How do you use Claude for business? Beginner to expert in 5 levels (2026)

The short version

Start on the Pro plan and climb five levels: learn where everything lives, build a Project that holds your instructions and files, get real files and research back, connect Gmail and Google Calendar, then hand Claude a job that runs on a schedule. Most people I see stop at level 1, asking it questions. If you only set up one thing, make it a Project. Mine took about ten minutes to set up, once, and every chat in it starts from my instructions and my files.

There are five levels to using Claude for business, and most people I see stop at the first one. At level 1 you're asking it questions. By level 5 it's doing the work while you watch.

I run BigWork AI here in Los Angeles and I build AI systems for small businesses, so I'm in Claude all day. This is the written version of the walkthrough above: which plan to buy, what to set up at each level, and the real numbers from the demos. Plan names, prices and limits were checked against Anthropic's own pages on November 9, 2026, and each one links to its source.

Free guide: Claude 101 for business owners

Eleven pages, one idea per page: how to brief Claude, make it know you, set up Projects and skills, connect your email and calendar, and hand off whole tasks safely. Every prompt is ready to copy. Free, no sign-up.

Download the PDF (1.2 MB)

Which Claude plan should a business owner buy?

My take: start on Pro. Upgrade when you hit a wall, not before.

PlanPriceWhat you getWho it fits
Free$0Sonnet, up to 5 projects, memory, web search and file creation. No Opus, Research, Claude Code, Claude in Chrome or scheduled tasks.Kicking the tires
Pro$20 a month, or $17 a month billed annuallyOpus, Research, Claude Code, Claude in Chrome and scheduled tasksMost business owners. Start here.
Max$100 a month5x Pro usageYou're in Claude every day
Ultra$200 a month10x Pro usage every week and no 5-hour limitSuper users running long jobs
TeamStandard seat: $25 a month, or $20 billed annually. Premium seat: $125 a month, or $100 billed annually.For 2 to 150 people. Central billing and admin, project sharing, and no model training on your content by default. A premium seat gets 5x the usage of a standard seat.Your team starts sharing work

Prices and features are from Anthropic's pricing page, checked November 9, 2026. Ultra's no 5-hour limit is for verified accounts.

On Free, Pro and Max, usage is limited every 5 hours and every week. My Usage page in settings shows both limits and when each one resets. If you keep hitting a wall there, that's your signal to move up a plan.

What are the five levels of using Claude?

Each level is one more thing you hand over. You don't need all five this week. You do need to know they're there.

LevelWhat it isWhat you set upWhat you get
1. The mapKnowing where everything livesNothing to build. Walk the sidebar, then check Usage and Privacy in settings.You stop hunting for things, and you've decided whether your chats can be used for training.
2. ProjectsA workspace that does one job, with its own instructions, files, memory and chatsInstructions, your reference files and one scheduled task. Mine took about ten minutes.Every chat starts from your rules and your files, so you stop re-explaining your business.
3. Real files and researchClaude hands back files you can open, edit and send, and runs longer research with sourcesAsk for the file you want, or switch on Research from the plus button.An Excel sheet with live formulas. A sourced research report (mine read about 330 sources in 5 minutes).
4. Your appsConnectors plug Claude into Gmail, Google Calendar and your other tools. Skills save your know-how.Connect each app once under Customize. Save one skill.It can read your calendar and your inbox, and a skill keeps it writing the way you do.
5. Let it workA real job with real steps, run on a schedule, in your browser or on your computerOne scheduled task. Then Claude in Chrome and computer use.For me, a lead report every Monday at 7:00 a.m.
10 min
is what the Project in level 2 took me to set up, once
330
sources Claude's research went through in 5 minutes and 5 seconds
56¢
for the 4-second talking 3D clip Claude made in level 5

Level 1: where does everything live?

Before you type a single word, find where everything lives. This is my sidebar on the new Claude layout. Anthropic is rolling it out in stages, so yours may put a few of these in a different spot.

  • Search. Your old chats and everything Claude has made for you, in one place.
  • Projects. A workspace that does one job, with its own rules, files and chats. That's level 2.
  • Artifacts. The docs, decks, dashboards and designs Claude makes for you all land here.
  • Scheduled. Jobs that run on their own: hourly, daily, weekly. They're on every paid plan and they run remotely, so your laptop can be closed. A task that needs files or apps on your computer only runs on your computer.
  • Dispatch. You message Claude from your phone and it does the work on your desktop, with the desktop app open and the computer awake. Anthropic lists it as a limited beta on some Pro, Max and Ultra plans, and it isn't open to new users. If a step needs your okay and you don't answer within ten minutes, the request is denied and the task carries on without it. If you've got it, use it.
  • Claude Duet. A proactive agent that works on its own computer in the cloud and messages you when something needs you. It's a research preview with a waitlist, for Pro, Max and Ultra plans in the US. I have early access.
  • Customize. Where you plug Claude in: connectors for your apps, skills for your know-how, and plugins. That's level 4.

Then the chat box. The plus button is where you add files, switch on Research and reach your skills and connectors. Next to it you pick the model and the permission mode. Manual means Claude asks you before each step. That's a good place to start, but it gets tiring. I keep mine on auto and let Claude make most of those calls. For any job that can send, spend, delete or change something live, stay on manual until you trust it.

