[00:01] together they cover a huge portion of everyday work. From information analysis collaboration, and full task automation. each one so you can see what Claude is truly capable of and which features you
[00:13] might be missing out on. We start with chat because it's the layer everything else runs on. I want to lock in three building blocks here and then show you when to hand the work off to the other features. Projects are folders with
[00:26] memory. I keep one project for ongoing thing in my life, one for my job hunt, apartment renovation I've been dragging through for a year. Inside each project,
[00:38] I drop the documents Claude actually needs, like my resume, a few job descriptions I'm targeting, or the contractor quotes I'm comparing. From that moment on, every chat I start inside that project already knows what
[00:51] I'm working on. The shift here is subtle but huge. You stop re-explaining yourself every morning. I open the job hunt project, I paste a new job description, and I just ask for a tailored cover letter. Claude already
[01:04] has my resume and the four roles I've been targeting, so the answer lands in the right shape on the first try. Inside each project, there's a custom instructions field, and this is where I install a persona. For my job hunt
[01:16] project, I tell Claude exactly what level of role I'm targeting, what industries I want to avoid, and that I want cover letters that sound like a human wrote them. For my recipe project, I tell Claude to first cook for two
[01:28] people, I hate cilantro, and I want grocery lists grouped by aisle. The field is not a wish list, it's a constraint sheet. The tighter and more specific you make it, the better the output gets. Mine runs about 150 words
[01:44] per project, and that paragraph does more work than any prompt I write afterwards. Claude has two research modes, and most people only notice one of them. The basic research toggle sits right above the prompt bar. You flip it
[01:57] on and Claude searches the web in real time before it answers. Useful for quick competitive checks like what are people saying about Claude versus GPT-4 right now. Results land in about 30 seconds. Deep research goes further. It takes 5
[02:12] to 10 minutes instead of 30 seconds, but it reads dozens of sources, cross-references them, and builds something closer to a real research brief. I use it when I actually need to understand the space before I commit to
[02:24] a video. Competitor landscape, pricing shifts, what a new model actually benchmarks on. The output is a structured doc you can save and reference, not just a chat reply. Next up is MCP. It stands for model context
[02:37] protocol. It is how Claude connects to the tools you already use. Notion, GitHub, Google Calendar, Slack. So it can read and write to them directly inside a chat conversation. You set it up once under settings, then customize,
[02:51] then connectors, and you pick which apps to authorize. Here's how that looks in practice. I open a regular chat conversation. I type "Add this topic idea to my content calendar for next Tuesday." and Claude writes directly
[03:05] into my Google content calendar without me leaving the conversation. No copy-paste, no tab switching, Claude just makes the entry. That one connector alone saves me about 10 context switches a day. Use chat for thinking out loud,
[03:20] for drafting, and for any one-shot task that fits in a single conversation. The moment the task needs to touch files on your machine, you move to code. The moment it needs to run on its own without you, you move to routines. That
[03:34] handoff is the whole game. Quick break. If you're launching anything visual, you usually pick between two ugly options. Either you shoot it yourself on a kitchen table at midnight, or you ask Claude or GPT and get that plastic AI
[03:48] slop. Stop wasting your time. Love art sits between those two and I'll show you on a real case. My friend Maya is launching Echo form, a premium over-ear headphone brand this month. Her starting point is just one product shot on a
[04:02] clean studio background and a screenshot of a brand she loves. I dropped both into Love Art's brand kit, and it instantly pulls the color palette, typography, and visual direction into one coherent system. Then, I prompt the
[04:16] full launch campaign, and Love Art builds the entire brand ecosystem on the canvas. Instead of generating just a few images, it creates a complete set of images, it creates a complete set of launch assets. A hero product render, a
[04:29] lifestyle campaign shot, a website banner, packaging design, a product feature sheet, and other brand materials, all following the same visual direction. Once the core campaign is ready, I expand it into marketing
[04:42] ready, I expand it into marketing content. I ask Love Art to turn the same brand system into social media assets, including Instagram creative, carousal posts, story formats, and paid ad variations. Every piece feels connected
[04:56] because they all come from the same campaign foundation, not separate AI-generated images. From the hero image, I use Love Art's multi-angles feature to control the camera perspective and generate multiple
[05:09] product views from the same shot. I can adjust the camera position, create a side profile, close-up detail shot while keeping the product, lighting, and brand identity consistent. This gives Maya all
[05:22] the visual variations she needs for her website, ads, and social content without recreating the product from scratch. Link is in the description with free credits attached. Sign up and run the same flow on your own brand tonight.
