The argument in one line.
Building faster with Claude Code isn't about learning more features — it's about widening what you can do across domains first, then compounding depth in the one vertical you already work in through a self-improving skill system.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude Code or Claude desktop regularly but keep bouncing between tools instead of building a repeatable system.
- You're a solo builder, freelancer, or team of one who wants a self-updating personal operating system instead of paying for more SaaS.
- You have real work to draw on (calls, docs, past projects) and want a proof-based way to turn that into custom Claude skills.
- You've built a few one-off automations and are ready to turn them into a compounding system instead of isolated scripts.
SKIP IF…
- You're brand new to Claude Code and haven't run a single skill or project yet — this assumes basic familiarity with skills, MCP, and plan mode.
- You want a deep technical walkthrough of MCP server internals — this stays at the workflow/strategy level, not the protocol level.
The full version, fast.
Austin Marchese lays out a six-step roadmap for building faster with Claude Code, split into two halves. Steps one to three build breadth: lock in a fixed AI stack (Claude desktop, skills, plugins, MCP, plan mode) and stop re-evaluating it, run three zero-to-one micro-projects to prove you can do more than you think, then wire up a self-improving system — a raw/wiki folder structure, a resource-ingestion skill that also mines your own Claude conversation history, and a weekly self-review loop. Steps four to six build depth: identify your archetype (builder, storyteller, or systems thinker), build a proof-based skill library across four skill types, then turn it into a personal 'niche command center' — a dashboard that triggers your Claude skills without burning API credits.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:23
01 · Cold open
States the promise: a six-step roadmap, drawn from frameworks used while COO of a $25M startup, to build faster on both technical and non-technical tasks.
00:23 – 01:03
02 · Part 1: Build the top of your T
Introduces the t-shaped-builder framework — breadth across every domain (the top of the T) plus depth in one vertical (the stem) — as the lens for the rest of the video.
01:03 – 02:52
03 · Step 1: Lock in your AI stack
Names the exact stack to stop re-litigating: Claude desktop app, Claude skills/plugins/MCP connectors, and plan mode for interviewing out plan gaps before building.
02:52 – 07:01
04 · Step 2: Expand your domains of execution
Three zero-to-one micro-projects — a calendar-review automation via Gmail/Calendar connectors, a personal 'niche command center' website, and AI-written branding guidelines saved into CLAUDE.md — to prove firsthand you can do more than you think.
07:01 – 11:51
05 · Step 3: Build a self-improving system
Sets up a raw/wiki folder structure as the project's memory, a skill that bulk-ingests outside resources and the user's own Claude conversation history, and a weekly self-review loop run as a scheduled routine.
11:51 – 12:57
06 · Part 2: Build your depth in a specific vertical
Pivots from breadth to depth, arguing augmentation (raising quality) beats automation (raising quantity), illustrated by a five-years-vs-five-days subscriber growth story.
12:57 – 14:54
07 · Step 4: Identify your vertical
Frames three winning archetypes — builder, storyteller, systems thinker — and argues you should go deep on whichever one you already do, not the one you wish you were, checkable via a Claude prompt against your own conversation history.
14:54 – 19:00
08 · Step 5: Build your archetype-driven skill library
Defines four skill types (utility, verification, data enrichment, orchestration), a proof-based build rule (only build skills from work already done), and a refine-immediately habit for fixing skills the moment their output disappoints.
19:00 – 23:27
09 · Step 6: Your niche command center
Upgrades the earlier proof-of-concept site into a full personal dashboard — one-click prompt generation and background-triggered Claude runs — and lists the reasons it beats juggling separate tools and SaaS subscriptions.
Lines worth screenshotting.
- Learning every foundational skill in a new domain used to take years — with AI, most people can get there in minutes and handle day-to-day tasks in that domain on their own.
- A t-shaped builder framework borrowed from software engineering — breadth across every domain, depth in one — is offered as the model for how anyone should approach AI.
- The most common AI failure mode is a 'tool graveyard': trying tool after tool, abandoning each one, and having nothing to show for any of it.
- The fix for tool overwhelm is a fixed rule: pick a stack, stop questioning it, and only reevaluate every three to four months.
- Claude 'connectors' are described as Anthropic's branding for MCP integrations — the same protocol under a friendlier name inside the Claude apps.
- A niche command center running locally uses a Claude membership instead of API credits, which is framed as why it's cheaper to operate than most custom AI tools.
