Where Do I Start With AI in My Business? The First Move Most Owners Skip

TL;DR: Most business owners start with AI by picking a tool or automating a task — and that's why it underwhelms them. The real first move is building an AI that actually knows you and your business before you point it at anything: fill out your account preferences, then create one dedicated project with memory, instructions, and a few key files. Do that, and the same tool that felt generic starts working like it actually knows what it's doing. Tools and agents come after the foundation, not before.

Why AI probably feels underwhelming right now

By now you've almost certainly used ChatGPT, Claude, or Gemini. And if you're being honest, you've probably been a little underwhelmed. It makes mistakes. You repeat yourself constantly. Sometimes it feels like it's taking more time than it saves.

Here's the thing: most people aren't tapping into even 10% of what these tools can actually do. Not because the tool is weak — because of how they're using it. And that gap is the whole reason people freeze. In fact, 96% of small business owners plan to adopt emerging technologies including AI (U.S. Chamber of Commerce, 2025) — yet most are stuck at the starting line. More than half still describe themselves as "explorers," and only a small fraction have gone deep enough to see real results.

The want is nearly universal. The starting point is what's missing. So let me give you the honest one — and it's not what the internet tells you.

Where should I actually start with AI?

The advice everywhere is "pick your biggest time-drain and automate it." That's not wrong, exactly — but it's not first. If you point AI at a task before it knows anything about you, you get generic output, because it's guessing. You end up with a tool doing work in nobody's voice, for a business it doesn't understand.

The real first move is this: stop using AI in a cold thread.

A cold thread is when you just open AI and start typing into any old blank conversation. That's fine when you're learning the ropes — everyone starts there, no shame in it. But the moment you want real leverage, the move is to build the tool to know who you are first, and then build dedicated spaces for the specific jobs you need it to do.

Think of it like making a hire. You're not opening a chat window — you're hiring a thought partner. And everything that follows is about onboarding that hire properly. I'll even give you a copy-paste prompt below to make the first step easy.

Step 1: Fill out your account settings and preferences

Almost nobody knows this exists. In Claude (and there's an equivalent in ChatGPT and Gemini), go into your account settings and preferences, and tell it everything you'd want a great assistant to know about you: how you work, your personality, what you're good at, what you need help with. If you've got ADHD, tell it. As much as you're willing to share.

It's like the bio on your social media account. Nobody knows who you are if you never fill out your bio. Fill this out, and every single response — even in a blank thread — gets filtered through that context.

Here's my favorite tell: people say "I hate that AI just tells me I'm great all the time — I want it to be honest." The second someone says that, I know they never set up their preferences. Mine is ruthless with me, because I trained it to be.

If you've seen Interstellar, remember how Matthew McConaughey's character constantly tweaks TARS, the AI robot — dialing his humor down from 95% to 60%, turning his honesty up? That's the mental model. When AI isn't responding the way you want, that's usually a settings problem, not a limitation of the tool. They shipped you a box; most people use it straight out of the box and never customize it. This is your chance to customize it from the start.

Not sure what to write? Paste this into your settings and fill in the brackets:

Here's what I'd like you to know about me so you can help me well:

- What I do: [your business and your role]
- How I work: [e.g. I think out loud, I move fast, I get overwhelmed by too many options]
- What I want help with: [where you're short on time or stuck]
- How I want you to talk to me: [e.g. Be direct and honest — don't just agree with me;
  challenge my thinking, and skip the long preambles]

From now on, filter your responses through this context.

That's it. Every response from that point on runs through it.

Step 2: Build a dedicated project (not one endless chat)

Once it knows you, give it a dedicated home for each job you need done. In Claude these are Projects; ChatGPT and Gemini have their own versions.

Think of it like Google Drive. When you throw every document into one pile, it's chaos. Instead you have folders — personal finances here, company files there, marketing over here — a structure that reflects how your business actually works. A project area does the same thing for your AI: "now that you know me, I'm going to build you into exactly what I need for this specific job."

