The argument in one line.
Build a single structured database as your business operating system, then use Claude Code routines as autonomous agents that read, decide, and act on your workflows—making your infrastructure compound with every AI model upgrade while you only approve or reject decisions.
Read if. Skip if.
READ IF YOU ARE…
- A solo founder or small team operator running 2-5 active business workflows (content, finances, ads, email) who wants them automated without hiring.
- An engineer or technical founder with 1-2 years of coding experience who's comfortable with databases, APIs, and Claude Code but tired of manually triggering tasks.
- A creator or agency owner generating enough content and revenue to justify 9+ months of infrastructure setup in exchange for hands-off operations at scale.
- Someone already using Supabase or similar structured databases who wants to see how AI agents can be wired into existing data rather than building isolated automations.
SKIP IF…
- You're a non-technical founder with no database or coding experience — this requires hands-on architecture work and assumes comfort reading SQL and API documentation.
- You need something operational in days or weeks — the speaker explicitly states this takes nine months of compounding work and iteration, not an afternoon project.
- Your workflows are one-off or episodic rather than recurring — this infrastructure pays off when you have 5+ daily/weekly/monthly processes that would otherwise demand manual attention.
The full version, fast.
Features rot, but infrastructure compounds � so stop chasing point automations and build one structured database that every AI model you'll ever use can read from. The method: centralize every workflow (inbox, finance, content, ads, CRM, analytics, social) as rows inside a single Supabase project, pull external data in through edge functions, then schedule Claude Code routines that act like employees � triggered on a cadence, equipped with MCP tools, memory, and write-back access to the same tables. You stay the human approver via a dashboard and Telegram pings. Practically: identify your repeating workflows, force their data into one schema, and let proactive coding agents read, decide, and draft so morning admin collapses from an hour to ten minutes.
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 – 00:29
01 · Cold open + promise
Hooks with 'my business finally started running itself' and promises you'll learn how to build the same.
00:29 – 00:46
02 · Disclaimer + nine-month truth
Honest reset: this took nine months. Features rot, infrastructure compounds.
00:46 – 01:50
03 · Four-layer stack reveal
Supabase Postgres, Edge Functions, Claude Code routines, pg_cron — the brain is the new top layer.
01:50 – 02:02
04 · Eight surfaces, one database
Inbox, content, competitor watch, finance, ads, CRM, Azure ops, social — every workflow is a row in a table.
02:02 – 03:06
05 · Anatomy of a routine
Not an API call — a colleague that clocks in. Trigger → Spin up → Read → Act → Write. Edge functions move data; routines do the thinking.
03:06 – 03:55
06 · Content engine: this video came from this loop
Mine → Score → Outline → Script → Package. The slides he just showed were generated by a skill in one prompt.
03:55 – 04:50
07 · Dashboard walkthrough — finance
Revenue, expenses, audience, agenda, inbox, monitor — all live tiles fed by Supabase, conversable with Claude.
04:50 – 06:05
08 · Content board + packaging demo
Pastes a YouTube URL into Claude, gets three thumbnail/title packaging options back instantly.
06:05 – 06:52
09 · Mobile app — promote a post
Same dashboard as a native iOS app; demos promoting an Instagram post for €10/7 days from his phone.
06:52 – 07:28
10 · Unified inbox — approve & send
Every email, partnership pitch, and YouTube comment becomes a drafted reply with one-tap approval.
07:28 – 08:05
11 · Infrastructure tab + soft pitch
Azure resource health, then first plug for build-loop.ai/founder-os done-with-you offer.
08:05 – 09:05
12 · Behind the scenes — 15 routines
Pops the hood on Claude Code: each row is a scheduled employee. Why routines beat trigger-then-LLM workflows.
09:05 – 10:12
13 · Morning review — LinkedIn carousel, IG story, YouTube post
One screen with the day's auto-drafted social posts; he reviews and clicks Post Now.
10:12 – 11:00
14 · Core lesson — context is the moat
Identify workflows, store data in one place, AI quality compounds with every model release because context is already there.
