Top 5 AI Fundamentals

AI is disrupting jobs, companies, industries, markets, and economies worldwide, in large part because of how easy it has become to go from idea to reality. Barriers that gated rapid innovation for years are beginning to fall. It has never been easier to build new products, share solutions, and stand up new business models, and access to AI reaches new highs nearly every day. Yet so far, a small few have captured most of the gains.

This shouldn't be the case, and it doesn't have to be. AI is unique as a disruptor. It doesn't just displace the old way of doing things; it's also a skill you can learn, and more importantly, a technology that can teach you the skills you need to take advantage of it.

If you leave with one takeaway from this guide it should be that you never needed this guide. Which brings us to the first and arguably most important fundamental:

1. Ask Outcome-Oriented Questions

AI is a useful instructor and guide, not just a building tool. Leveraging AI to learn how to better use AI is one of the foundational skills for making the most of it.

This requires you to be aware of what you don’t know and to have the confidence to ask the questions. You do not have to be a computer science major or career software engineer to become AI native. In fact, one of the best parts of AI is that you need not know the precise methods, patterns, architectures, acronyms, (you name it), to be using. So it’s important you don’t let that be a barrier to you experimenting with AI.

The key to making the most of AI’s instructive power is to ask outcome oriented questions - that is, to ask about how to achieve your particular goal or need. Ask AI for its recommendations and a walkthrough for how to achieve your outcome, and don’t get lost in feeling like you need to know the methodology. Push and prod further on the AI’s responses when you get answers that still don’t make sense or go over your head. Be aware that there’s stuff you don’t know, and constantly ask to learn it.

Good and Bad Examples of Outcome-Oriented Questioning:

Good = How can I build a website for my gumball business, that my friends and family can access at this URL I want (CustomGumballz.com) where they can sign in to an account, design a custom gumball using my color palette, and pay to place an order from me?

Bad = Can you build a website for custom gumballs?

2. Planning and Project Organization

Shape and flesh out your idea deeply before you run into building it.

AI can build fast. This is a double edged sword. Just as well as you can turn a great idea into reality, you can quickly get lost wasting both time and tokens building a poor or distorted version of your idea. That’s why it's incredibly important to get organized from the start and plan thoroughly. Again, this is a responsibility that both you and AI can share.

Start by making a folder on your computer to work out of, name it accordingly, and point it as the project file you want to be working in. This helps keep your work in one place and keep your chat threads organized, whether you’re in Claude Code, Codex, etc.

Most importantly, use /plan mode to help you flesh out your idea before you prompt AI to start building. Instruct the AI to relentlessly ask you several rounds of clarifying questions before it calls the plan complete, so you can be sure that you are aligned on the vision. It also helps to ask for the AI to provide a recommended answer for its questions. This can help speed up the planning process, either discovering good ideas from the AI or simply speeding through the simple ones.

And to harken back to the first skill: you may well encounter steps or recommendations that don’t make sense to you. Just ask about it - what exactly does that mean and why does it matter for your ultimate goal?

Good and Bad Examples of Planning and Project Organization:

Good: /plan Help me make a plan to build a website for my solo custom gumball business. I want to be able to have people find me at a URL I choose, sign in to my website, design their custom gumballs using the colors I have available in my inventory, and pay me to create and ship it to them. Ask me several rounds of clarifying questions to sharpen my concept, and provide recommendations as well as explanations for why you think that’s the best choice in each step.

Bad: Build a website for my custom gumball business. I want to design custom gumballs and sell them online.

3. Understand Where Your Work Lives and Runs

Get a good understanding of where your work lives and how it runs.

Too often users will use AI to build a wonderful website that only exists locally on their device, and never goes live. Or push a site live that doesn’t have the proper storage configured to save users’ inputs or actions across sessions. Or collect sign ups in a registration form that doesn’t fire off any confirmations or save any of the information submitted.

Following steps 1 & 2 (asking outcome-oriented questions and planning and organizing thoroughly) will already help you avoid these kinds of pitfalls. But the important lesson is to build a good mental map of where things are, what’s running where, and your workflow for iterating on it.