- Automation is framed as producing average output at scale; augmentation — using AI to raise quality while also raising quantity — is framed as what actually compounds.
- Five years of posting 190 YouTube videos produced about 9,000 subscribers; a quality-focused pivot is claimed to have then produced more subscribers in five days than the prior five years combined.
- Every skill someone builds is framed as overhead: unused skills should be deleted and overlapping skills should be merged, not stockpiled.
- Verification skills are cited as having the most measurable internal impact on Claude's own output quality, more than adding skills in general.
- Decision fatigue is framed as 'the new cigarette' — removing unnecessary choices is treated as a direct lever for building faster.
- A functional skill library is sorted into four types: utility (do one thing well), verification (check output quality), data enrichment (pull in external data), and orchestration (chain other skills together).
- Skills are meant to be built only from work already done, not from theoretical projects — proof-based creation is meant to force the library to match a person's real archetype.
- The fastest way to fix a skill is framed as writing the correction the moment its output disappoints you, because the context that made the fix obvious won't exist later.
Six steps to make Claude Code a system, not a chat window.
WHAT TO LEARN
The gap between using Claude Code and building faster with it is a deliberate two-part system: broaden what you can do across domains, then compound depth in the one vertical you already work in.
03Step 1: Lock in your AI stack
- Fix your AI tool stack deliberately — Claude desktop, skills, plugins, MCP connectors, plan mode — and stop re-evaluating it until three or four months have passed.
- Treat MCP connectors as the way Claude reaches outside itself: Gmail, Calendar, Notion, and other services plug in without you having to log into separate tools.
- Use plan mode before any nontrivial build: have Claude interview you for gaps in the plan first, so the actual build goes faster and matches what you wanted.
04Step 2: Expand your domains of execution
- Run three zero-to-one micro-projects — a calendar-review automation, a personal website, and branding guidelines — to prove firsthand you can do more than you think in each domain.
- A calendar-review automation that cross-references email history before a call is a fast way to get comfortable with Claude's Gmail and Calendar MCP connectors.
- Writing branding guidelines into your CLAUDE.md file gives every future AI-produced asset a standard to check itself against, instead of relying on someone else's approval.
05Step 3: Build a self-improving system
- Give every project a raw folder for original source material and a wiki folder that indexes it, so AI can navigate a growing archive instead of drowning in it.
- Build a skill that automatically files new material into that structure, and treat your own past Claude conversation history as a legitimate source of skill ideas and lessons.
- Only automate a task into a recurring routine after a skill version of it has proven reliable — automating something unverified just repeats the same mistake on a schedule.
06Part 2: Build your depth in a specific vertical
- Automation optimizes for more output; augmentation optimizes for better output — chasing volume alone produces more low-value content, not more value.
- A multi-year plateau can flip fast once you shift from quantity to quality — five years of steady output produced less growth than five days after the pivot, by this account.
07Step 4: Identify your vertical
- Pick the vertical you already spend your time in — builder, storyteller, or systems thinker — not the one you wish described you; existing skills are the lowest-friction path to depth.
- Knowing what you're deliberately not focusing on removes decision fatigue as effectively as knowing what you are focusing on.
- If your natural lean isn't obvious, ask Claude to analyze your own conversation history for which vertical you've already been gravitating toward.
08Step 5: Build your archetype-driven skill library
- Only build skills from work you've actually done — proof-based creation keeps the library grounded in real tasks instead of speculative ones.
- Sort skills into four types — utility, verification, data enrichment, orchestration — to see what's missing instead of just accumulating more of the same kind.
- Delete skills you don't use, merge ones that overlap, and fix a skill's prompt the moment its output disappoints you, before the context is lost.
09Step 6: Your niche command center
- A locally-run personal dashboard that triggers your own Claude skills runs on your Claude membership instead of metered API credits, which is what makes it cheap to keep running.
- Build the dashboard to bend to your workflow instead of adapting your workflow to fit someone else's software — the reverse of how most tools work.
- Treat small time costs as investments: spending an extra 30 seconds wiring a task into the dashboard now saves that time every time after.
Terms worth knowing.
- MCP (Model Context Protocol)
- A protocol that lets Claude reach into external services like Gmail, Calendar, or Notion; Anthropic's 'connectors' inside the Claude apps are MCP integrations under a simpler name.