11:00 – 11:48
15 · CTA + sign-off
Pitches a call via build-loop.ai/founder-os, asks for like/subscribe, sign-off.
Lines worth screenshotting.
- Features rot but infrastructure compounds: a Supabase database plus Claude Code routines gets better with every model release without the founder touching the system.
- When every workflow becomes a row in one structured database, the AI can read, decide, and write back across all business functions with a single session.
- A founder OS built over nine months is not replicable in an afternoon — the value comes from the accumulated context the system has built about the business over time.
- 15 Claude Code routines acting as employees across inbox, finance, content, ads, CRM, and social eliminates the manual coordination tax that consumes a solo founder's day.
- pg_cron and Edge Functions running on a schedule create a business that triggers autonomously — the founder's role shifts from executor to approver.
- The human-in-the-loop model (accept or reject, not initiate or execute) is the correct abstraction for a solo founder who wants leverage without losing quality control.
- Telegram notifications for exceptions mean the founder's attention is summoned only when something goes wrong, not consumed by monitoring what's going right.
- Embedding content with pgvector and storing it in Supabase allows the AI to search across all previous YouTube output to identify patterns and generate better packaging options.
- An iOS app that wraps the same Supabase data into a mobile interface means the founder's dashboard is accessible without switching to a laptop.
- Infrastructure ownership means you are never locked into a third-party tool that can sunset, reprice, or deprecate its API — the system lives in your database.
- A Claude Code routine that generates slide decks from a single prompt is a creator-facing productivity tool that compounds as the prompt gets better over time.
- A founder who builds their own OS rather than subscribing to a SaaS stack accumulates institutional knowledge inside their system rather than inside a vendor's database.
Steal this stack — the brain is the layer you don't own yet.
Founder OS playbook for Joe
MCN already has 138 Supabase tables and a Contabo VPS — you've owned three of the four layers for years. The unlock is naming the fourth layer (Claude Code routines), turning every JACE/REESE/SAGE/RYDER agent into a row, and treating that table as the org chart.
- Open a `routines` table in MCN's Supabase tomorrow. One row per agent (JACE, REESE, SAGE, RYDER, plus the unnamed ones). Schedule + job description + last-run + status. That table IS the MCN+ marketing asset.
- Adopt the five-step anatomy as the standard agent lifecycle: Trigger → Spin up → Read → Act → Write. Every Paperclip agent should fit this shape, including the Telegram-style ping at the end.
- Borrow the eight-surface grid for the MCN+ sales page hero. Yours is bigger (340K LOC) but the message is identical — kill the SaaS sprawl, one DB to rule them all.
- Build the content loop yourself, scored: Mine → Score → Outline → Script → Package. Joe currently mines and packages, but doesn't score against pillars before drafting. Add the score step.
- Steal the 'features rot, infrastructure compounds' line directly. It's the philosophical center of the $6 Stack pitch. Put it on a sticker.
- The CTA pattern is gold: $10K done-with-you at the top of the title, ‘helping three other entrepreneurs' scarcity, three CTA beats spaced 4 minutes apart. Mirror this for the LFB Line.
- Killer self-referential proof: 'this video came out of this loop.' Joe should record one MCN demo where the demo itself is the artifact the system produced. Trust skyrockets.
Terms worth knowing.
- Supabase
- An open-source backend platform built on PostgreSQL that provides a hosted database, authentication, storage, and serverless functions — used here as the central structured database that stores all business workflows and data in one place.
- edge functions
- Serverless functions that run close to the end user on a distributed network — used here to fetch data from external services (email, analytics, finance APIs) and write it back into a central database without a traditional server.
- pg_cron
- A PostgreSQL extension that schedules SQL queries or functions to run automatically at defined time intervals — used here to trigger automated workflows (every five minutes, hourly, weekly) directly inside the database.
- Claude Code routine
- A scheduled, recurring Claude Code session configured to run autonomously at set intervals — acting like a persistent AI employee that reads data, makes decisions, and writes back to a database without manual initiation.