[05:35] let's move to Claude code. Let's say I want a tiny one-page web app I can actually use at the gym. It's a rest timer between sets. There are three preset buttons for 60, 90, and 120
[05:48] seconds. A big countdown sits in the middle and the page beeps when the timer hits zero. I want everything in a single HTML file with plain JavaScript and CSS so I can just double-click it on my laptop and it runs. I open the Claude
[06:03] code in the project folder I just made and I paste that brief. Claude reads my and I paste that brief. Claude reads my claude.md and asks one short question about the visual theme. Then it gets to work. It creates index.html and writes a
[06:15] small styles block at the top. It drops in the timer logic and wires up the three preset buttons. Finally, it hooks up a short beep for the web audio API. The whole build takes about 90 seconds and I watch every file as it lands. Then
[06:30] I do the thing that makes code feel different from chat. I tell it to remember the last preset I used and select it again on the next load. Claude opens the file again and adds two lines that touch level first hand. The page
[06:44] now remembers 90 seconds between visits and I never had to open a text editor or paste a single snippet. That is what code actually does. It reads and edits every step. In regular chat, you would have copied a snippet, pasted into a
[06:59] text editor, refreshed the browser, and then asked Claude why it broke. In code, you stay in the chair as the reviewer and the agent runs the loop. There is one more thing worth calling out here. Claude code can update dozens of files
[07:14] in a single pass. That is actually one of the biggest reasons people switch from chat to code. You describe a refactor once, say rename a function or change an API endpoint across the code base, and Claude finds every file that
[07:27] references it and updates them all. In chat, you would have been copying, pasting, and hunting for the third place you forgot to change. In code, that whole sweep happens while you watch. The single biggest unlock in Claude code is
[07:40] a file called claude.md. You drop it at the root of your project and the agent reads it before every task. Mine is split into four short sections. It says what the project is and what it does. It lists the stack,
[07:54] the libraries and the conventions I follow. It tells Claude which commands to run for tests and builds. It also names the files Claude should never touch without asking first. That file is the difference between an agent that
[08:07] guesses and an agent that behaves. Spend 20 minutes writing yours once and you save hours every week after that. Three rules I put in every Claude {dot} md and you should, too. Ask me before any destructive operation, like deleting
[08:21] files or force pushing. Always run the test suite after a change. Summarize every diff in plain English before applying it. Those three lines keep the agent useful instead of scary. Quick note before we move on. AI Master is the
[08:33] platform my team built for ourselves and we teach the exact workflow we use every single day. Inside you get a full library of courses on the tools that deep track on Claude covering chat, Cowork and design. There's an image
[08:48] generation track with lessons on Midjourney, Nano Banana and GPT image, too. And there's a video track covering Seed Lens, too, Vio and Kling. On top of the courses, you get a sandbox to practice on the strongest models the day
[09:01] they ship. Right now there's 30% off the annual plan and that's the version most of our students actually use. Head to AI Master if you want the same setup without piecing it together yourself. Now let's keep moving through the stack.
