex Finn's 9-Rule CLAUDE.md

  1. Think through the problem, read codebase, write plan to tasks/todo.md
  2. Plan = checkable todo list
  3. Check in before executing
  4. Work through todos, mark complete as you go
  5. Give high-level explanation of each change
  6. Make every change as simple as possible — minimal code impact
  7. Add review section to todo.md
  8. DO NOT BE LAZY. NEVER BE LAZY. FIND ROOT CAUSE. NO TEMPORARY FIXES.
  9. SIMPLICITY ABOVE ALL. MINIMUM CODE IMPACT. ZERO NEW BUGS.

Nine rules delivered as prompts in CLAUDE.md that prevent lazy loops, whack-a-mole bugs, and over-engineered solutions. Rules 8 and 9 are all-caps by design — it changes the model's behavior.

Steal forDrop directly into any project CLAUDE.md. Rules 8-9 as all-caps is the key detail most people miss.
16:29model

Brainstorm Mode → Plan Mode

Before entering plan mode (shift-tab x2), first run a free-form brainstorm: ask Claude for options, what features to add, what it would change. Only then switch to plan mode with your chosen direction. Separates idea generation from execution planning.

Steal forJoeFlow Sessions orchestrator: run brainstorm before each batch job prompt to surface better approaches before committing to a plan.
05:56model

Sonnet Plan + Haiku Execute

Use /model to select Haiku. Claude Code still uses Sonnet for plan-mode planning (shift-tab x2), then drops to Haiku for the implementation pass. A thorough Sonnet plan is detailed enough for Haiku to execute without quality loss.

Steal forAny high-volume JoeFlow session burning through credits. Saves ~60-80% token cost on implementation-heavy tasks.
§ · Quotables

Lines you could clip.

03:34
Stop listening to all the gurus. You don't need to use every single feature inside Claude code.
Contrarian opener that immediately hooks the over-complicated crowdTikTok hook
08:30
I saw this tweet from Y Combinator yesterday, and it literally made me sick to my stomach.
Visceral reaction to the Plinko IDE — turns into a rant that's highly clippableIG reel cold open
09:23
If you spend 80% of your day scrolling TikTok, your brain's melting out of your ears.
Vivid line, no setup needed, universal painTikTok hook
12:46
Prompt engineering is the most overrated concept in history.
Punchy thesis, guaranteed engagement from both believers and skepticsTikTok hook
17:40
It's not just some junior dev going out writing code for you. It's your partner.
Reframe that shifts mindset — newsletter pull-quote territorynewsletter pull-quote
§ · Pacing

How they spent the runtime.

Hook length45s
Info densityhigh
Filler8%
§ · Resources Mentioned

Things they pointed at.

08:30productYC Chat IDE (Plinko IDE)
10:19linkAlex Finn CLAUDE.md rules (in video description)
§ · CTA Breakdown

How they asked for the click.

19:00newsletter
Sign up for the number one AI newsletter on planet Earth. It's completely free. Link down below. Alexfinn.ai/subscribe.

Soft — recap comes first, then subscribe ask, then newsletter. No hard sell. Mirrors the video's 'keep it simple' ethos.

§ · The Script

Word for word.