One setting to check on day one is Privacy. On the individual plans, whether your chats can be used to train Anthropic's models is a switch you control: Settings, then Privacy, then Help improve our AI models. If you work with sensitive or client material, turn it off. Team plans don't train on your content by default.

Level 2: how do you set up a Claude Project?

If you only set up one thing from this post, make it this one. Think of a Project as a filing cabinet with a brain. It has four parts: instructions, knowledge, memory and chats. Free accounts get up to five projects.

I built a real one on camera, the BigWork AI Proposal Builder. Here's every piece.

PartWhat I put inWhy it matters
InstructionsWho we are and who we serve. What to do with a call transcript: pull out the client's problems in their own words, write a one-page proposal, use the offer and pricing in the project files. And one rule: if something important is missing, like a budget, a timeline or who decides, ask me before you write.Every chat in the Project starts already knowing the job.
FilesFour of them: our brand guidelines, a client-facing FAQ, a past deck and our pricing.It writes from your real offer and your real prices.
A scheduled taskI called it Project Checker. Every day at 8:53 AM it looks for new call notes and drafts proposals from them. It runs on Sonnet 5.5, pulls calls from my note taker through a connector and sends me a push notification.The work starts before you ask for it.

Then I tested it. I asked for a proposal for a demo prospect, Ridgeline Family Law, and it didn't write one. It pushed back: "Who signs off on Ridgeline? Maria said 'I'd have to check,' and the proposal depends on that." That's the instruction doing its job. I told it Maria had confirmed by text, and it wrote the one-page proposal, then a finished version in our brand.

That was about ten minutes of setup, once. Now every chat in that Project starts from the same instructions and the same files, so I'm not re-explaining the business each time. Then look around your company for the next job that deserves one.

How do you make Claude answer the way you like?

A quick pit stop before level 3. Out of the box, Claude loves the details. I kept getting bogged down in blocks of text, and when I'd sifted through it all, only a couple of sentences had helped me make a better decision.

The fix is one setting. Go to Settings, then Your Claude, then Instructions. Whatever you write there applies to every conversation. Mine has six parts: who Claude is talking to, a 10-second rule, plain English, hard limits, depth on demand and decisions.

The global instructions I use (copy them)

These are the rules I read out in the video, written so you can paste them in as your own. Click the block to select all of it, copy it, and change the first section to describe you.

How to talk to me

These rules override your default behavior in every response, every project and every subject.

Who you're talking to
A busy entrepreneur with strong business sense who works across many disciplines and isn't a specialist in any of them. I make fast decisions and ask for depth when I need it.

The 10-second rule
Every response must be readable and understandable in 10 to 20 seconds. Give me:
1. Bottom line: one sentence.
2. Why it matters: two to three short bullets.
3. Next step: one clear recommended action.

Plain English, always
Zero jargon. Explain everything the way you'd explain it to a sharp business owner, never to a fellow specialist.

Hard limits
- No walls of text.
- Never dump full code, logs or configs into the reply.
- Don't over-explain your process.
- One question at a time.

Depth on demand
When I ask you to go deeper, go deeper, and we'll dig in together.

Decisions
Frame everything as a decision I can act on. If there's a judgment call, make a recommendation and say why in one line. Don't hand me more homework.

The 10-second rule does most of the work: a bottom line in one sentence, why it matters in two or three short bullets, then one clear next step. When I want more, I ask for it.

Under Instructions in the same tab is memory. Claude can remember context from your chats and carry it into new ones. If you want more privacy you can turn that off, but it limits how much Claude knows about you, so it's a judgment call. Manage shows you what it remembers, and you can edit or delete any of it. You can also import memory from another AI tool you've been using. Anthropic calls that import experimental, so check what came across.

Is an incognito chat safe for client data?

It's safer, but it's not completely safe. The ghost icon in the upper right corner opens an incognito chat. It isn't saved to your chat history or your memory, Claude won't find it later when it searches old chats, and it isn't used for training even if you have training switched on. That's all safer means here. It doesn't make it okay to share something you're not allowed to share.

Incognito doesn't mean nothing gets stored. A few limits to know before you rely on it:

  • Anthropic still retains incognito chats for 30 days. An Enterprise organization can set a longer period.
  • On Anthropic's consumer plans, a chat that gets flagged for breaking the usage policy can be kept for up to 2 years.
  • Your instructions still apply, so it still answers the way you like.
  • It doesn't work inside Projects. On the new Claude layout it also can't create files or run code.
  • Once you close it, it's gone. You can't reopen it or turn it into a normal chat.
  • On a Team or Enterprise plan, account owners can still export incognito chats.

My rule: memory on for business. Incognito for private stuff you're allowed to share. And the stuff you're not allowed to share, like client health records, Social Security numbers and legal case files, never goes in any chat window.

The same goes for what you connect. Once Gmail or your call notes are plugged in, Claude can read what's in them, so connect only what you're allowed to share. Turning training off doesn't change that.

Level 3: can Claude make real files and do research?

Yes. Claude doesn't just talk. It hands you real files you can open, edit and send, and that works on every plan.

In a demo Project for Ridgeline, I asked for an Excel pricing sheet built from the FAQ and the transcript of the last call. It came back as a working spreadsheet. It led with a first-win fix, every option was tied to the prospect's own numbers, and the totals were live formulas I could edit. From the chat I sent it to Google Drive, opened it in Google Sheets with the formulas still working, and I could have downloaded it for Excel instead.