- MCP tool
- Model Context Protocol tool — a standardized interface that allows an AI agent to call external APIs and services (email, calendar, database, social media) in a consistent format, enabling the agent to take real-world actions beyond just generating text.
- Founder OS
- A personal operating system for a business owner — a custom-built stack of connected infrastructure, automated routines, and a unified dashboard that replaces scattered SaaS tools with one structured database the AI can read and act on.
- human in the loop
- A system design pattern where an AI agent completes the work and surfaces a decision or draft, but a human reviews and approves or rejects each action before it executes — maintaining oversight without requiring manual execution of every step.
- embeddings
- Numerical vector representations of text or data generated by an AI model — stored in a database to enable semantic search, allowing a system to find conceptually similar content rather than only exact keyword matches.
- LinkedIn carousel
- A multi-slide document post on LinkedIn that users swipe through — commonly used for educational content, frameworks, or summaries, and referenced here as one of the automated content formats generated from existing video transcripts.
Things they pointed at.
Lines you could clip.
00:00
“My business finally started running itself. And for the first time in history, I think yours can too.”
00:36
“Features rot, but infrastructure compounds.”
02:02
“This is not really just a simple API call. This is really like a colleague that clocks in.”
03:06
“This video came out of this loop.”
08:26
“Every single row in here is basically an employee.”
10:27
“The single most important thing that makes the output of AI high quality is context.”
10:45
“Every model that gets upgraded, the output gets better because the context is already there.”
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
00:00My business finally started running itself. And for the first time in history, I think yours can too. And in this video, I will walk you through every single detail.
00:08And at the end of this video, you will know everything you need to know to build the exact same for your business as well. So, this is my dashboard, but first, I've made some slides to make things a bit more clear, so my business actually runs itself. And this is all done by code.
00:23And because of Cloth, this is finally possible. But this is something I've been working on for nine months already. Also, little disclaimer, this is not something you can create in just an afternoon.
00:33This takes a lot of time. So, the whole thing is .
00:39With AI becoming better and better, and with the amount of features that Claude is shipping, this all compounds with your infrastructure. And what I mean with infrastructure is the following. I have this super based database, and this contains a lot of different tables containing everything.
00:54So, it's a database, it's edge functions, it's cloth code routines, which I will show in a moment, and everything runs automated. So, I don't do a lot of stuff manually. I am only the human in the loop and I just accept or reject something.
01:07Every workflow I had is now a row in my table, in my super base right here. What I mean with that is I had a lot of emails coming in daily. I need to create a lot of content, both new ideas and repurposing.
01:19I need to know what my competitors are doing, so I have competitor watching. All my finances, my income, and my expenses are also in my super base. Then, I run ads, so all my ads are also run by Claude.
01:30Then, I have clients, so I have a CRM for that, so I can keep track on all the specific income that I get because of a client. Also, I run stuff in Azure. That is basically a place where you host code and I needed better insights on everything that was happening inside of Azure.
01:44Also, I post on YouTube, Instagram, TikTok, Twitter, LinkedIn. That was also a lot of manual work. So all of that is automated.
01:51Like, I don't need to open the apps anymore. So this is basically the infrastructure that I'm talking about. So this is not really just a simple API call.
01:59This is not something on the surface.
02:03This is really like . So, there's a trigger. So, I want stuff done every five minutes, every hour.
02:10I want something done once a week. We have the trigger. It spins up a ClothCode session because .
02:17It is not limited to the code we just wrote. It can read everything, so it uses specific Cloth skills. It can access all the stuff in my super base right here, and it uses the memory that it stores over time.
02:30Then it can act, So it has access to every single MCP tool I need. Think about my Gmail. Think about my calendar.
02:36Think about my Supabase. Think about Higgs field for thumbnail generation. And then it writes back to Supabase and Telegram.
02:43So, on my phone, I just get pinged whenever something goes wrong. So, here's the thing. We have add functions in Superbase, they get all the data.
