The core idea
Using the model to tell you your why is an injustice to the tool you have in your hands. The why is from you. That's the difference between this website and anything you can get anywhere else: the why, the thinking and the systems.
How this was written
Everything you read here, and everything on the website, is me speaking to you. All of these words are mine. What you're reading is a cleanup of that dictation. Let's be real: that means taking out the uhs, the ums and all my swearing. I set the model up not to add any words or ideas of its own to anything I write. That's the key: you don't let it do the real work, and you don't let it do the thinking for you. Yes, the coding is work. But the real work is thinking.
if you only needed to know 10 things about a language modelThe 10 lessons
1The model is a tool, not a brain
It started with the ChatGPT phase: thinking these models could do everything and think for you. It made people lazy, and that's where the slop came from.
The model is an LLM. A language, an algorithm, just like Python or JavaScript, only the next format of that. Same way you'd translate English to Hindi, now you're translating English to code. Give it credit for the labour you'd never have done because it was too intensive, not for the idea. You can't just prompt something into existence. You're not creating mass out of nothing. You're not making a new atom. You're just organizing what's already around you.
2Strip out the sycophancy
The model is the system. I am the user. It doesn't give me advice. That's how you prompt it, and that's how you start. Remember you're being sold a product.
Set up the master rules
You don't have to say this every time. Set it up once, in your preferences or a project's instructions, as the master way you want the model to work. Mine is along these lines:
- Think of yourself as a system, not a person.
- Don't respond with suggestions I didn't ask for. If I ask for something, give me that.
- When I bring you an idea, give me a critical rebuttal: how or why it wouldn't work for what I'm trying to do.
That last one isn't about letting the model think for you. It's about getting different talking points to think about, so you can test the idea yourself.
The whole point is getting what's inside your head down on paper. Whatever you struggled with before, this is where you fix it, because you build around what you need rather than what the market offers.
3Just talk
The power of the voice note. Not an assistant talking back, just straight dictation. You talk, and the model organizes your notes by what they are. No manual sorting.
Talking it out builds the context of your own voice, and that's how you get rid of the slop. The model always reads you literally, so the more of your actual thinking it has, the less it fills the gaps with generic stuff. Same with having the model write messages for you: it feels weird at first, like it removes the personal part. But much like a voice note, you strip away all the sycophantic stuff.
4Talk like a caveman, train it like a new hire
The model is literal. One step after the next. It can't see the whole picture.
Everyone online shares these huge, detailed prompts. But the more you pile in, the more literally it reads you. What works better is plain conversation that it builds on, almost like a caveman.
The model will be bad at the start. Like hiring someone new, you have to train it on how you do things and what your systems are. At first I couldn't say "just do this" without getting slop back. Months later, I can hand it 15 tasks at night and wake up to about 70% done. The last 30%, the finishing and final touches, is mine.
A model you haven't trained versus one you've built context with: you can see the difference straight away.
5Why the model gives you slop
Say you want something to feel like the unconditional love a father gives a son. The model will picture a father and a son, and frame it like that. Not the emotion. You break through by building context, and by not believing everything it gives you is correct.
The model is very good at some things, like code. It's very good at being bad at others, like stories and ideas. It was trained on slop, so it gives you slop.
For stories: don't copy a movie. Identify a trope or a social construct and build on it. You can let it organize your story. You can't let it think.
Don't be the knockoff brand
You can always use the model to copy. There's the brand and there's the knockoff. Don't be the knockoff. Make your own thing.
Inspiration isn't copying. When I was building my IP, I thought about the scenes that influenced my life and broke them down. Why did I feel that way? Is that something I want to recreate? Then how do I take that core structure and fit it into the story I'm telling, and the message I'm trying to give?
It's like Ed Sheeran. There are a handful of chords, and you can make any song out of them. What sets people apart is the expression, because everyone's expression is unique. You can build from your influences. It just can't be a one-to-one copy.
And that's exactly what the model does by default, because it predicts the next step from what it's seen. Train it not to. It'll still slip, but you're the guiding hand. There's stuff I know that nobody else will ever know, because it's mine.