Research is the other half of this level. Click the plus button, then Research, and a blue indicator shows it's on. It's on the paid plans, and web search has to be turned on.

My question was: who sells AI setups to small law firms in Los Angeles, and what do they charge? It went through about 330 sources in 5 minutes and 5 seconds. It came back with what other agencies charge, how my pricing compares and where the gaps in the market are.

It also surfaced something I didn't ask for: California's SB 574. That's a California law about attorneys. It was signed on September 30, 2026 and takes effect January 1, 2027, and it requires attorneys to disclose the use of generative AI to the court for documents submitted to the court. The same Holland & Knight alert says the wording leaves open how far that duty reaches, so attorneys should read the bill themselves. This isn't legal advice. If you're about to pitch AI to a law firm in California, you want to know the law exists before the call.

One tip. Run your big research jobs on your computer. Started from the phone app, research only searches the web and skips your connected apps, and it won't tell you it skipped them.

Level 4: how do you connect Gmail and your calendar?

This is where Claude starts to feel like an assistant. Connectors live under Customize, next to skills and plugins. You connect an app once, and Claude can read from it and act in it. There's a directory of connectors to browse, and if the one you need isn't listed you can add your own.

Two quick tests from the video. I asked, "Can you tell me about some of my important meetings coming up next week?" It pulled up my Google Calendar, put the meetings that mattered at the top, and then offered to write a pre-call brief for one of them. Then Gmail: "Can you go through my emails and surface the best AI newsletters that have come out this last week?" A minute later I had a short list with a line on each.

What are skills, and do you need them?

A skill is saved know-how: instructions Claude loads on its own when a task matches, once the skill is turned on. Think about your favorite meal. There are certain ingredients, a certain process and a certain timing. A skill is that recipe, written down once.

Mine is my voice. It's a SKILL.md file with the rules for how I write, and I've rebuilt it several times from podcasts, sales calls and client calls until the output sounded like me.

Anthropic has its own packages too. If you run a small business, look at Claude for Small Business. Its listing today shows 43 ready-to-run workflows and says it works with 37 partner connectors, and Anthropic says Pro covers one person and Team covers a business with more than one.

You can also record a skill. In Cowork on the Mac app, Record a skill captures your screen, your clicks, your typing and your voice while you do the task, then Claude proposes a skill for you to review. It's on Pro, Max, Ultra and Team plans.

Level 5: how do you let Claude do the work?

This is where you stop chatting and give it a real job with real steps. My prompt was one sentence: "Build my Monday lead report based upon my Gmail activity this last week, and please put it in a Google Drive folder and turn it into a Google Sheet."

It went into Gmail, searched my Drive and built the sheet: priority, lead, company and the stage each one is at. It reported back the way my instructions ask for, with a bottom line, a few bullets and the next decision. I could read it in about 30 seconds.

Then one more sentence: "Please turn this into a scheduled task that runs every Monday at 7:00 a.m." Done. It even pushed back, because I already had lead gen set up for Mondays.

One tip on scheduled tasks. Don't go overboard right away. Set one up, use it, find the value in it, and then add another on top. Sometimes they break and you have to set them up again, so start with one that matters.

Can Claude use your browser and your computer?

Claude in Chrome is a browser extension on every paid plan. It opens a side panel in Chrome, and it can read, click and type on the page. Think of a really smart employee with access to your keyboard and your browser. I told it to go into my Make account and tell me which scenarios (Make's name for an automation) I had running. It signed in, clicked through and reported back. I've sent it in before to fix a bug in one of them. It fixed the bug and relaunched it without me touching anything.

One caution before you copy that. Anthropic's safety page says a website, an email or a document can hide instructions meant to trick Claude into doing something you didn't ask for, and that the risk is not zero. Start with sites you trust, keep it off sensitive accounts, and watch what it does on anything live.

Computer use goes further. Claude clicks, types and opens apps on your own machine. It's in beta on Pro, Max and Ultra plans, in the desktop app on Mac and Windows. Team and Enterprise plans don't have it yet. You turn it on under Settings, then Capabilities. Run it in the background (the default on macOS 15 or later) so you can keep working while it does.

For the demo I had it open Blender, a 3D program I don't know how to use, and build a cartoon version of me from photos on my computer. About three minutes later there was a character on screen. One more prompt and it was waving. A third prompt sent it through fal.ai for animation and my ElevenLabs voice for the line. We got a 4-second talking clip, and it cost about 56 cents.

The character is not my favorite. But those were single-sentence prompts, and it did all of it inside real desktop software.

Before you turn on computer use

Claude sees whatever is on your screen. Anthropic's guidance is to keep computer use away from banking, healthcare and government apps, and it strongly advises against using it for financial accounts, legal documents or contracts, and medical or health information. Close anything confidential first and start with simple tasks.

What should you set up first?

So that's the ladder: the map, Projects, real files, your apps, and Claude doing the work while you watch. If you're at level 1 today, build one Project this week for the job you repeat the most. Then paste in your global instructions. Then connect one app, and set up one scheduled task.

One at a time. Each one earns the next. Now go build something.

Want it all on paper? Download the free Claude 101 guide (PDF, 11 pages) and keep it next to you while you set up.

Frequently asked questions

Is the free Claude plan enough for a business?

