you don't have to learn AI. you have to use it. this is the practical stuff, in order: set it up, prompt it right, give it your business, and turn your best work into something you reuse. the room covered four moves. this is the other thirteen. bookmark it, it's your map.
two parts. the basics get you running. the knowledge base is everything to come back to. tap anything to jump.
most people use claude wrong on day one. not because it's hard. because nobody showed them the switches that matter.
Settings, then Memory. Now it remembers how you work so you stop repeating yourself every chat. Do this first. And the line that clears up most of the confusion: Memory is Claude remembering you everywhere. A Project is a folder for one job. You want both. They are not the same thing.
Everything we did in the room works on the free plan. When you're ready, the paid plan with the desktop app is where Projects, Skills, Cowork, Code and Research live. That's where the real leverage is. Don't buy it until you've used the free one for a week.
Tell it who you are and how you like to work. It carries across every chat, so you brief it once, not forever.
New chat, model picker, web search toggle, Projects, Settings. That's the whole dashboard. Nothing else to figure out.
Click the model name next to the send button. Effort goes Low, Medium, High, Extra high, Max. High is the default. For a proposal, a contract, a pricing decision, go up. For a quick email, leave it. It's the one dial most people never find.
a prompt isn't a search bar. you're not searching. you're briefing. the people who get great answers aren't smarter. they just give more context.
try it. start with a lazy prompt, then tap each layer to watch it turn into something that actually sounds like you.
if you re-explain your business every chat, you're working twice as slow as you need to.
a project is a folder that remembers. set the context once. every chat inside it already knows your brand, your voice, your files.
Name it for the job. "Sales emails." "Client X." "Weekly content."
Who you are, your tone, the rules. This is the brain it works from.
Brand docs, past work, pricing, SOPs. It reads them and works from them.
Now every answer starts from your world, not a blank one. No more re-explaining.
The instructions field is where most people freeze. That's the whole game. Here's the structure that works. Swap anything in [brackets] for your details.
Instructions are rules. Files are reference. When they disagree, the rule wins. And what you type in the current chat beats both.
The trap: two files that contradict each other have no reliable tiebreaker. You can get one answer today and the other next week. So keep it clean:
claude doesn't just describe the thing. it builds it, right there next to the chat.
ask for a tool and it appears live. don't like it? tell it what to change and watch it update in front of you. no design skills, no code.
you don't want to write the perfect prompt every time. you want to write it once and reuse it forever.
a skill is a saved workflow. a repeatable job you taught claude to do your way, on command.
A set of instructions claude follows when you ask. The job runs the same way every time.
Write out exactly how you want it done. Save it as a skill. Now it's one ask, not a fresh brief.
Ready made skills for marketing, sales, ops, and writing. You don't have to build from scratch.
Drop in a skill and it runs your process without you re-explaining a thing.
most people only ever use the first one. the other two are where it stops answering and starts doing.
A conversation. You ask, draft, brainstorm, or think out loud. Claude answers right in the chat window.
A workspace. Point Claude at a folder and your tools, describe the outcome you want. It plans the work, does it, and delivers.
An agentic coding tool. Claude works inside a codebase, editing files, running tests, and shipping the changes.
chat is where you start. cowork and code are the next two gears, and they live in the desktop app on a paid plan. get sharp on chat, projects, and skills first. reach for the other two when you're ready to build something real.
the real unlock isn't claude doing a task. it's claude doing it every week without you asking.
scheduled tasks live in cowork and run on a cadence you set. the command is /schedule. two ways in: type /schedule, describe the task and how often, and review the prompt it drafts. or the cleaner path, do the job once, confirm it looks right, then /schedule to turn that exact run into a recurring one. pick hourly, daily, weekdays, or manual.
one catch: it only runs when your computer is on and claude is open. miss a window and it catches up the next time you're back.
claude doesn't have to stay in a chat window. it can work right inside your browser.
claude in chrome is an extension that sits in a side panel while you browse. it reads the page, clicks, fills forms, and runs multi-step tasks for you. add it from the chrome web store. it's in beta on every paid plan, so everyone here can use it. pair it with cowork and it gathers from the web while cowork builds the file.
heads up: it's beta and it takes real actions on sites for you. it pauses at logins and anything sensitive so you stay in control. give permission deliberately.
the part above is what we covered in the room. everything below is yours to keep. the prompts, the templates, the how-tos. come back anytime.
the people who get the most out of claude aren't better prompters. they just do two things everyone else skips.
The first answer is a draft, not the final. Don't open a new chat. Tell it what's off. "shorter." "more like me." "drop the third point." The second prompt matters more than the first.
It can sound completely sure and still be wrong. For anything that counts, check it and ask it for its sources. Confidence is not proof.
most bad results aren't claude's fault. they're one of these five habits.
Treating it like google. One-line searches get one-line value. Brief it instead.
Taking the first draft. That's the rough cut. Push back and it gets better fast.
Being vague. "make it better" means nothing. Say what better looks like.
Starting over when it goes sideways. Correct it in place. It remembers the thread.
Pasting confidential stuff on autopilot. Know your company's rules first.
you don't have to describe what's on your screen. show it.
drop in a screenshot, a pdf, a photo, a spreadsheet, a contract. it reads them and works from them. paste an error message, a chart, a whiteboard photo. if you can see it, so can it.
you don't have to start from scratch. here are six that do real work. tap to copy.
the biggest jump isn't a better prompt. it's the day claude can open the tools you already pay for.
Gmail, Google Drive, Calendar, Mailchimp, Shopify, QuickBooks, Slack, Notion. You connect it once, in Settings then Connectors. From then on you say "check my calendar" or "pull last month's orders" and it does, instead of you copying and pasting between windows.
HOOPTEQ is a basketball shooting machine company I'm CMO of. Claude is connected to its Mailchimp account. I told it what the email needed to say. It designed the email and sent it to the list. That's the one that was on the screen. A real send, to real customers, from a prompt.
Your email or your calendar, whichever you open most. Connect it, then ask Claude to do one thing you'd normally do by hand in that tool. Summarize this thread. Find every message from this client. When that works, add the next one.
Every connector asks your permission and you can disconnect it in one click. Start with reading jobs, summarize and find, before you let it send or change anything. Then widen the leash as it earns it.
this is the one paid feature a beginner actually feels on day one. it's the difference between an answer and a report.
Normal web search does one lookup. Research runs search after search, each one building on the last, across the web and your connected files, then hands you a written report with sources you can click. Five to fifteen minutes of work you'd otherwise do in twenty tabs.
Who distributes bike locks to big retail in Canada. Which Edmonton trade shows are worth a booth. What my three competitors charge and how they position it. Who the decision makers are at a company I want to pitch. Suppliers, grants, regulations, a market you don't know yet.
Paid plans only (Pro and up). In a chat, click the plus button at the bottom left, choose Research, and make sure web search is on. Ask the question the way you'd ask a smart assistant, with what you already know and what you want back. Then go do something else.
Research is for finding out. Writing, replying, planning and the four moves from tonight don't need it and it'll slow you down. If the question fits in one search, use one search.
the fastest way to get good is to use it on what's already on your plate. start here.
the jargon is the only hard part. and it isn't even hard.
it's powerful. a little judgment goes a long way.
Know your company's AI policy before you paste client data, contracts, or anything private.
Drafts are drafts. What goes out the door is on you, not the tool.
Verify facts, numbers, and quotes before they leave your hands.
You've got everything you need. Pick one thing from tonight and build it before you close the laptop. That's how it sticks. The four moves and the exact prompts from the stage are on the Tonight tab.
Stuck on something? Email navin@ramharak.com
@navinramharak