- T-shaped builder
- Someone with broad working knowledge across many domains (the horizontal bar) and deep expertise in one (the vertical bar), a framework borrowed from software-engineering hiring.
- Plan mode
- A Claude Code mode where Claude interviews you to find gaps in a plan before writing any code, producing a more accurate build plan up front.
- Niche command center
- A locally-hosted personal dashboard with buttons that trigger a person's own Claude skills, built to run on a Claude membership instead of metered API credits.
- Zero to one (in a domain)
- Getting a foundational working understanding of a new skill or domain for the first time, typically by completing one small real project in it.
- Skill-driven loop creation
- The practice of only turning a task into a recurring automated routine after it has been proven to work reliably as a manually-triggered skill.
Things they pointed at.
Lines you could clip.
00:00
“Most people know what Claude code is, but they have no idea how to actually use it to build faster.”
12:15
“I'm anti automation. I am pro augmentation.”
12:17
“Automation leads to shitty output. Augmentation leads to next level output.”
14:47
“Decision fatigue is the new cigarette.”
12:32
“I spent five years posting a 190 videos on YouTube, which got me to like 9,000 subscribers... in a span of five days, I got more subscribers than I did in five years.”
16:20
“Generic skills are extremely worthless... the value comes from the personalization of each skill.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogystory
00:00Most people know what Claude code is, but they have no idea how to actually use it to build faster. So, this video, I'm breaking down the exact roadmap that will help anyone use Claude code to build 10 times faster on both technical and nontechnical tasks. The roadmap is six actionable steps that you can implement today that are inspired by the frameworks that I used while the COO of a $25,000,000 tech startup and what I've taught millions of other people like you on YouTube.
00:23So part one is build the top of your t. You wanna start doing more. Well, was at my first job as a software engineer at JP Morgan.
00:30They would stress the importance of being a t shaped developer. And what this means is you have breadth in a lot of areas, but depth in a single area. With AI, the same applies to every person.
00:40You need to become a t shaped builder. So that's breadth in every industry, which is the top of the t, and it's now a 100 times easier with AI. And then you want depth in one specific area, that is the vertical part of the t, and that's where you're an expert.
00:52And in this video, the first three steps we'll go through will help you establish the breadth at the top of your t. And then the last three will identify what archetype you fall into and how to build the depth needed to differentiate yourself. So step one of building the top of your t is lock in your AI stack.
01:07The biggest issue with AI is feeling overwhelmed by the number of tools out there. And this leads to what I call AI tool graveyard, where you have a bunch of tools you've tried, spent time on, and then you just abandon it with little to show for it. So in order for you to build faster, we need to identify our AI tool stack and stop questioning if these are the right tools to use.
01:25Make a decision, use it in your day to day, and then every three to four months reevaluate. My general rule of thumb is I don't need to be first, second, or third to a new tool. I wait until it's so undeniable that I just have to use it.
01:34So here's what we will use and what we won't use. Because explicit do's and don'ts reduces scopes and helps you have laser sharp focus. First, we're gonna use the Claw desktop app.
01:44In my eyes, this is the simplest way to interact with Clawed code while giving it access to all of your local files. You just download the app, select the code tab, and then you select the project you're working on. If you are more comfortable using it directly in the terminal, that's entirely fine too.
01:56The second is we'll be using Claude skills, plugins, and MCP connections. At a high level, skills are repeatable workflows you can have Claude run for you. Plugins are bundles of skills and commands that you install, and MCP connections like Claude reach into external services like Gmail, Calendar, Notion, whatever you're using.
02:12And the beauty of all three is they extend what Claude can do. And the general rule of thumb is we want Claude code to be our interface. We don't wanna have to juggle between a bunch of different tools and logging into different web pages.
02:22One last thing that I consider as part of the stack because it's one of the only universally agreed upon things in the AI ecosystem is using plan mode. In Claude desktop, click next to the plus and select plan whenever you're working on anything robust. Then you could say, task, interview me to identify any gaps or issues in my plan before building.
02:40And I'll then go ahead and create an exact plan to help Claude build it quicker and make sure it's exactly what you want. This is universally agreed upon so it will help you build faster. Now, the first step, set the foundation, but the second step is where the fun begins.
02:52Step two, expand your domains of execution. Think of your output like a factory. Each step requires a specific skill, and every step that you can't complete yourself reduces your ability to get done.