It's enough to test it. Free includes Sonnet, up to five projects, memory, web search and file creation. Opus, Research, Claude Code, Claude in Chrome and scheduled tasks start on Pro, which was $20 a month, or $17 a month billed annually, on Anthropic's pricing page on November 9, 2026.

Do scheduled tasks keep running when my laptop is closed?

Yes. Anthropic's help page says scheduled tasks run remotely, so they keep their schedule when your computer is asleep or the desktop app is closed. The exception is a task that needs files or apps on your computer. That one only runs on your computer.

Do I need the Claude desktop app?

Not to get started. File creation, Research and connectors all work at claude.ai in a browser. The desktop app matters at level 5: computer use and Dispatch both need the Claude desktop app open with your computer awake, and recording a skill is Mac only for now. Claude in Chrome needs the Chrome extension.

Can my team share a Project?

On Team and Enterprise plans, yes. A project can be shared with specific people or the whole organization, with view or edit access. On Free, Pro, Max and Ultra, a project is yours alone.

Full video transcript

This is the video word for word, apart from the meeting names on my calendar, which are left out. Where a plan or a number here differs from the article above, the article has what Anthropic's pages showed on November 9, 2026.

0:00 The 5 levels of using Claude for business

There are five levels to using Claude for business, and most people stop at the first one. At level 1, you're asking it questions, and by level 5, it's doing work while you watch. I'm Dave O'Rourke. I run BigWork AI here in Los Angeles. I build AI systems for small businesses, the kind that handles leads, follow-ups, and ties your team's knowledge together. Let's climb all five.

0:22 Which Claude plan to buy (start on Pro)

First question, which plan to buy? Free is fine if you're kicking the tires and just getting your feet wet. You get Sonnet, you get a handful of projects, you get memory, and you get web search. However, for most business owners, Pro is gonna be the one you're gonna wanna go with. You get Opus, which is a bigger model, you get Claude inside Chrome, and you also get scheduled tasks. Not to mention, you get Claude Code, which eventually you're gonna wanna be on.

Next up is Max. It's the $100 plan, and you get 5 times the amount of usage. If you're in Claude every day, this is where you're gonna end up. There's also Ultra, which is $200 a month, and if you're a super user, you will definitely end up there. Once your team starts sharing tasks, you might wanna look at the Team and Enterprise section here, $20 a month for a standard seat. Premium seats are $100 a month.

My take, start on Pro, upgrade when you hit a wall, not before.

1:16 Level 1: Where everything lives in Claude

Level 1, before you type a single word, let's find where everything lives. So we're gonna take a quick look down the sidebar. So first off, we have Search. This is where you can see everything that you have that's searchable. We have artifacts in here. We also have old chats that we have. We have different sessions. So this is a one-stop shop to directly navigating yourself around Claude.

Next up is Projects. Projects is a workspace that does one job. It has its own rules, its own files, and its own chats. This is a big one, and we're gonna build one of these in a minute. So Artifacts, everything Claude makes for you lands here. That's docs, slide decks, dashboards, designs. They start private. Nobody sees it until you share it, but it all winds up here. Scheduled Tasks, jobs that run on their own.

They're either hourly, weekly, daily, certain days of the week. They run in the cloud, so your laptop can be closed, unless that job running needs files in your computer, but for that, you're gonna have to have a paid plan. Dispatch, so you give it one job, and it's like briefing a colleague. It splits the work up into tasks. It runs them on the background on your computer while you're doing something else.

It can start from your phone too, so if you need something spun up in your desktop, as long as your computer's open and awake, you can hit the Dispatch button, task it, and it gets to work. If a task needs your okay and you don't answer it within minutes, it skips that step. It's made for the Pro and the Max plans, and it's made for the desktop app only. And a heads-up, if you have a new account, you can't turn this on anymore.

If you've got it, use it. The Design tab is where you chat, you make visuals, a one-pager, landing page, mock-ups, even clickable prototypes. Give it your brand once and everything else comes out in your colors and fonts. Next up is Claude Duet. This is an always-on agent that's working with you proactively throughout your day. As of this video, I have early access. It will be released widely soon.

And last up, Customize. This is where you plug Claude in, connectors for your apps, skills for your know-how, and plugins. We'll come back to this here in level 4. So let's take a brief look at the chat box here. We have a plus over here where we can upload files. We can add folders. We can record a skill. This is our access to our skill folders, connectors, design systems, plugin.

We can set it on research, and we can look at the memory. We come over here, and we can choose from the project folder we wanna work out of. And over here we have our model choice. This is where you're choosing your models, and then you can also come down here and choose effort levels. Another cool thing, they give access to old models. So there might be some scenarios where you might want to use one of the older models. Last up is permissioning, where you're gonna either decide whether Claude gets to make the decision or whether you wanna be kept in the loop.

Manual means that you're gonna be interacting with Claude on each decision that has to be made. That's a good place to start, but it does get tiring. I keep mine on auto and let Claude do most of the major permissioning.

4:17 The settings worth changing first

While we're in here, though, let's walk through settings real fast. First up is the General tab. This is where we're gonna play with appearance, tasks, voice, and notifications. If you come up here, you'll notice that you can change it to light or dark themes. Anthropic even made a dyslexia-friendly font. Try saying that three times in a row. But it's huge if you read all day. Usage, this shows your limits, a 5-hour window and a weekly one, each with when it resets.