HOOKopening / re-engagementCTAthe pitchmetaphoranalogystory
00:00HOOKOn 02/24/2025, Claude Code released, and not a single day has gone by where it hasn't been the best AI coding tool on the entire market. Since then, I have been using Claude Code twelve hours a day without fail. I've absolutely mastered it. I've built and released full scale apps making over $300,000 a year now, and I've also built up this entire YouTube channel, over 50,000 subscribers, almost exclusively talking about Claude Code. Over that time, I've learned eight major lessons using Claude code, many of which I've never heard anyone else talk about before. Whether you're a day one beginner or you're an absolute Claude code professional, I promise you, you're going to learn a ton in this video and be a master by the end of it. Now let's lock in and get into it. So the first lesson I learned is is the extension inside Visual Studio Code is the absolute best way to get the most out of Claude code. There are many different ways to use Claude right now. You can use it through the CLI. You can use it through the CLI inside a cursor. You can use it extension inside cursor. You can use it on the web. You can use it in the mobile app. They've made Claude code very accessible many, many different ways out of all the ways you can use it for, and I've used it for hundreds of hours every single way you can use it. The extension
01:16inside of Visual Studio Code is the best experience. I've tried it, the CLI in cursor. I think most people are using the CLI in cursor to use Claude code. I've had many issues with that. One is the UI is just better inside the extension than in the CLI. You can open up multiple chats very easily, switch between them. It's formatted incredibly well. It's very easy to turn on and off a lot of the features, get options like the slash commands opened up really easy. It's just a way better experience than the CLI. But the reason why I use the extension
01:50inside of Visual Studio Code specifically is I've just had zero issues with it at all. As you probably know, Cursor and a lot of these other AI IDEs are just forks of Visual Studio Code. So Visual Studio Code is just going to have the most up to date updates and work the best out of all of them. I have a lot of time when I use the other AI IDs where the extension just doesn't open up correctly. It opens up in different places. It doesn't work correctly.
02:17I have never had a single issue using the extension inside of just your basic free Visual Studio Code. It's completely free. You don't need to pay anything monthly for it. Download Visual Studio Code for free. Get the Claude code extension, which is right in the extension store over here on the left. You open that up, I promise you're gonna have the best experience ever with Claude code that you can get anywhere else. My second tip for you is this. Stop listening to all the gurus. You don't need to use every single feature inside Claude code. Every time a new feature releases such as MCP,
02:53sub agent, slash commands, you get all of these videos, the top 15 slash commands you need to implement. Here are 40 sub this is my favorite one. Here are 40 sub agents you need to implement in order to get the most out of Claude Codens. These outrageous videos of people implementing literally a 100 different sub agents that you manually call as you go as you're building your apps. Stop listening to all these people.
03:21I get the best experience out of Claude Code when I just keep it as simple as humanly possible. There's a lot of gurus in content that will just make up these work flows that actually don't add any value. They just complicate the hell out of everything you're doing just so they have some more content to build. But I promise you, the main message I have in all my videos and all my livestreams is just keep it simple. When I'm using Claude code, I'm very rarely using any sort of slash commands. I, at most, just use kinda clear or new conversation.
03:53I'm not setting up these complex sub agents where, oh, I'm in this specific scenario. Let me spin up these four sub agents. You're just complicating things. Claude code in their main model, SONNET four five, is so freaking smart. It can do most of the things you do with sub agents on its own. And if it needs sub agents, it calls them for you automatically. You don't need to create a 100 different complicated custom sub agents to get the most out of Claude Code. I don't use a single MCP.
04:21I know there's some good MCPs out there. People tell me x is great. You want the most up to date documentation. I think SONNET four five is so smart that a lot of the time when I say, hey. Can you build out this new feature functionality for me? It's able to find the documentation by itself. I don't need MCPs for every single piece of documentation out there. For instance, the other day, I asked Claude Co. To build an app using the Kimi k two thinking API.
04:46This was a brand new model that just dropped literally that day. So I came here. I said, please build me an app with Kimi k two thinking. I hit enter, and it just went and found the documentation. It just searched the web and found the documentation in a second. I don't need to set up four MCP servers so that it has the latest documentation.
05:08The models are smart enough. If you are adding 20 MCPs and 30 sub agents and using 40 different custom slash commands, I promise you you're overcomplicating things. For 99.9%
05:22of people, Claude code just works out of the box. So stop listening to all the AI gurus. All you need to do to get the most out of Claude code is just tell it what you want to build, and it will figure it out on its own how to build things. So tip two, keep it simple. 99.9% of people don't need MCPs and a 100 sub agents and 30 different slash commands. Keep it simple. Tell Claudeco what you wanna build, and it will build it for you. Just as a side note, before we get into tip three, make sure to leave a like if you learn anything at all. Subscribe. Turn on notifications.
05:52All I do is create amazing videos around using AI. Tip number three, one of the biggest complaints I get is that Claude code's kinda pricey and credits go by quickly. Well, here's a tip for you. I use this, and this gets me, like, 10 x more usage out of Claude code. You wanna go into your model picker. You wanna select models. You do slash model, and you wanna choose Haiku.
06:13Haiku is the cheapest, quickest model for Claude code. But here's the thing. Here's how it works when you select Haiku. When you select haiku, it actually still uses SONNET four five for all the planning. Right? So when you're in plan mode, which is shift tab shift tab, when you make the plan, it actually uses SONNET to build the plan, but then it uses Haiku to do the execution. So if I build out a feature here, let's just say add a dark mode to the app and hit enter, it's gonna use Sonnet to build the plan for this feature.