02:50So, it gets all my emails, all my Revolut data, all my YouTube analytics. Everything is pulled out of the external apps. Then I have routines.
02:58So, for example, draft my emails, come up with video ideas, and build up context about me and my business over time. So, for example, this video, the idea, the context behind it was made with AI. Like, this is still me.
03:11I'm a real person, and these slides are created with just one prompt because I created a skill that makes those slides on autopilot. So what happens? I have sources.
03:20So, for example, my revolute, all my expenses, all my income, it is fetched into my super base right here and then shown right here. And for now, I blurred my financial. Like, that is the only thing that I'm blurring in this video.
03:32Right here, I have my revenue. I have my audience on all the different platforms. I have my agenda.
03:37I have everything in my inbox coming from YouTube comments to my email inbox, to community posts where I need to react on. Then I have content. I have the monitor.
03:47I have to do list. So I will now just show you everything in detail. This is the main dashboard, but I also have a very simple to do.
03:54I have finance, where I have the revenue this month, the expenses of the month, the averages, and everything I need to understand, but also my revenue in the last years.
04:04So now that I have this infrastructure, I can actually just have a conversation with Claude, it can help me understand what is happening with my finances. So, I have everything by purpose.
04:14So, what I spent on travel, what I spent on my team, what I spent on SaaS, what I spent on private, on ads, everything is in here. And I can also invite my team members to access this. Then I have content.
04:27So, right here, I have this content board storing all my content, and it compounds over time. Whenever I post a YouTube video, it gets the YouTube video with the transcript, and the AI actually start to understand my type of content better. It has everything, automated.
04:41So it has the hook, it has the promise, it has the proof. But what I can now do is, for example, I can get the URL right here. So I just copy the URL, I go to my Cloud, I paste the URL, and I just say the following, please create three packaging options for those, for this video with thumbnails.
05:02Like, this is a short, so that is not the best example, but I can show you, for example, this one which actually has thumbnails that I can just use immediately. I have multiple packaging options right here, but those are really good already. So, everything is just automated in my Cloud Code and it can access my infrastructure right here.
05:18So, what happens is I have a table, for example here, called YouTube videos, and I have everything right here. All my content with descriptions.
05:28I have the duration. I have the key topics. I have embeddings.
05:32So I can actually search through all my content, and everything shows in here. So, I have my telegram, which is just texting me stuff that I need to know about, and this is my main thing. And the nice thing about this is, and I will show you that directly, right here on my phone, I actually have an iOS app.
05:49As you can see, this is my logo. I can now click it, and right here, I have everything. So it's the same.
05:55So what I can do is I can go to content, and right here, I can just edit whatever I want. So this is just a a side note, like it's an iOS app as well, and it just has everything I need. So, for example, the calendar, but also my analytics right here.
06:09How is everything growing, but also my meta spends? So I do promotion for Instagram, for example, I can then see everything from here. What I can also do is the following.
06:18I want to promote something. I can click right here and I can just promote it. So I can do, for example, €10 for seven days.
06:25I want engagement, and I can now just promote it like this. It has been promoted. As you can see, it is live.
06:31I can now click it, and right here, I have the actual post that is being promoted. Then I have my inbox, and honestly, this is the one that is most important to me. I was getting a lot of emails for partnerships, requests, client contact.
06:45Everything is now in here, so I have one unified inbox for everything. So, for example, this is called outreach to me about a video that I created and they want to give me a chair. Now, I have this reply.
06:56Not a fit right now. I don't do sponsored chair integrations. Thank you for the note though.
07:00I can now click on approve and send and it will actually send the email. But also, for example, right here. On my video, I got a comment.
07:07So, it's someone saying, if you're watching this, this is how I answered your comment, but I've been programming for fifty years in the vibe coded programs I've made so far, not follow this pattern. You just iterate slowly. Okay.
07:19Nice. That is crazy. Fifty years is a long time.
07:21I can now approve and send, and it will actually respond to the comment. I also have infrastructure. So, right here, everything that I'm running, for example, for build loop right here, is right here shown, so everything is clear to me.