When you start building a project with AI (e.g. creating a marketing site), you’ll want to give your code a place to live that’s not just on your computer. For this, GitHub is your friend. If you’re totally unfamiliar with it, think of it like Google Drive for your code (or even better, ask AI to explain what GitHub is and why it matters for your project goals!). While you iterate and work on your project, you can localhost to get a private preview into how your code is shaping up, and push your work to your GitHub repository as you progress. Once you’ve reached a state you’re happy to go live with, you’ll want to get your site running. For this, Vercel is your friend. This will get you a URL that you, or anyone else, can access your build from.

This is the simplest example and you will quickly outgrow it. As soon as you start thinking about a custom URL, user accounts, accepting payments, storing user inputs and data, etc. - you’ll need to expand to even more tools. But it's important to work with your AI to keep a strong mental map of your project, and to establish a blueprint for how you grow and expand.

4. Avoid Doing Everything By Hand

If you're doing it manually, ask whether the AI can just do it.

Akin to the first skill - when AI instructs you to manually complete steps for it (e.g. creating a GitHub repository, opening a new Vercel project, configuring environment variables, etc.) question everything. More often than not, if the AI has told you to do a manual step, it can probably do it itself. Sometimes it's as simple as asking "Can't you do that for me?"

Over time, AI tools will likely improve to anticipate these manual steps and avoid asking or instructing you to complete steps it could do for you (with the right authorization). But for now, you may well encounter instructions from the AI to complete steps across other tools (e.g. GitHub, Vercel, Figma, Google Drive, Supabase, etc.) that it could do for you.

To take advantage of this, be sure you’re working in your AI application’s coding workspace (Claude Code, Codex, etc.). This way, your AI is able to access your files (with permission), run commands, and operate tools. And get the CLI or MCP for a particular tool you are working across and authorized within your AI. If you want to know more about CLIs or MCPs, ask your AI! But what you need to know is that it will enable your AI to take actions in another tool on your behalf (given you’ve made an account already and can link/connect it with your credentials).

Also, avoid hand-building complex infrastructure parts for your project. Try to reach for building blocks when you can, like Stripe for payments, Clerk for SSO, Supabase for data storage, Resend for emails, etc. By asking the right outcome-oriented questions, you can get ahead of this as well!

5. Avoid AI Slop

Put more of your own content into your work.

No matter how hard you fight against it, AI models have a distinct signature - voice, style, patterns, and habits that permeate most of its outputs. This can work against you in many ways. People are increasingly turned off by AI ‘slop’ and are only getting better at detecting it. But more importantly, it can also trap you in unproductive loops - endlessly prompting over and over again with the hope that the AI will eventually, magically arrive at exactly what you wanted.

That is not an effective strategy. The best approach is to write some elements yourself, and put in as much of your own contribution as possible. This can give the AI more to work off of, produce better results, and help avoid an endless conversation where you dig deeper and deeper into a hole of ad hoc revisions.

Challenge yourself to give your AI positive instructions with examples of what ‘good’ looks like. Negative instructions (what not to do) is good too, and often easier to provide, but is not an effective way for getting the best out of your collaboration with AI. This is again a responsibility you could share with AI - instruct it to challenge you to request and prompt you for your input as you work with it, instead of defaulting to simply rewriting it every time.

You can also provide it with a guidance document, where you keep a catalog of the tones, styles, phrases, patterns, punctuation, etc. that you want it to avoid in its outputs. These approaches will help you produce more authentic outputs, of better quality, and with less time and effort. This matters especially for knowledge work, like blogs, articles, essays, and marketing materials, like product messaging, value propositions, website copy, etc.

Just Get Started

The best way to learn AI is to get started. Think about a problem you live with, a market need that you understand deeply, a customer pain point you see everyday… or, start with someone you know. What’s something they need and would make their life better? And just start building. Keep these 5 principles in mind as you get started and be ready to learn.