[09:14] Cowork is the part most solo creators skip and most teams under use. It's a shared workspace where multiple people share the same project context, the same memory and the same custom instructions. Here's the problem Cowork solves. Say
[09:27] two-week trip to Japan. You both open Claude on your own laptops. You add a bunch of context about Kyoto temples you want to see. Your partner adds notes about ramen spots and hotels in Osaka. Neither of you sees what the other one
[09:41] itineraries [music] that don't talk to each other. Catwork fixes that by putting the context in one place. You both see the same notes, the same constraints, and the [music] same prior decisions. Claude builds one itinerary
[09:54] that actually reflects what both of you want. Let me show you a real one. This project is called Japan trip October, and there are three of us in it: me, my for the second week. We each have different priorities and different
[10:08] travel dates. Pinned at the top of the project, I have our flight confirmations, a rough budget doc, and a shared list of must-do versus would-be nice activities. Every chat any of us starts in here already reads from those
[10:23] three files. My partner can ask for a Kyoto day plan that hits three temples and a tea house. Her sister can ask for a Tokyo evening plan inside our budget. I can ask Claude to merge them into one calendar. The other thing Catwork added
[10:38] recently is concurrent editing. Two people can work in the same conversation at the same time, the way you would in a Google Doc. In our trip planning, my partner is editing the Kyoto day while I'm tweaking the Tokyo one, and Claude
[10:50] keeps up with both of us in real time. Use Catwork the moment more than one human is touching the same body of work. A couple planning a trip, a small team writing a proposal, a founder and a virtual assistant managing a calendar,
[11:05] anywhere the context needs to live in one place instead of in everyone's head. Next one is where it gets fun. Design is the feature that surprised me the most. It produces slide boards, landing pages, social posts, and even short motion
[11:18] graphics, and it does it inside Claude. No Figma round trip, no separate tool. Before I show you anything pretty, I want to give you the most practical tip in this whole video. Design burns tokens about twice as fast as regular chat.
[11:33] Every generation renders a full visual, and that is expensive under the hood. If you walk in and start vibe designing, you can chew through your monthly allowance in a week. [music] So, here is the rule I follow. I plan in chat, I
[11:47] build in design, I write and refine the prompt in a regular Claude conversation first until the brief is tight. Only then I bring it into design and generate. That single habit cuts my design token use by more than half. The
[12:01] biggest mistake people make in design is vibe designing without a design system. You type, "Make me a landing page for my AI tool, modern and clean." [music] And Claude gives you something that looks like every other AI landing page on the
[12:15] internet. Generic gradient, generic font, generic hero. It's fine. It's also forgettable. The fix takes about 15 minutes, and it changes every output you ever generate after that. Here's the move that changed how I use design.
[12:29] Inside the design systems tab, you can import a screenshot of any site you love. Claude reverse engineers the system straight from that image. You get the full palette and the typography back in one pass with the component styles
[12:41] already mapped. I picked a landing page I actually like the look of. One screenshot, nothing else. I tell Claude, "Hey, build me a design system based on this site." That is the entire prompt. Now I want to use this system on
[12:55] something real. So, I'm going to build a landing page for AI master with it. I open a new design canvas, [music] attach the system I just made, and I prompt for a hero section, a feature breakdown, a pricing block, and a CTA. One
[13:09] generation, and it lands close enough that I'd only tweak copy from here. No restyling, no fighting the model on colors. That is the whole point of front-loading the system. Same design system, different output. I ask for a
[13:23] five-slide pitch deck for AI master. Title, problem, solution, traction, call to action. About 2 and 1/2 minutes later, I have a deck that already matches our brand. No font cleanup, no color correction, no re-aligning boxes.
[13:37] This is the part that converted me. Decks used to eat a full afternoon. Now they take the time it takes to write the bullets. Next one is Claude routines. It is the reason this stack stops being a set of tools and starts being a system.
[13:51] A routine is Claude running on its own on a schedule or on a trigger without there are two flavors. Local routines only run when your laptop is on and awake. Remote routines run in Claude's cloud 24/7, whether your machine is open
[14:06] or closed. For anything you actually want to rely on, use remote. Local is fine for tinkering. You open the routines tab, you click new routine, you write the prompt, which is just what you want Claude to do every time it runs.