If you're hitting a wall, you need to look here first, then this might be your signal to upgrade. The Privacy tab, this is where you allow Claude to train their models on your data or deny them. That's this toggle right here. If you're working with sensitive information or client-facing things that you might not want Claude to see, this is a good thing to turn off. So one more thing to note about the Memory section here is this is also a place where you can export all your data if you wanna share with another LLM or file system.

You can also import data here as well. The Reflect tab is great. It shows where your time's going, what your peak hours are, what most of your work is going into. It's a great way to see where you're spending your time on Claude. So that wraps it up for now for the Settings page. We're gonna come back to Claude Code and Claude in Chrome, as well as systems, extensions, and developers, and then the customization piece here.

5:35 Level 2: Build a Project that writes your proposals

Level 2, Projects. If you only set up one thing from this video, make sure this is the one. Think of it as a filing cabinet with a brain. A project has four parts, instruction, knowledge, memory, and chats. I'm gonna build a real one right now in front of you, show you every piece, see what it does, and why it matters. All right, first up, we're gonna come over here. We're gonna click the New Projects button. What are you working on?

We're gonna call it the BigWork AI Proposal Builder. Next up is what are we trying to achieve? This is describing the project. What are the goals, the subject, et cetera. So we're gonna paste this in right here. We're gonna say: Writing a proposal for BigWork AI, my consultancy. We help business owners pick the right AI tools, set them up, and get them results fast. Our main clients are behavioral health rehab centers, real estate, and law firms.

When I give you call transcripts, pull out client problems in their own words, write a one-page proposal, use the offer and pricing in our project files. If something's important and missing, like a budget timeline or who decides it, ask me for it before you write it. And the look and the feel. We're gonna use Modular Joy, the brand guidelines by BigWork AI. Those are good project instructions and descriptions.

You'll wanna do the same with your own. Now next up, very important step, we need to add context to this. Any files and other documentation you may have relevant to what we're trying to accomplish here is gonna be absolutely crucial. So let me go ahead and pull these documents in. So we have a brand guideline, we have a client-facing FAQ, we have a deck that we're used to using in the past.

We also have our pricing models here. We're gonna upload those. We have four files of context. Once you're in a project and you start a chat, every time that chat loads, it's gonna load with the instructions and the goals you're trying to accomplish and the context in which you've attached to it. That is massively important in getting quality outputs. Next up in projects, we have the ability to schedule tasks.

So I'm gonna schedule a task called Project Checker. This is gonna run and look for new call notes, and it's gonna draft proposals based upon finding these new notes. I have the option to let it run either manually, automatically, or skip all approvals. I'm gonna have it run on auto, so it's gonna do most things without asking for my permission. However, it will ask if it'll erase or delete files or do any major changes to my account.

The other thing we have over here is the default model. So right now we get to pick our model here. This task is fine to run on something like Sonnet 5.5. Next up we have our trigger, what schedules it. This is gonna run on a daily timer. We're gonna have it run, let's say, at 8:53 AM. Next up below is the advanced settings. This is a place where we can add connectors.

So if you're using a note taker like Fathom like I am, you can connect it here. This will scrape Fathom, find, and process those calls. We also have the ability to show how you wanna get notifications. Would you like to be notified by a push notification, an email, or Slack? We're gonna give it push notifications for now, and we're gonna say Create. All right, our project is now set up. Now we have proper instructions, we have proper context files.

Since this is new, there's no memory that's been generated yet. And we have a scheduled task that we just set up, which is that prospect checker. All right, let's go ahead and test this out right now.

Please create a proposal for this new prospect we just got off the phone with. Let's see what we get back. And it's working, and boom. Here's what we get back. This is what good looks like. Based upon our instructions and the context we gave it, the AI pushed back and asked for an answer from us.

Who signs off on Ridgeline? Maria said, quote, "I'd have to check," and the proposal depends on that. So it's outlining the problems here, why this decision matters, and what we do next on the call. So I'm gonna say, "That's okay. You can write it for Maria now.

I had a conversation with her via text after the call, and she said she's good to sign off on it." Here we go. Ridgeline Family Law proposal prepared for Maria, office manager, November 2026. This is the outline of the call we had and talked about. Here's our price. This is the next steps. The next step, which was specified in our instructions were, "Give me a branded deck done in the style of Modular Joy," which is BigWork AI's branding.

So let's go ahead and say, "This looks great. Please present a finished version." And the final step delivered. Here's the proposal. Let's push this up on the big screen. Perfect. This is exactly what we wanna see. This is BigWork AI branding. We've now built a system that gives us a quality output every single time. So the takeaway from this is 10 minutes of setup once, and you have repeatable quality outputs.

If you're not using projects, this is the reason to start. Think about all the other places that you can build these in your company.

10:55 Make Claude answer the way you like

A quick pit stop before level 3. Let's get Claude to answer you the way you'd like to be answered. Global instructions. We're gonna come over to your Claude, tap that, and you're gonna get instructions. These are your global instructions, and this is how you can control how Claude responds to you. These are my global instructions, and I'll leave a link for these in the details below. How to talk to DaveO. Master communication skills.

These are rules that override all default behaviors in every response, every project, every domain. So who are you talking to? You're talking to a busy entrepreneur with strong business intelligence who works across many disciplines, not a specialist in any of them. He makes fast decisions and asks for depth when needed. So the 10-second rule, every response must be readable and understandable within 10 to 20 seconds.