06:46But when I hit yes on the execution, it's gonna use Haiku for the implementation. And as long as you have a nice thoroughly detailed plan, which is what SONNET four five does, you don't need the most advanced models to do the implementation. Haiku can handle it really, really well. And by using it this way, Sonnet for the planning, Haiku for the implementation,
07:08you're going to be so efficient when it comes to using credits and tokens. So you wanna make sure if you're on the $17 plan or even the $100 a month plan, make sure you do model and make sure you select haiku. You're gonna get Sonnet for planning, haiku for implementation, and that will be just as good as using the default Sonnet for everything, I promise you. Tip number four, and this might be the most important out of all of them, to be quite honest with you, and that is make the most of your time when the AI is working. I saw this tweet from Y Combinator yesterday, and it literally made me sick to my stomach. They're releasing an IDE called chat IDE,
07:45HOOKwhere while the AI works, it brings up, like, video games and gambling in Plinko, uh, as the AI work. And it shows these guys just playing video games every time the AI works. This is quite literally, a, the worst product ever made in history, but, b, the stupidest possible thing you could be doing while your AI works. 80% of the time while you're vibe coding, you're in a mode like this where the AI is building a plan or it's implementing code. And if you use this 80% of the time to do what vast majority of people do, which is play video games, get dopamine hits, scroll TikTok, watch Netflix, look at Instagram reels, look at thirst traps on Instagram.
08:21HOOKYou are doing yourself and your app such a big disservice. Here is what I do when I have my AI working for me while it's vibe coding, while it's building code, whatever it is, I have the desktop app for ChatGPT open. You can have Claude, g b GPT, Gemini, whatever your favorite model is just to chat with. Have the desktop app open, And what I do is I tell it I want it to be my Copilot CEO product manager for my app. And anytime the AI is working, I'm literally talking to the AI in my desktop app and saying, hey. This is the feature I'm working on. What do you think? What should we add next? What should the road map look like? What should my first tweet be like? I'll even just ask for life advice. I'll be like, hey. I'm thinking about picking up meditation.
09:07What's the best way to do that? And now this makes that 80% of the time, the AI is working inside Claude code so much more productive. I'm coming up with new ideas for features. I'm coming up with improvements for the app. I'm coming up with marketing for the app. I'm improving my life. So instead of scrolling TikTok and getting dopamine hits or playing Plinko or gambling on the football game that night, you are using ChatGPT
09:33or whatever AI chat model you wanna use to plan and improve and then get your app in a better place. If you do this and you spend 80% of time being productive doing this, I promise you, you are going to be you're gonna be so much more productive. You'll be so much more mentally healthy. If you spend 80% of your day scrolling TikTok, your brain's melting out of your ears. You're gonna be so much more mentally healthy doing this. You're gonna get so much more done, and you're gonna be ahead of the 99% of your competition
10:02that just waste their day doom scrolling while their AI works. If you want to win, you need to take this tip. Do not let yourself doom scroll while the AI works. Instead, open up a Copilot CEO product manager chat and take advantage of it every single time. Here is my next tip for you. You need to have a good Claude rules file. So you need to have a good Claude dot m d file. These are the rules that get sent to Claude with every single prompt you make. And if you have the right rule set, you're gonna get so much better performance out of Claude code. Over my hundreds and hundreds of hours of using Claude code over the last year, these are the nine rules I use. These are the nine rules I've put in where I get the absolute best performance out of Claude where I really don't get any more issues or errors or bugs anymore. I don't get any of those nasty bug circles where you're going around and around in circles. Here are the nine rules. I will paste them down below so you can just copy and paste them and put them into your own Claude rules file. Just as a quick overview, uh, number one, I make sure it slows down,
11:03thinks through the entire problem, and reads the code base for relevant files, and builds out an entire to do list for itself anytime I give it a prompt. Sometimes these models are just lazy. I'm gonna be honest. Sometimes all the models are just very, very lazy. This first rule prevents it from getting lazy. It makes a detailed to do list based on all relevant files that has to follow step by step. I make it stop and give me explanations before every single little change it makes. One, this is just good for me because I like to understand how the entire app works. But two, again, this is helping the model because it's forcing it to slow down instead of being lazy and speeding through problems. It forces to slow down and explain issues to me, which actually makes it smarter itself. I say make every task and code change you make as simple as humanly possible. I have too many times have given Claude a task, and then it goes and it makes a change that impacts 20 other features that I then have to play whack a mole with to fix all the other features.
11:58This rule makes it so it keeps it simple, and it only touches the code and files it needs to touch to fix any issues. Then I reiterate a lot of the important points and rules eight and nine here. Do not be lazy. Never be lazy. If your context window starts filling up, it starts getting lazy. It starts hallucinating. This prevents that from happening. No temporary fixes. I see this a lot with every AI
12:19where I'll say, here's an error, and it'll do, the most temporary fix humanly possible. This makes it so it only does long term fixes that address the root cause. And, it stresses simplicity, simplicity, simplicity, so it doesn't overthink it and write way too much code. These nine rules have improved the performance of Claude by 10 x. It really is amazing. So make sure copy paste this, create a new Claude dot m d file in your project, paste them in, and you're good to go. You'll get way better performance. My next tip fits in with the general theme I think of this video, which is keep it as simple as humanly possible.
12:52This tip is prompt engineering is the most overrated concept in history. There are 28,000 videos on YouTube telling you how to prompt Claude the right way and get the most out of the prompting. Throw it all away. It doesn't matter. Prompt engineering is so overrated. Again, the models in the app have gotten so good and so smart. You don't need to do these complex