03:03And before AI, learning every skill in your factory took years. So, you had to work with experts called salespeople, lawyers, designers, engineers, marketers. But today, with AI, you can get a foundational understanding in minutes, and then most day to day tasks become doable on your own if you take the time to learn the basics.
03:19And suddenly, you'll stop needing permission and expand the list of things you can complete end to end. And the process of learning foundational skills in a new domain is what I call going zero to one. So how can you go zero to one in different subjects?
03:31The truth is, with any new skill you're learning, seeing is believing. I can tell you all day that you can do this stuff, but until you actually do it yourself, you won't believe that you can actually do it. So no matter what field you're in, these three mini projects will help you go zero to one quickly in domains that matter.
03:45The first micro project is build a calendar review automation. This is to get operational basics. This project will expand your knowledge into operational automations and give you confidence when building out systems from scratch.
03:56In this case, we're gonna set something up that reviews your calendar, checks any email conversations with that person, and gives you an overview of the relationship so you can be prepared for the next call. To do this, first, to Claude desktop, hit plus, go to connectors, go to Gmail and Google Calendar, and then connect them.
04:11For your own understanding, connectors are Anthropix branding for MCP integrations inside the Claude apps. It makes it super simple for you to make these connections.
04:19There's really no need to get boggled down in how it actually works. Then go to Claude and type this prompt in to create a skill called email brief. At this point, when you hit enter, you've created a custom Claude skill that does the same thing every single time.
04:32And suddenly, when you do this, you'll see how easy it is to set up operational automations and the realm of possibilities expands instantaneously. And the reason is you no longer have to ask anyone to do this and you realize how simple it is. The second micro project is launch a personal niche command center.
04:47This is to get the basics of building a website. In 2026, there's absolutely no reason that you can't build your own website. And this project is important because you will see how effective AI is at building websites, and you'll understand that you can now build your own tools from scratch.
05:00And if you are technical, pay attention because there are some things here that I bet you haven't thought through. Type this prompt into Claude and then hit enter, and this will go and create a website and launch it for you. But there is something really important here that I wanna call out.
05:12We're able to call the email brief skill and run that from our machine. We're essentially creating a command center that can let us interact with Claude skills we've created without spending a ton of money on Claude API credits. This opens a door where you can create a command center where you can trigger these automations to run automatically for you.
05:29We'll get into this later in this video, but this is super powerful. Now, all of a sudden, you don't need anyone to help you make a simple website and you can start thinking about making your own personal command center to help you with your day to day tasks. Again, I will cover this later when we go through the niche and the archetype that you fall under.
05:45The third micro project is create branding guidelines. This is for basic branding and marketing. Now, no matter what you do, whether you're an employee or a business owner, branding is critical.
05:54I personally would hate spending time on something and then have someone say, that's not on brand or this doesn't follow our required guidelines. So to help mitigate this so you can start building and getting things actually approved quicker, let's start thinking about it proactively. We're gonna update the niche command center that we created in micro project number two by writing this prompt.
06:11This will then go ahead and create a branding page that you can reference in all future projects to make sure that whatever you're doing is on brand. What this will do will create a branding guidelines and update the Claude dot m d file. And by establishing clear guidelines and updating the Claude MD, you're gonna constantly remind your Claude code system whatever your branding guidelines are, so anything it produce follows these guidelines.
06:31Suddenly, you don't have to ask for branding and design help. Once you do these three micro projects, you are going to have a different swagger about what you can and can't do. And recently, I actually sent these to one of my buddies and he sends me this text.
06:44He said, I guess it's my first true experience with it and it's really blowing me away. Seeing and doing is believing. And until you do it yourself, you're just not gonna believe that you're capable.
06:53Now, at this point, you've proven to yourself that you can do more than you thought. But as you do more, you need to create an environment that continuously improves over time. That brings us to step number three, which is about building a self improving system.
07:04In order for us to truly build faster, we need to set up a system that learns and doesn't make the same mistake twice. . Ninety nine percent of people using AI have a leak in the bottom of the bucket.
07:15They use it and context fills up, but over time, the bucket slowly drains and they end up having to repeat themselves or AI continues to make the same mistakes. We need to fix that leak so the system learns automatically and self improves over time. Now, have a video on my channel where I dive deep on this, but here are the three most important features that you need.