I set this because I found myself getting bogged down in reading blocks of text, and once I sifted through all that information, I felt like there was only a couple of sentences that actually were really important that helped me make a better decision. So this is how I crafted this. Give me a bottom line in one sentence, why it matters in two to three short bullets, and then a next step, one clear recommended action.

Next up, plain English always. Zero jargon. Claude loves to get into the details and the weeds. That is too much for most of us and really not needed. Explain everything in a way that you would explain to a sharp business owner and never a fellow specialist. Next up is hard limits. No walls of text, which we talked about already. Never dump full code, logs, configs into the reply.

Don't overexplain your process and handle one question at a time. Depth on demand, there are some times where we do want Claude to go deeper, and in those cases, we can ask it and we can dig in together. Next and last is decisions. Frame everything as a decision I can act on. If there's a judgment call, make a recommendation and say why in one line. Don't hand me more homework.

12:53 Memory: what Claude remembers about you

Underneath those global instructions in the Your Claude tab, you have memory. So we can search and reference old chats. We can also generate memory from chats. So if you want privacy and you don't want Claude remembering things about you, these are going to be places you're going to want to turn off over here. But that will limit the fluidity in which you can work with Claude and how much context it's going to have and remember about you.

So this is a judgment call. Toggle these on and off as you see fit. Now, if you want to go deeper, you have your memory right here. We can manage your memory. You can make adjustments to your memories. You can erase things if you don't want them in here. You can also add things as well. The last thing in this section that's really important is this button here, Import Memory. You can import memory from any file systems you may have.

You can also export memory from other LLMs that you've been using and upload that data here.

13:49 Incognito chats: safer, not safe

Now, if you want to keep your memory on, but you're looking to have a quick chat with Claude that's off the record, there's a great little tool for this here, which is this little ghost icon up in the corner. This is your incognito button. I'm sure everybody's familiar with this. Google's been doing this for a long time. Claude has one of their own. Anything you type here in this chat history, Claude won't find it later when it searches your old chats.

It doesn't use your memory. It doesn't add anything to it. It's never trained on their models, even if you have training switched on. If you want to get a private chat in that doesn't record to your memory or take up any data, this is the place to go to do that. Now, one quick caveat to know about the incognito tab is the incognito doesn't mean that nothing gets stored. Anthropic still keeps that chat for 30 days for safety reasons, then deletes it.

If a chat gets flagged for breaking usage rules, they can hold it for up to 2 years. A few limits before you rely on it, your instructions still apply, so it still sounds like you. It doesn't work inside projects. It can't make or run files and run code. And once you close it, it's gone. You can't reopen it. You can't turn it into a normal chat. If you are on a team or enterprise account, your account owners can still export incognito chats, so keep that in mind.

So is incognito safe for sensitive data? Well, it's safer, but it's not completely safe. Use your incognito for personal stuff you don't want in your history. Never paste in data that you aren't allowed to share, client health records, Social Security numbers, legal case files, et cetera. And one more thing, if you're using Claude Code, there is no incognito there. So my rule, memory on for business, incognito for private stuff you're allowed to share, and the stuff you're not allowed to share never goes in any of these chat windows.

15:35 Level 3: Get real Excel files you can send

Level 3, let's put together some real files. These are spreadsheets, decks that you can actually send clients. Claude doesn't just talk. You can hand it real files, open it, edit, and send. Here's what I'm talking about. This is a project, a demo project we're doing for this, and it's for a client called Ridgeline Law. I'm going to give it a simple prompt to make me an Excel pricing sheet, and let's see what happens.

Please make me an Excel pricing sheet based upon the data, FAQs and transcripts from the last call. It's spinning up an Excel file right now. Something else you can do is click in and see the work that Claude's actually doing. Loading skill. It ran a request. It failed, but it's trying it again. We're going to let this work for a sec. And here we go. Here's the output.

It leads with a first win fix. Every option is tied to her numbers. The totals are live formulas in which we can edit. So let's push this up onto the big screen. Here we go. A fully functional Excel file that we can edit and work on. We can also come over here too, and we can export this directly to Google Drive. And now we have this in our Google Drive folder. Once we're done editing this, once we make sure that it's buttoned up in exactly the way we want it, this is ready to send.

And as we click into Google Sheets, we can come down here and see some of these numbers and see that there's fully functioning formulas up there that can be edited. Lastly, when we come back to the main chat, you have the option to download this and open it in Excel.

17:06 Research: 330 sources in 5 minutes

Let's jump in and talk about how research works, how to find it, and how to use it. You're going to click the plus button. We're going to come down to the Research button. You're going to see this little blue icon pop up. That means research is now turned on. So now let's send it out in a research task.

Question, who sells AI setups to small law firms in Los Angeles, and what do they charge? Okay, you can see it's working now. We can open this up. We can see a bit more about what it's doing here.

It's cited 50, 78 sources now and counting. We're going to let this work and come back in a few minutes. It's still working. It's gathered 271 sources now and counting. So the research is back, LA law firms in the AI market. It went through about 330 sources, and it took 5 minutes and 5 seconds. It shows pricing and how relevant my pricing is compared to other agencies out there.

It shows gaps in the marketplace, and then it surfaced something really interesting. California's SB 574 takes effect January 1. So what is that? That's a new law that lawyers have to disclose when they've used AI to the courts. So two things to know here. Research, you need your web search tool turned on, and if you're doing it on your phone, it skips your connected apps.

You want to run your big research audits on your computer.