13:14structures of prompts. Right? I see this all the time where they'll be like, okay. In order to get the most out of Claude code, you need to first have a context section where you explain everything inside your app. And then your next part is you go a UI section. In the next part, you do a UI section where you explain in detail what the you want the UI to be. And then the next part, you have a database section. You explain clearly what the scheme and it gives you this crazy
13:41long prompt engineering technique. You have to do it. No. No. No. No. No. I think it's actually the opposite. You should keep these prompts as simple as humanly possible, Claude. Just tell it exactly what you want to do. For example, I'm building out this AI journaling app. Right? Say I want to add a folder structure to the app where I wanna be able to add different journal entries into different folders. A lot of the gurus on YouTube and on x will tell you, oh, you gotta use prompt engineering here. Make sure you include all the context in these 40 different structures and spin up the sub no. Just tell it what you want. I want a folder
14:16system in the app so I can put entries into folders. That's it. Hit enter. You're good to go. And as long as you have it in plan mode, so shift tab twice,
14:28it's going to figure out the context for you. It's going to look at the right files for you. You don't need these advanced prompt engineering techniques. Every time I see someone on Twitter say, oh, here's how to do prompt engineering, I laugh in their face, because I know they're just trying to get a little bit of engagement. There is no need anymore for prompt engineering. Prompt engineering's extinct. These models are good enough. Just tell it like it is. It will build it for you. Keep it simple, stupid. The next tip I have for you is use screenshots
14:58as much as possible, especially for UI. My biggest issue with Claude Code and really every AI out there is the UI is almost always the same. It's always bad. It's a miracle. You don't see any blue and purple gradients in this app that it built out. The reason why is this. I try to include screenshots with any UI change I make. Right? I use screenshots of inspiration
15:21for the UI. How do I do that? One of my favorite ways to do that is through v zero. So v zero is another AI coding tool. They have a design system section here. If you click on design system, they have a whole bunch of designs in here you can click on and customize. If you just choose any of these and then you bring it over and you screenshot it, you can just screenshot it. You use command control shift four on Mac that takes a screenshot and copies it to your clipboard so that you can easily just go in a clawed code and paste it in here. And so if you're building out a UI, you're building out a v one of your app, you always want inspiration.
15:58This v zero feature makes it really easy to get inspiration. Include the screenshots. Other things I like to do is I found some inspiration on Pinterest before. So Pinterest, which I thought only chicks use, apparently not. I go on there now, and I search for, like, app designs, and I find really good inspiration for app designs. So always make sure you have some sort inspiration or screenshot of UI whenever you're building out UI inside of Claude Code because you get way better UI instead of that annoying, gross, blue and purple gradient you always get. In the last tip I learned about Claude code using it almost twelve hours a day every day for the last nine or ten months is
16:38treat Claude like a creative partner. Not many people I talk to ask a lot of questions to Claude. I am constantly asking questions to Claude. These AIs are much at least much smarter than me. They may not be smarter than you. You're pretty smart. But for me, they're a lot smarter than me. And so I am constantly asking questions and treating it as an equal, treating it as a creative partner. So I'll do a lot of things like this. I'm not even in plan mode yet. I'm in regular edit mode. I'm saying I want a folder system in the app so I can put entries into folders.
17:10And I'll ask things like, what are different options for designing this? What are different cool features
17:19we could add? And I'll hit enter on that, and I'll send it. And now I'm just bouncing ideas off of Claude. Right? Before I do any coding, before I do any sort of planning, I'm just saying, hey, Claude. What are you what's your opinion on this? What would you add? What would you change? And I do this this is like almost a preplan mode. This is like a brainstorm mode. I wish they would put in this a slash brainstorm mode where we can just bounce ideas off each other. I mean, I'm just doing it now, but I bounce ideas. I kind of figure out what I want this to be,
17:50and then I go into plan mode after I say, perfect. Make a plan for building that out. I love that idea. So as you can see here, it came up with a whole bunch of ideas. There's kind of the flat categories folder system, which is what I was thinking about. But then there's also a tag base. I didn't think of that. Using hashtags to tag the journal entries,
18:09I think that's brilliant. And then you can search for the hashtags in there. I didn't think of that. That's great. Now we just came up with an even better way to implement that feature. Uh, and then there's a hybrid approach. Best of both. I like that. You have both folders and hashtags. Now I can switch to plan mode with shift tab, and now I could say, perfect. Implement number four, please. Right? It gave me a menu item of ways to implement it. I choose the item I want to say, perfect. Now we're in plan mode. Let's plan that out. You should be treating Claude as your creative partner. It's not just some junior dev going out writing code for you. It's your partner.
18:43CTAEvery app I've built, including Creator Buddy, which is now making over 300 k a year, I built completely off just bouncing ideas back and forth with Claude, and he gave me such brilliant, brilliant ideas. So to recap everything we went over, it basically comes down to this. It basically comes down to a, keeping it as simple as humanly possible. You don't need to use every frigging feature on planet Earth with Claude code. Keep it so simple. Number two, don't waste your time. 80% of time using Claude code is downtime. Don't waste that downtime. Talk to the AI. Get the most out of it. And three, the AI is smarter than you think. Bounce ideas off of it. Trust it. Use it as your partner, and you'll get the most out of Claude Code. If you do the eight things I talked about in this video, you will get so much more out of it. Feel free to go back now. Write down notes, whatever you wanna do. You'll get so much out of this. Claude code is brilliant. It's amazing. It's my favorite tool to use on a daily basis. Make sure to leave a like if you learned anything at all. Subscribe if you haven't yet. Turn on notifications. I livestream, like, three times a week so you get alerted on that. And sign up for the number one AI newsletter on planet Earth. It's completely free. Link down below. Alexfinn.ai/subscribe.
19:50CTAWe'll get you on it. Hope you learned a ton in this video. Love teaching you guys new things, and I'll see you in the next video.
§ · For Joe