07:33Also, if you want to know more about how to make this, make sure to check out the link in the description. At this point, I'm helping three other entrepreneurs build this for themselves. Whether you're a solopreneur or someone with a very large business, you can now start automating every single workflow that you want.
07:48So, that is amazing. Make sure to check out the link in the description. Also, if you enjoyed this video, make sure to like and subscribe.
07:53Okay. Let's go on. So, I think by now you really want to understand how I created this.
07:58Right? So, this is basically just a web app, but a web app is really simple to create. Right?
08:02It's just you show data, it comes from the super base, we have all these tables, it stores everything. But right now, I want to show you what's behind it because we can now go to Cloud right here and it shows routines. If I click this, right here, you see 15 routines.
08:18And this is exactly where the intelligence come from. How you should see a routine is . So, for example, email drafts twice daily.
08:28That is exactly what is answering my emails right here. Build loop YouTube sync. So, it checks for new comments and writes responses to that.
08:36But the whole thing is where we previously had a trigger, we then had a large language model, and we had some output, we still trap the large language model into a, then b, then c. It is not really an agent, and that was exactly why these workflows were no employees yet.
08:52Right now, when you use a routine, what is actually happening is it is spinning up a ClotCode session. So, clot code is super proactive.
09:00What I mean with that is if something goes wrong, it won't just break down and say, sorry, I couldn't do that. What it would actually do then is, hey, I tried to access this URL, I found that it didn't work, I will try another URL to eventually achieve the end goal.
09:14That is the big difference. So, for example, when I now go more in-depth in the app, I have the start morning review. Like, the whole walkthrough, this is the actual part that brings everything together even more.
09:25So, what happens is I have a LinkedIn carousel completely automated for me. So, it look looks like this. Images are added.
09:33Nice. Looks good. So, what I now want to do is I can click on post now.
09:38I have the actual text right here. So, this is really cool. Right?
09:42I don't even come with the ideas. So, it uses all my content from my YouTube, from my other social media accounts, and from there, gives me the ideas and already executes them. I can now just click a YouTube community post, an Instagram story, so the day I stopped blaming the model.
09:58Looks pretty good. What I can now do is I can click on post now. And as you can see, I can now go to the Instagram, and right here, you will see Build Loop.
10:08When I check the story, here we have our story. So, this is all automated, but the main point about this is you have a business, probably. You have a lot of workflows that you could automate.
10:19Identifying what workflows you need, identifying what data is important to your business, storing that into one infrastructure, aka a super base database, is the best decision you can make for the coming two years.
10:33The single most important thing that makes the output of AI high quality is context. We give it context by having our second brain, our digital twin, whatever you want to call it. Because we have everything in tables, we have everything structured right here in one main database, it can access everything forever.
10:52Every model that gets upgraded, the output gets better because the context is already there. Everything is done for me right here, and in the morning, I just open this, and in ten minutes, I've done everything already.
11:05And I am ready in five minutes what previously took more than one hour every single day. So, the core message of this video is make sure your data is in one place and you store everything there. Not in 10 different systems, but all in one.
11:18Then from there, you can do anything. You can automate anything with Cloud Routines. Just explain to Cloud Routines.
11:24I have more videos about this. Long story short, just explain your workflow. Say that you want a scheduled run.
11:31And if you need more guidance in actually making a system like this, whether it's for personal use or for your whole team, make sure to check out the link in the description. We could do a call to discuss this further, but this is really the best move that you can make right now. So without any further ado, make sure to like and subscribe, and then we'll see each other in my next
The Hook
The bait, then the rug-pull.
Luuk Alleman opens with the line every solo founder wants to say out loud — 'my business finally started running itself' — and then earns it across the next eleven minutes by clicking through the actual system. The pitch isn't 'automate one workflow' — it's 'collapse every workflow into one database and let Claude Code be the brain on top.'
Named ideas worth stealing.