07:32The first is you set up the project structure. You need to give every output you produce a home. We need to create a project that you can easily store data so that it can capture specific contextual information based on whatever you're working on.
07:43To do this, open ClawedCode and make sure in the folder where you created the mini projects from step two, and then paste this prompt in. As part of that prompt, it will explain what each folder does and why it's needed. But in short, we have a raw folder that takes in any raw information that you want.
07:57This could be a call transcript or a YouTube transcript. This is the original source of truth. And then it creates a wiki, which is basically a table of contents, which tells AI what is in each raw file.
08:07A mental framework here is slash raw is holding the entire book, whereas slash wiki is the table of contents so AI can quickly navigate. The second feature is bulk data ingestion. We wanna fill this project with data that we've already done so it can learn quicker.
08:20And so first, we're gonna create a Claude skill that systematically adds new resources so every new file gets handed properly. Here's a prompt for a skill called add new resource. This skill will bring any raw file into slash raw, then automatically process it into the wiki folder.
08:34And one question you may have is what resources do I add? Really, anything specific to what you're working on. So this could be blog posts from people you trust, call transcripts from your manager, whatever.
08:43But you don't wanna just stop at other people's materials. This is your system, so you have to make it you. To do this, what I do is I record myself constantly.
08:51Voice memos, end of day rambles about what I worked on, and this add new resource skill will take those rambles from you, your lived experience, and then ingest it into the system so it can constantly improve. So that pulls an external data, your own data, but there's one other piece of data that you have to actually look at.
09:07Claude Code desktop stores all of your conversation history locally in a file. And what this means is that Claude Code can look at your past conversations, identify ways to improve the system, and suggest custom skills while surfacing any information that's worth ingesting.
09:21So what you can say is slash add new resource, bring in my clawed conversation history, and process any learnings from it. Suggesting improvements, custom skills to create, or lessons learned that we can add to our system. This now creates a clear way for you to bring in data in bulk, uploads it with the most valuable information, your actual conversation history.
09:38Feature number three is self improvement loops. So we set up the structure, we've ingested data, and now we need to make it so it continuously analyzes data and improves the system. To do this, we'll follow skill driven loop creation.
09:48Essentially, anything that we make that we wanna run automatically or in a loop has to be derived from an existing skill. The reason for this is before you automate anything, you need to make sure it actually works consistently, and skills are the best framework to do this.
10:01So for a self improving loop, we need to create a skill called improve system. To do this, write this prompt into Claude, will create the skill for you. This skill itself will look at your past history, look at your conversation, and suggest improvements.
10:14I have a version of this in my plugin build partner.ai if you want a head start on setting this all up. Once you have that, you can go to Claude desktop, select routine, and then it creates a routine that calls this skill every Friday. And the beauty here is that if you update the skill, the routine itself that automatically triggers will update automatically.
10:31In the case of this routine, if you wanna create a system that's fully automatic, you could do slash improve system, auto approve every suggestion. Personally, what I like to do is manually review improvements to make sure that the system is self correcting in the right direction.
10:44So what I'll do is I'll say slash improve system, create a file formatted for Obsidian that I can approve or deny changes. And Obsidian's just a way to view these files locally that's entirely free. So thus far, have covered lot.
10:55We've established our AI tool stack. We've outlined three micro projects that can help you expand the top of your t so you can start doing more, and then we've laid the groundwork for a self improving system. But the reality is just having the top of the t in the AI world isn't enough.
11:10You have to have depth. And before we dive into the three step process to know what to double down on. If this is your first video, welcome to the channel.
11:18But if this is your second or more, here is our anti slap agreement. The visuals, the testing, the me writing on this piece of paper, this is all for humans, not for AI scrapers. So all that I ask from you is that you subscribe as part of this agreement to help this content reach more people so that I can keep making videos like this.
11:35Also, I do wanna congratulate Tom Brooks twenty four for winning our Claude Max giveaway. He's building an automation system for his construction business, absolute legend.
11:43To enter the next giveaway, comment below with what you're building or a recent feature that you made. And, yes, each video that you comment on gives you an extra entry to this giveaway. Part two is build your depth in a specific vertical.
11:54One of the biggest issues with how people think about AI is they always wanna do more. More social media posts, more code written, more websites, more, more, more. But the reality is that's only one side of the equation and it's something that doesn't really interest me that much.