18:24 Level 4: Connect Gmail and Google Calendar

Level 4, we plug Claude into your emails, your calendars. This is where it starts to feel like an assistant. Customize, connectors. I can connect Gmail, Google Calendar. You'll see in a second. We can do it in two ways. We can come down here to this plus button. We can come scroll down through here, and we can see the connectors and the tools that we have access to already. But how do you set this up?

We're going to come back over to this Settings button here. We're going to go down into Customize, and here we go. We have connectors, skills, and plugins. Let's focus on connectors first. So these are my connectors that I have currently set up. If you wanna discover a connector, we can come over here and hit the Discover button. These are the connectors that Claude can give you access to that you can connect to your account.

Look through this list, it is extensive. And a cool bonus, if there's a connector that's not in here that you'd like to add, you can add custom connectors right here. This is a little bit more advanced, and we're not gonna get into it right now, but you should know it's there. So let's see the connectors in action. I'm gonna come over here to my chat window and say, "Hey Claude, can you tell me about some of my important meetings coming up next week?"

I'm gonna let it do some work here. It's pulling up my Google Calendar right now. And here's the results. It surfaced a few different meetings to the top [meeting names left out of this transcript], and you see down here next steps, Claude's already trying to be proactive and asking if I need a pre-call brief for [one of them].

Let's use another connector. Let's use Gmail, and to keep client information out of this, I'm gonna give it a generic prompt. "Hey Claude, can you go through my emails and surface the best AI newsletters that have come out this last week?" So you can see it's tuning up Gmail right now. We're gonna get a response back in a second. Okay, and here's what we get back, full context by Derrick Hicks, my boy.

This is a great newsletter, by the way. I'd highly suggest subscribing to anybody out there that's looking for getting the best news and best up-to-date information on AI. We also have Eric over here at the Lean Startup. Also worth skimming, Nate Herk, another legend in the space. So you get the point. You saw the full connectors page. You saw how many are in there. You also saw that you can create your own, so sky's the limit here.

This is a big unlock for making Claude more like an assistant and less like a chatbot.

20:55 Skills: save your know-how once

Next up is skills. Skills is saved know-how. It's instructions Claude can grab on its own when it needs them. Think about it like making your favorite meal. There's a certain bunch of ingredients that go into that. There's a certain process, a certain timing to get the result of your favorite meal. Claude ships with its own built-in skills, but let's find where these skills live first. Very close and similar to the connectors, we come down here to Skills.

These are all my skills. We have this list here of skills that I can use. Some of these are actually Claude skills as well. Let's tap into the Customize button. We're gonna come up here to the top. We're gonna hit the skill. We're gonna go to My Skills first. We're gonna start off with my personal voice. This is DaveO's voice, and this is how I like content to be written. So you're gonna have a SKILL.md file here, and this is what this says, "Apply DaveO's authentic voice to any outward facing content.

This skill provides tone and style." So this is the rules on how to sound like me. I've recrafted this skill multiple times. I've trained it on a lot of podcasts, sales calls, and client calls that I've been on to really get a quality output that I'm happy with every time. Come back over here, we have Discover. So Claude also offers their own set of skills and some packages.

If you're running a small business, I'd highly suggest the small business package. You get 44 skills, 35 connectors. This is available on any Pro plan and up. One other tip and trick is you can record a skill based upon your actions on the computer. So you hit the plus button here. We come down to record a skill. When you record a skill, it records your screens, your clicks, your typing, your voice inputs.

It sends it all to Claude, and it turns it into a repeatable skill.

22:42 Level 5: Give Claude a real job (Monday lead report)

Level 5, this one's fun. This is when we stop chatting and let it do the work. We give it a real job with real steps. Here's what we're gonna prompt it with. It's gonna be, "Build my Monday lead report based upon my Gmail activity this last week, and please put it in a Google Drive folder and turn it into a Google sheet." We have two permission modes here, manual, which you're gonna have to okay each step that Claude does, and auto, where Claude's gonna make a ton of the decisions on its own.

I highly suggest turning auto on and letting it go. Here's Claude working. It's in my Gmail right now. If we wanna see what Claude's doing in more detail, we can click into this, and we can see the steps that Claude's taking. Searching my Google Drive right now. It's building the lead report, and here's my response. If you remember earlier in the video, the response style I like is clear-cut, what stands out, give me three bullet points.

If there's any pushback, I'd like to hear this, what it covers, and then give me the next step. What's the next decision that I need to make? It's clear, and I can read it within about 30 seconds. As we can see, I'm over in my Google Drive here, and here's the Monday lead report that was just generated. Let's click into this and see what it looks like. Excellent. Priority levels, leads, companies, the stage we're at.

Now, this is a great output for just a generic prompt I gave it. If you spend more time with this, you can craft it and get something that really makes sense for your business.

24:11 Make it run every Monday on its own

Now, we can do something really cool that turns this into a scheduled task and runs it every Monday at 7:00 a.m. So I'm gonna say, "Claude, please turn this into a scheduled task that runs every Monday at 7:00 a.m.," and it's off to work. It's done, and again, here's my output style, clear response, a couple of supporting bullet points, a pushback here, which is we already have lead gen set up for Monday. This was obviously for the purpose of doing this for the video here.