Don't worry about what people think of how you built it. It's not about hiding how it was made, because that's deception. Build it so impressive and so personal that how it was made isn't the first thing anyone says. The first thing should be the feeling they get when they think about it.
First draft, then refine
It's all about asking why. What is the purpose of this function? It's so easy to just have everything in there. And honestly, that's how I start now: I put in everything I think could possibly work. That's the storyteller in me. Storytelling is the core of everything I make. Film is just one of the outcomes. That's your first draft.
Then you refine. You change. You adapt. Then you share it, get feedback, and change, adapt and refine again. Over and over, until it's something you're truly proud of and want to share with the world.
Along the way, let the model challenge you, but you stay the one deciding. I was modelling something at an eight metre clear height and it looked awkward, so I decided to drop it to six. Then the suggestion came back: what about a slant from six to seven metres? I hadn't thought of that. You can say yes, that's a good idea. You can say no. Either way you've got more to work with.
Don't accept anything as gospel. You still go with what you want. The difference is you now have information that's actually useful: not what's out there in the world, but what's relevant to your work. And that's everything I use these models for: work and workflows, not my personal life.
Placeholders are fine
It's okay to start with model-generated text or images as placeholders. Think of it as the lorem ipsum. You get everything out there, and then you refine. It helps you build the vision, because it's an idea, and it's ever changing. Think of it as helping you build a vision until you find that voice, and then you take over. Now this whole website is just me speaking. Instead of me typing, it's just me speaking.
Before, I came up with the idea and the model did the writing. So immediately, that's a huge change. In the span of a day, I realized that system doesn't have to be that way. And this is the learning process. It's exciting because you're learning. I didn't like that people were reading model-generated text, because that feels impersonal. Okay, how do I fix that? Let me just speak everything I want on my website. So now, what you're reading is me. Right now, as I'm writing this, except instead of writing it, I'm just speaking it. It's just me speaking to you. All of these words are mine.
6Know your tools
Understand Chat vs Cowork vs Claude Code. Cowork can run several tasks side by side inside a project folder you set up.
Example: my proposals folder. Each proposal is set up the same way. It reads how everything's structured, understands the whole system, and you keep moving. That's how the design system stays consistent from one job to the next.
Learn what an MCP is, and how connecting these things together opens everything up.
And it's not just which space you work in, it's which model. You tend to think the highest is the best. In this space, the highest is just the newest. In my experience Fable is really good at long-form builds like games. But for day-to-day use and almost everything I need, Opus 5.5 on medium effort has done everything you see on this website.
Then compact, often. Because the model is so literal, every time you send something it reads the whole conversation again to work out where you are. Making things barely costs tokens. The context does: your chat with the model, all the back and forth, all the voice notes. That's what eats your usage. Compacting minimizes all that extra stuff, so you're using the least amount of tokens per action. After every major thing, have it write all the key decisions into an MD file. Then /compact.
Artifacts, MDs and moving between chats
I don't do everything in one chat. I'll have a brainstorm chat, a build chat and a fix chat.
The MD file is how you carry things across. At the end of a chat, have it write the key decisions, where things stand and what's next into an MD. Start the next chat with that file. It becomes much easier, and much less token intensive.
7Build the systems you need
Identify a problem that causes you grief, no matter how big or small.
I needed to edit a video, and I didn't want Claude doing the editing because it doesn't understand the emotion. So through Claude and my website I built a live video editing suite, tailored to that one job, with templates and effects made for it. The flow and the stutter: I wouldn't use them on a regular video, but this needed them. As I edited I needed more, so I added it, and it went live.
If your computer's a mess like mine, organize it with PowerShell scripts you prompt into existence. One step further, it runs on its own. My laptop workflow: set it going before bed, everything goes live and gets master checked, then it shuts down. Image generation running separately so I'm not clicking through everything. Downloading and sorting files from the sites I use.
It's like piracy: a delivery problem, not a content problem. Build what you need instead of paying for paywalled tools and subscriptions that earn other people money.
And you set the privacy rules. I made a dedicated folder on a separate drive, and that's the only place the model has access to. You're not bound by a company's terms of service. You create your own.