12:08What interests me is how AI can be used to improve the quality of the final product. And I'll go even a step further. I'm anti automation.
12:16I am pro augmentation. Automation leads to shitty output. Augmentation leads to next level output.
12:21And in a world where you can do infinite things with AI, there is really zero advantage of just cranking out more slop. The advantage is increasing quality while you increase quantity.
12:31A personal example of this, I spent five years posting a 190 videos on YouTube, which got me to like 9,000 subscribers. Right? I had the quantity.
12:39But then in the past six months, I shifted focus and I pivoted to quality. Recently, in a span of five days, I got more subscribers than I did in five years.
12:47In the world of AI, there is a surplus of quantity, but there will always be a shortage of quality. So how do we double down on what we do best to 10 x our output in a specific vertical? Step four is identify your vertical.
13:00The first thing you need to know in going deep is where to go deep. And there are two ways to do this. Right?
13:04The first is by the archetype, what you identify as. And the second is by auditing, like, do you actually do? In terms of your archetype, there are three types of people that are winning in the AI era.
13:13And most successful people will land in one or a combination of two. The first is the builder. This is someone who ships products, automation tools.
13:20The second is the storyteller. This is content narrative, brand distribution.
13:25And the third is the systems thinker. This is processes, workflows, ops, infrastructure. This is anything that turns one good run into a repeatable system.
13:33For me, I'm primarily focused on storytelling. For my business, this YouTube channel and the content comes first for me. And, yes, I do spend a ton of time building out systems and creating products for businesses I work with.
13:43But right now, I'm primarily focused on storytelling vertical. So the natural question that you may have is which vertical are you? Well, to start, what are you currently doing the most?
13:51If you're a software engineer, you're likely a builder. If you're a marketer, a storyteller. If you're a project manager, a systems thinker.
13:57The trap that most people fall into is they focus on things that they wish they were, but they don't actually spend their time on it. Instead, if you're a software engineer, get really good at using AI for software engineering. Same for marketing, same for operations, etcetera.
14:09These skills are what got you this far, and it's the lowest barrier of entry to get deep in that vertical. So those are the types of archetypes and which to lean into. But if it's not yet obvious, let's look at the data.
14:19Go to Claude and type this prompt. It will look at your conversation history and then identify which verticals you have naturally biased towards. If you don't know where to start, start where your energy has naturally been drawn to.
14:29At this point, you know which archetype you lean towards. But why does this actually matter? Well, this builds on something I said in step one.
14:34Sometimes it's just as effective to know what you're not working on as it is to know what you are working on. So once you've labeled yourself and identified where you wanna focus, you know what you don't wanna focus on and that helps remove decision fatigue. .
14:49Remove decisions and you're gonna start building faster. Now at this point, you're ready for the next step. Step five is build your archetype driven skill library.
14:57Based on your archetype, you're going to want to build out your skill library. And to do this, we're going to take a proof based skill creation approach. Only build skills from work you've already done.
15:07And this makes sure that every skill you create is grounded in real work, not theoretical. And it also forces you to converge on the archetype you identified in step four. You can't build storyteller skills if you're never doing storyteller work.
15:19And so by default, you're gonna start focusing on what you're actually working on, not science projects someone told you would help change your life. And while you work on these skills, there are four different skill types to keep in mind. The first is utility skills.
15:30These are small reasonable skills to do one thing very well. For example, draft email response could be a utility skill for drafting email responses. Second is verification skills.
15:38These are skills that check the quality of the final output. From Claude directly, verification had the most measurable impact on Claude's output quality internally. The third is data enrichment skills.
15:48These are skills that pull external data into our system. For example, I have a funnel digest skill that pulls traffic data from all of my websites. The fourth is orchestration skills.
15:57This is a skill that chains multiple steps together. For example, you could have a skill that says end of day review, which could run the slash funnel digest skill and the slash draft email response skills as part of an end of day checklist. So you understand the four categories, but how can we start creating skills that are specific on what we're actually working on, in turn actually helping us build faster?
16:15First, audit your system to suggest new skills and fix existing skills. Use this prompt which will map your skills into categories and give you an understanding of what you're already doing and suggest improvements plus new skills from your past conversation history.
16:30You have to know where you're at to know where you should go. And before you just go buck wild creating a ton of skills, every skill you make is overhead for your system. So if you aren't using skills, just delete them.