The nine rules that end lazy loops.

JoeFlow / Sessions playbook

Alex's CLAUDE.md is the most copy-paste-ready artifact in the video — and the Brainstorm Mode → Plan Mode sequence is the single most underused pattern in Claude Code.

  • Drop Alex's nine rules into your project CLAUDE.md verbatim. Rules 8-9 in ALL CAPS — that's not style, it's a behavior change.
  • Before entering plan mode on any JoeFlow Sessions run, spend 2 minutes in brainstorm mode. Ask Claude what it would change. The hashtag idea came from Claude, not Alex.
  • Switch to /model Haiku for any session burning credits fast. Sonnet plans + Haiku executes = same output at a fraction of the cost.
  • The 'copilot CEO during idle time' framing maps directly to the JoeFlow Sessions cockpit — the gap between agent runs is exactly where this applies.
  • Drag a v0 design-system screenshot into any UI prompt. Stop accepting blue-purple gradient defaults.
  • The meta-lesson: Alex's channel wins by talking back to over-complication. Every new wave of tooling (MCPs, sub-agents, slash commands) creates a new cohort of beginners to address. That's a repeatable content formula worth stealing.
§ · For You

How to actually get more done with AI tools.

If you're using Claude Code or any AI coder

The single biggest productivity unlock isn't a new feature — it's changing how you use the 80% of time you spend waiting for the AI to finish.

  • While your AI codes, open a separate chat (ChatGPT, Claude, Gemini — your pick) and treat it as a thinking partner: what should you build next, what's the marketing angle, what life decision are you circling.
  • Don't start with a prompt — start with a brainstorm. Ask the AI what options exist before telling it what to build. You'll get ideas you wouldn't have had.
  • A tight rules file (CLAUDE.md) isn't bureaucracy — it's the difference between an AI that loops forever and one that delivers clean, minimal changes.
  • Screenshots are the fastest way to get good UI. Grab a screenshot from any app you like the look of and include it alongside your prompt.
§ · Frame Gallery

Visual moments.

§ · Watch next

More from this channel + related dossiers.