I have one project for my main company. Others for side projects. And one that does nothing but hardcore auditing — I've trained it specifically to pressure-test everything I'm about to launch. (Name them whatever helps you remember. I'm a sci-fi nerd, so mine get pop-culture names. Have fun with it.)

What goes inside a project? Memory, instructions, and files

This is where the magic actually happens. Each project has three layers.

1. Memory — what the AI needs to know about you and your business for this specific job. Not just you the human (that was your account preferences) — a layer deeper. If it's a business project: your company details, who's on the team and their roles, how long you've been in business, who you serve, what a customer looks like, your revenue milestones. So when it talks to you, it already understands your world. You can guide what gets stored, add to it, and change it over time (I update mine whenever something shifts internally). Any time something important happens in a chat, you can just say "save that to memory."

2. Instructions — this is the job description. If you hired someone and never told them their role or what you needed them to do, you wouldn't expect good work. Same here. This is where you tell it who it needs to be for you and what it needs to do. This is the step that makes people's jaws drop — because they realize how much they were repeating, or losing, by skipping it.

3. Files — deeper context and how you like things done. But be careful here, because this is where people shoot themselves in the foot.

What files should I add — and what's a mistake?

The classic mistake: treating your project like Google Drive and dumping everything into it — your entire book, giant files, your whole business history. Don't. Every time you start or continue a conversation, the tool re-reads all of it. That means your token burn goes through the roof, and you risk "context chaos" — feeding it so much that pieces start to contradict each other or bury what matters. It's like onboarding a new hire by spending all day dumping your family lineage and every past employee on them. They don't need that to do the job. Would you do that to a human you hired? No. Same rule here.

What does belong (these are the ones I use):

• A voice/tone doc — samples of how you actually sound, your phrases, your cadence, so it writes like you instead of guessing.
• An ideal-customer profile — who they are, what they care about, their objections, what they say when they buy.
• Your goals / plan for the year — so it's not aimlessly generating; it knows what actually moves the needle.
• Your offers and product suite — what you sell and the purpose each one serves.
• Your tool stack — what you use and how.

The test for anything: does this project need this file on a daily or weekly basis? If yes, it belongs. If it's just "nice to have someday," put it in Google Drive instead. (When I finished building my community app, I archived all those working docs straight out of the project and into Drive.)

Do I need custom software, or is ChatGPT enough?

For this foundation? A general tool is plenty. General-purpose AI is exactly that — general — which is why the foundation matters: the account preferences, the project, the memory and instructions and files are what turn a generic tool into one that knows your business. You don't need a developer or custom software to start. You need to onboard the tool you already have.

(For perspective on what becomes possible once you're operating this way: I recently built a full custom community app for our clients with AI in under 14 days — software engineer friends told me that would've taken a full dev team six months. That's not the starting point. That's what the starting point leads to.)

What about connecting my tools and using AI agents?

That's real, and it's powerful — connecting your AI to your other tools (via what's called MCP) is like plugging a USB-C cord straight into the systems you use, so it can not just see your information but actually do things: draft the email, update the record, run the task. And agents that run work for you are the layer on top of that.

But hear me clearly: that's Phase 2. Connecting tools and deploying agents before you've built the foundation is how you end up with automation you can't steer. Build the core first. The rest gets dramatically easier once you have.

What do I do this week? (Without the overwhelm)

The reason AI feels overwhelming is that people try to change everything at once — seventeen tools, thirty automations — then quit two weeks later, more confused than when they started. Don't do that. Do this, in order:

1. Fill out your account preferences.
2. Create one project for your most important job.
3. Give it memory (about you + your business), instructions (its role), and a few key files.

That's it. One foundation, built properly, beats ten tools bolted on badly. You've already got the starter prompt above — use it today, and you'll get more out of your AI this week than you have all year.