16:39If you have overlapping skills, combine them. So once you create these skills, it's time to refine them over time. In my eyes, generic skills are extremely worthless.
16:46And this is why I kinda roll my eyes when someone says that they're sharing hundreds of skills with you, this whole folder with hundreds of skills. The value comes from the personalization of each skill. Austin Lau, who runs Growth at Anthropic, said it best about generic skills.
16:58It's good scaffolding, but you still need to spend the time to right size the template to your stack, your edge cases, your workflow. So what you need to do is use the skill, identify holes, and enhance it over time. Using a skill we actually created earlier in this video, whenever I run a skill and I find that the output wasn't exactly what I wanted, I'll write this in a clause.
17:15This will take the skill that I just used, the back and forth to get a output that I wanted, and it will suggest ways to improve the skill to get a better final product the next time I use it. And a key here is that I try and reduce the time for every feedback cycle. If I use a skill and it's not perfect, update it right then and there.
17:31Don't wait because you're gonna forget or your self improving system isn't gonna have all the context that you have at the current moment. I'm getting jazzed up about this, but this stuff is so important. Now the third thing, after you've audited your existing skills and have refined them over time, you need to enhance skills so that they can verify the output.
17:47There are a number of ways to actually do this, but here are my favorite ways to verify outputs. You can update your Claude MD with explicit verification language. So you can have it say, before returning any work, verify that it works and the task is complete.
17:59And if you can't verify the result, fix and rerun. The second is that you wanna connect any external services your skill use. For example, if it deploys an app, connect wherever it's deployed so it can verify successful connection.
18:10Here is a prompt that you can see for which MCP servers would be valuable for verifying any of the existing skills that you use in your setup. Think of this like saying, hey. Here are my skills.
18:19What are ways that I can improve my system to empower them to verify the output they produce? And the last for nontechnical stuff, I use buildpartner.ai to verify my outputs against experts.
18:29For example, if I want Gary Vee to review my content strategy, I call slash b p expert advice to review the response. So in this step, for the skill library, we've identified how you can determine what skills to work on, how you can actually improve those skills, and then ways that you can help the skills to verify the output.
18:44Before we get to this last step, at this point, we've established the top of the t. We're able to tackle tasks in different verticals, and we've also gone deep started creating custom skills that are proof based with verification baked into them. Now, we need to create our own interface that helps streamline everything we do, which is step six, your niche command center.
19:02In step two of this video, we briefly created a niche command center, which was a local web site with one button that runs a skill. That was a proof of concept just to showcase the power. In this step, we enhance it for the vertical you spend the most time on.
19:14Now, if you're wondering why this is so powerful, there are four key reasons. The first is that you're creating a fully customized workspace. Every section, every button, every feature you build is designed around your workflow.
19:25This is the opposite of how you normally use software that helps you in your day to day. So if you're using another tool, you're bending your workflow to make it work within that ecosystem. In this case, you're actually creating a tool that bends to fit you and your exact needs.
19:38For me, this meant creating a YouTube dashboard with every single feature that I need, and I don't have to go to external services. Instead, it's all in one.
19:46I've shaped it to my exact business and the exact needs that I have. The other benefit of a niche command center is it uses clawed membership instead of your API credits. Most AI tools that you create cost a ton of money to run because they hit the API on every request.
20:00And as somebody who's built tools for thousands of users, this change of using your account instead of API usage can really save you a ton of money. The third is that you can dynamically create prompts based on dashboard data. So here's what it looks like in practice.
20:13So on my niche command center, I have a copy prompt button. I click that, and I'll get an exact prompt with the context from the dashboard loaded. So I can maneuver this dashboard and change the data on it to generate an optimized prompt that I can then bring into Claude code directly to run.
20:30I copy it. I paste it in Claude. I hit run, and it streamlines everything I do.
20:34The key here is I'm making it to help me work faster. I don't care about anyone else. And this alone can 10 x your output.
20:41The more you dive into your vertical, the better you're gonna start getting in it because this niche command center is going to unlock your productivity. The fourth is that you can trigger Claude to run-in the background on a single button click. So the dashboard can become the orchestration layer that you can then manage to trigger these agents.
20:57When I use this, I visualize being in, like, Star Trek command center and you're clicking buttons to trigger these clawed agents to spawn for you. And the fifth benefit of this is it can save money on unnecessary SaaS products. In my example, there are tools out there that I was paying for to visualize thumbnails in the context of YouTube.