Next, though, we're gonna come over here and show you where the scheduled task lives. So these are the scheduled tasks I currently have. There's templates here that you can run yourself. There's tasks right here that are your tasks that you can make, and then you can create also a new task here. One last tip for scheduled tasks, don't go overboard with it right away. Set one up, use it. Find the value in it, and then set another one up on top of that.

I will say, sometimes these break, you do have to pay attention to these, sometimes you have to set them back up again. That's why I'd suggest one valued scheduled task and then build upon that.

25:10 Claude in Chrome: let it work in your browser

This is a big one, Claude using your computer. This lives in the desktop app, Mac and Windows. It works on Pro, Max and Ultra accounts, and it's still in beta. So where do you find this? We're gonna come down here, we're gonna go to Settings again, we're gonna go to Capabilities. It's in the Capability section here, and we're gonna scroll down, and you're gonna see Claude in Chrome. The title of the section is Claude in Other Apps, so this is gonna be Claude in Chrome.

We're gonna wanna toggle this on here, and this is what it does. It opens up a side chat in your Google browser. Think of this like a really smart employee that has access to your keyboard and browser. So we could say something like, "Go into my Make account and tell me which scenarios I have running currently." So it's gonna get to work, and it's gonna take over my browser here in a second.

You can see it up here. It's working. Here's the magic. It went ahead and signed in. Now that I'm in, it's got my browser, it's taken over again, and it's clicking through my Make account. Pretty cool. Just to show it working a little further, I'm gonna give it a prompt. "Click into My Bully, Shared Image, Personalized Roast. Tell me a little bit about that scenario."

Clicking through. Here you go. It's got my browser, it's pulling up the scenario. What's cool and magical and crazy about this all at the same time is that it can click into each one of those scenarios and actually fix bugs for me. I've literally sent it out to fix a bug in here. It fixed the bug for me, relaunched the app without me touching anything.

That's super cool. So that's Claude in Chrome. Now, coming back over to the Settings section, there's two other tabs that we need to check out here, which is Site Permissioning. We're now allowing Claude to work on all sites. We can actually permission Claude and have it block from sites that we don't want it to be on. You just click this button here, you add a web address, and you are good to go.

27:09 Computer use: Claude builds a 3D me in Blender

Next up, the big one, computer use. This is in Settings. Again, Capabilities. We're gonna come down underneath this. We were just at Claude in Chrome. We're gonna scroll down here, and we're gonna find Computer Use. This is in beta, and it's only available to run in your local machine. We're gonna click Enable Computer Use, and we are going to run this in the background.

Otherwise, what'll happen, it'll take over the computer, and you won't be able to work on it. All right, back in chat now that computer use is turned on. I'm gonna give it a prompt. I'm gonna say, "Open Blender and create me a fun 3D-style character called Ultra DaveO. This is gonna be based upon my likeness, so feel free to find any files and pictures and images on the computer that look like me or are me, and build me a fun 3D character in Blender."

And we're off. It's loading tools, searching and finding files on my computer, and it's got access to Blender now. As you can see, here's an update. Here are some screenshots. It's currently working in Blender, and this is what we have. You know, not the best, but hey, we just single-shot prompted that, and we had that about three minutes later. Pretty cool.

It's also using an incredibly complex platform called Blender, which does a lot of animations. You see this in a lot of movies. If we come over to this screen, we can actually see it inside Blender right here.

We can play around with him a little bit. Let's come back over, though, to our Claude chat and say, "This works for now. Please go ahead and animate him and give him a little bit of a handshake arm wave." So we're back in here. Let's click in and see what this is actually doing. It's executing Blender code here, and just like that, it's already done.

Let's come over here and see what's happening. Oh, and we got our little character waving at us now. Pretty cool. Hey, let's keep playing with this. Honestly, this is fun. This is why I do this all the time. Please go ahead and run this through fal.ai and animate it, and then use my DaveO ElevenLabs voice to say, "Hey, what's up?

I'm DaveO. I just built this whole thing in Claude." It's using the connectors. It's connecting to my directory right now. Here's what's happening. It pulled up my fal account, and it also pulled up my ElevenLabs account. Another cool feature of computer use is it has access to your file folders and your desktop apps if you let it. It just asked for that. Let me give it permission for that. Here we go.

Quick behind the scenes. This is Claude working in ElevenLabs right now. All right, and it's in. Let's check the results. We have a 4-second video. It cost about 56 cents for us to make this. Next step, if the lip sync looks off, I can give him bigger, more animated mouth and rerun it. All right, moment of truth. Let's give it a look. Here we go. Hey.

What's up? I'm DaveO. I just built this whole thing in Claude. Ha. Let's give it one more shot. Hey. What's up? I'm DaveO. I just built this whole thing in Claude. You know, the lip sync isn't so bad. The character is, you know, not my favorite at all, but this was for demo purposes anyways. But you guys get the point. Obviously, you can see the power that you can wield with computer use.

30:38 The ladder, and what to do next

So that's the ladder: the map, Projects, real files, your apps, Claude doing the work while you watch. Most people stop at level 1, so if you made it this far, that's definitely not you. Good job. Now, if you want this set up inside your business with your tools, your voice, that's what we do here at BigWork AI. Grab a free roadmap. The link is below. And if you're a ChatGPT person, I'm gonna do the same walkthrough in another video.

Now go build something.

Want this set up inside your business?

With your tools and your voice. That's what we do at BigWork AI, and it starts with a 30-minute fit call.

Book a 30-minute fit call