[14:19] You pick the model, you toggle adaptive thinking on if the task needs reasoning, off if it's simple fetch. That is the whole set up screen. Before a routine can do real work, it needs access to your tools. Under more customize
[14:33] connectors, you connect what you use. Slack, Google Calendar, Gmail, Notion. The connector list is long enough to cover most stacks. You authorize once and every routine in your account can use those connections. First demo, I
[14:47] want a routine that runs every morning at 6:00 and tells me what my day looks like with research already done on anyone I'm meeting. I call it prep buddy. Schedule triggered daily 6:00 a.m. The prompt does five things in
[15:01] plain English. Pull my Google Calendar for the next 24 hours. Bucket events into personal, internal, and external. Skip the personal and solo ones. For external meetings, research the company and the attendees. Post one Slack
[15:15] message per event to my prep channel with the meeting time, the attendees, and and brief. I hit save. Next morning at 6:00, before I open my laptop, my Slack has one tidy message per real meeting with the background already
[15:29] done. That used to be 45 minutes of manual work. Now it costs me zero. And if you want to go further, you can wire your CRM in as another connector, and the brief gets even richer with deal stage, last contact, and open issues.
[15:44] Second demo, and this one ties back to the gym thread from earlier. I want a tells me how I actually did, and suggests what next week should look like. I call it weekly gym recap. Schedule trigger every Sunday at 8:00
[15:57] p.m. The setup is almost embarrassingly simple. I log my sessions in a plain Google Sheet, one row per set with date, exercise, weight, reps, and a notes column for how it felt. Nothing fancy, no app. The Google Sheets connector is
[16:12] already authorized in my cloud account, so the routine can read that sheet directly. The prompt does four things in plain English. Pull every row from the last 7 days. Group the volume by muscle category, so push, pull, and legs.
[16:26] Compare the working weights against the previous week, and flag anything that moved up or stalled. Then suggest a loose plan for next week, respecting that my left shoulder gets cranky on overhead press. Last step in the prompt
[16:39] is to email me the recap. Subject line is the week number, the body is a short summary on top, then the volume breakdown, then next week's suggested loads. I hit save and pick Sunday at 8:00 p.m. as the schedule. Sunday at
[16:52] 8:00 p.m., the email lands in my inbox before I even think about training. I open it on the couch. Push volume is up 9% this week, pull is flat, legs are down because I skipped a session. Bench moved from 100 to 102 and 1/2. Overhead
[17:07] press is flagged as stalled with a note to drop the weight 10% and add a set. Next week's plan is right there, ready to copy into the sheet. 20 minutes of Sunday night planning gone, and the routine ran while I was on the couch.
[17:22] The same shape shows up in a dozen different stacks. A client posts in your shared Slack channel, and the routine spins up a task in linear with the message attached. New lead fills out your Typeform, and the routine drafts a
[17:35] welcome email plus a CRM row. The trigger is the event, not the calendar. That shift is what makes the rest of the stack feel autonomous. Schedules handle the every morning jobs. Event triggers handle everything else, which is most of
[17:49] handle everything else, which is most of the work. Before I wrap, one mistake per feature. These are the ones I see the most, and they are the easiest wins on the way out. Most people live in the default chat and re-paste the same
[18:01] context every morning. Spin up a project for anything you touch more than twice. Drop the docs in once, write a tight custom instructions paragraph, and every chat after that starts smarter than the last one. If you open Claude code in a
[18:14] repo without a claude.md, you are asking the agent to guess your stack, your conventions, and your guardrails. Spend 20 minutes writing one. Four short sections, the three rules from earlier, and the agent stops
[18:28] improvising on you. Teams open co-work and treat it like a shared Slack thread. That misses the whole point. The value lives in the shared project. Pin docs, shared memory, shared custom instructions. Put the context in the
[18:40] project once, and every teammate's chat reads from the same source of truth. A local routine only fires when your laptop is on and awake, which is exactly when you do not need it. If the routine matters, flip it to remote so it runs in
[18:53] Claude's cloud around the clock. Local is for tinkering. Remote is for anything you actually rely on. You type modern and clean, and you get every other AI landing page on the internet. Spend the 15 minutes on a design system first,
[19:08] import a screenshot if you have to, and every output after that lands on brand instead of on the default gradient. Fix any one of these this week and the stack starts paying you back immediately. Five features, one workflow. You think and
[19:24] draft in chat, you build and ship code in code, you collaborate with humans in design, you automate the repetitive parts in routines. That is the system. And if you want the same workflow without piecing it together yourself, AI
[19:39] Master is where we teach all of it with a sandbox on the top models so you can practice the second you finish a lesson. Link is in the description. See you in Link is in the description. See you in the next one.