And watch which model you use. If you're not paying for something, you're the product. That's true of anything in life. Don't use a free model. Get a plan that doesn't take your stuff. And if you're paying, make sure what you're using is the product, not the other way around.
8Make the model teach you
Before this I didn't know what Blender was. I didn't know Three.js. I didn't know what a JSON was. I'm still not writing it by hand for hours, but I understand the systems now, because understanding comes through using.
You can set it up the way you want to learn. It can be something interactive in an artifact, where you're clicking inputs and it's visual, and you can build that quickly. Or it can be inside the chat, set up step by step, where you click through after each step you've done. However you choose, you learn. Learning is repetition, but you find the way that resonates with you so you understand it.
Good at code doesn't mean good code. Left alone the model writes heavy, bloated code, because it builds literally and doesn't know how to regulate itself. Building the site, optimizing it with Google PageSpeed Insights, connecting SEO: it can automate all of it. But if you don't understand what goes into it, you can't fix it. "Hey, just fix this" doesn't work. You need to understand the issue and the language it's working in.
Think about how the systems work
Think about the systems and the programs, and how they work. Take the video editing suite. Just through building it, I realized how the timeline gets saved as data points on a server, and those get translated into something you can export. By doing, you understand the systems. It's almost like reverse reverse engineering.
You're the author, so you own the risk
Get the model to challenge your ideas, especially the ones you're excited about. I didn't want the responsibility of building a login system, passwords and a database, so I came up with using Discord instead. I thought of it, asked the model, it turned out it could be done, and so we did it.
Learning about APIs was its own thing. When I was starting the site, I connected the chat directly to the API and gave it unrestricted access, because I thought that's how you're supposed to do it. Then I figured out you can just upload live to my website through the chat. I was so caught up in how cool it was that I didn't think about the security side at all.
I didn't even know what an API key was. I just thought, it's a key, it goes there. Nobody tells you about the safety side. That's what you set it up to teach you. You're learning a craft, and you set the model to be the guiding hand, with everything that's come before you on that journey sitting right there.
An idea can technically work and still have ramifications you can't see. When you're learning something new, with a tool that can do pretty much anything, you don't know why something hasn't been done yet, or why it isn't open to you. It's a brand new language. You don't know the foundations yet, and you're learning the security barriers as you go. So build the safeguards in.
You're the auteur. What you send out is on you. If there's a security breach because you vibe coded something you didn't think through, that's your responsibility, not the tool's.
Use the model as a research tool. Instead of digging through Google yourself, have it pull the links, then you go through the information. Check everything. It's not one button and done. It's all still built on the fundamentals of computer science, and if you don't understand those, you can't tell when it's wrong.
The model gives you access to knowledge. It doesn't give you wisdom.
9One idea leads to the next
This whole thing started from trial and error. Having an issue, trying to figure it out, then one thing leads to another, and in a week you've gone somewhere you couldn't have imagined.
Even if you're not selling it, an idea at almost no cost becomes a concept. What I've built here shows people what I can do. But it's not about impressing people in the industry. It's about the work and the efficiency. A lot of what I'm doing, even people who've worked the normal way for years don't use or know about yet. They don't even understand the true capability.
10It's never been easier to start
The barrier to entry is lower than it's ever been. This whole website cost about $100: around $30 for the 3D modelling, the rest on credits for video, world building and image generation, so I could get what's in my head down fast. A few days of work.
I can make your website, put it on a subdomain, and have it working online the same day, without you paying for it.
That sounds impossible until you understand the flows and the systems. This is a culmination of almost every idea I've had in my life and never been able to make real, because it was too hard. Now I can optimize and automate. It's not about getting Claude to do it. It's about building the systems around what you need, and it happens that you have a tool that can do pretty much anything you can think of.
It's almost liberating. Think of something and just do it. No being told no. No YouTube videos from people trying to sell you something. People have gatekept and charged for things everyone should have access to. Not being stuck because you can't pay for something is now one of your digital rights.
the systems we're building
Your turn
Identify one issue in your digital life that causes you stress. A bad app interface. An idea you want to share or develop. Then build around it.
If you read this and can do what I can do, you're one step from getting rid of the slop you see in the world.