21:13Instead of paying for it, I just built that feature into my niche command center and now I'm saving money. There are probably a bunch of different tools out there that you're using that you're paying for that you could probably just recreate. Now, this video to expand on this niche command center, I'm gonna give you a structure to help you build this all out.
21:27In step two, we built the shell, but now we wanna build the features. Type this into Claude, which will create a list of all of your skills plus an input field where you can type and build a custom prompt. Once you have this run, and here you can see it running on my system, it'll create two buttons.
21:41One that you can click and it'll grab the prompt that you can then bring into Claude code, and the other will trigger Claude in the background, which you can then watch the output stream on your screen. This is like seeing the Claude code agents running that you just triggered. Now these are foundational tools that I use for clients that I work with, my own internal operating system, and the this establishes the foundation that you can build on top of.
22:02So let's say you're building this out and you wanted to have something that pulls in your Google Calendar, you could just say, add a tab to the page that is synced with my Google Calendar and show me my upcoming meetings. You would hit enter and then it'll build for you or ask for any information that you need to provide it.
22:15And because it's all running locally, you don't have to worry about any of the technical challenges like deploying an app. And depending on the vertical that you're diving deep on, on screen, I show each of the archetypes and different niche command centers that may inspire you. So if you're a storyteller, a builder, a system thinker, so use these as a source of inspiration to drive whatever you're working on.
22:34And as you're building out this niche command center, a question that you may have is, I can just do this quicker manually myself than adding it to the center. Sure. Maybe.
22:43But this is where you want to lose the battle but win the war. If a task normally takes you thirty seconds and it takes maybe sixty seconds to add it to the dashboard so that it's there forever. If you get 1% better every day, it'll be crazy how much better this command center will be for you.
22:56I can't stress how impactful my command center has been for my business as we start to take off. At this point, you know exactly what it means to be a t shape builder, and you know how to build 10 times quicker. The theme is get good at many things first by going zero to one.
23:10Then get great at one archetype, and that archetype should be the one you already are, not the one that you wish you were. And if you like this video, you'll love this video where I dive deep on Claude's skills. Everything I cover there builds on what we just covered in this video.
23:22And if you pair the two, you will build 10 times faster. I'll see you in the next one.
The Hook
The bait, then the rug-pull.
Most people know what Claude Code is but have no idea how to actually use it to build faster — this lays out a two-part, six-step roadmap: build breadth across domains first, then compound depth in the one vertical you already work in.
Named ideas worth stealing.
00:23concept
T-Shaped Builder
Breadth across every domain (top of the T) plus depth in one specific vertical (stem of the T), borrowed from a software-engineering hiring framework.
Steal forframing a personal skill-development or career-positioning plan
13:10list
Three AI-Era Archetypes
- Builder
- Storyteller
- Systems Thinker
The three roles claimed to be winning in the AI era; most successful people land in one or a combination of two.
Steal fordeciding which AI skill to specialize in first
15:20list
Four Skill Types
- Utility skills
- Verification skills
- Data enrichment skills
- Orchestration skills
A taxonomy for auditing a personal skill library: small single-purpose skills, quality-checking skills, external-data-pulling skills, and skills that chain other skills together.
Steal forauditing which category of automation is missing from your own setup
20:04list
Reasons a Niche Command Center Beats a SaaS Tool
- Fully customized workspace built around your own workflow
- Runs on Claude membership instead of metered API credits
- Dynamically generates prompts from your own dashboard data
- Triggers Claude to run in the background from a single click
- Replaces paid SaaS point-tools you'd otherwise be renting
The case for building a personal dashboard instead of adopting more off-the-shelf software; the on-screen card reads '4 Benefits' but a fifth ('saves money on unnecessary SaaS products') is added verbally right after.
Steal forjustifying time spent on internal tooling instead of subscribing to another app
How they asked for the click.
VERBAL ASK
11:42subscribe
“All that I ask from you is that you subscribe as part of this agreement to help this content reach more people so that I can keep making videos like this.”
Delivered as a comedic 'anti-slop agreement' bit right after pointing at the handwritten whiteboard work as proof it's made for humans, not AI scrapers, then immediately rolls into a giveaway plug (comment what you're building for an extra entry) — softens the ask by wrapping it in a joke and an incentive instead of asking cold.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