00:36concept
Features Rot, Infrastructure Compounds
Don't chase the latest shiny AI feature — those decay. Build the data + tables underneath, and every Claude release upgrades your business automatically.
Steal forJoe's $6 Stack manifesto — slot this directly into the 'own your tools' chapter as the philosophical anchor.
00:56model
The Four-Layer Founder OS Stack
- Supabase Postgres (the store) — one source of truth for every workflow
- Edge Functions (the pipes) — pull data in from Gmail/Revolut/YouTube/Azure
- Claude Code Routines (the brain) — read, decide, write back
- pg_cron + scheduled-tasks (the clock) — triggers every 5 min / hour / week
Architectural blueprint of the whole system. The top layer (Claude Code Routines) is the new one — historically you only had three layers, and the brain was you.
Steal forDiagram this exact 4-layer stack for MCN+ pitch. It's the same shape Joe is already running on Contabo + Supabase.
01:10list
Eight Surfaces, One Database
- Inbox OS
- Content engine
- Competitor watch
- Finance
- Ads
- Clients & CRM
- Azure ops
- Social distribution
Every workflow that used to live in a separate SaaS becomes a row in one table. The product surface (the dashboard) is just a thin web app reading from that one DB.
Steal forThis is the 'kill ten SaaS subscriptions' argument made visual. Use the eight-tile grid as a sales-page hero for MCN+.
02:10list
Anatomy of a Routine — Trigger / Spin up / Read / Act / Write
- Trigger (cron / scheduled-tasks)
- Spin up (full Claude Code session in cloud)
- Read (skills + Supabase + memory)
- Act (every MCP — Gmail, calendar, Supabase, Higgsfield)
- Write (back to Supabase + Telegram ping)
Why a Claude Code routine beats a trigger-then-LLM-then-output workflow: the LLM is no longer trapped in a-then-b-then-c. If a URL fails, it tries another. It's an employee, not a script.
Steal forThis is the missing piece for every JACE/REESE/SAGE/RYDER agent in Paperclip. Adopt this five-step shape as the standard agent lifecycle.
02:46concept
Edge Moves the Data. Routines Do the Thinking.
Clean division of labor: edge functions are dumb pipes that ETL external apps into Supabase; routines are the smart workers that reason over the data and write decisions back.
Steal forUse as the one-line architectural rule on any 'build your own AI stack' page. It tells readers exactly where to draw the line between code and Claude.
03:06list
Content Engine Loop — Mine / Score / Outline / Script / Package
- Mine — pull from competitor channels, his own comments, Transcripts
- Score — ideas ranked against his pillars; top three land in Telegram
- Outline — hook formula, retention beats, open loops, written before he sits down
- Script — plain-words pass, voice rules, speakable cadence baked in
- Package — three thumbnail + title variants generated and scored
The actual five-stage assembly line that produced THIS video. The proof is self-referential — he's showing the loop with an artifact made by the loop.
Steal forAdopt this exact 5-stage loop for Content Forge / Killing Excuses. Especially the 'Score' step — Joe doesn't have a ranking layer yet.
08:30concept
Routines = Employees
Every row in the routines table is an employee with a name, a schedule, and a job. 'Email drafts twice daily,' 'Build loop YouTube sync,' 'Morning review.' Reframing automations as employees changes how you manage them.
Steal forName every JACE/REESE/SAGE/RYDER routine as 'employee X' with a one-line job description. Show the org chart in the MCN+ pitch.
10:39concept
Context is the Moat
The single most important thing that makes AI output high quality is context. When everything you do lives in one structured database, every future model release upgrades you for free.
Steal forFrame the $6 Stack as 'context insurance.' You're not buying SaaS, you're insuring against future model upgrades by owning the data layer.
How they asked for the click.
VERBAL ASK
07:39product
“Make sure to check out the link in the description. At this point, I'm helping three other entrepreneurs build this for themselves... We could do a call to discuss this further.”
Done-with-you positioning at $10K-ish implied price (title says 'worth $10.000'). Scarcity baked in ('three other entrepreneurs'). Soft first mention at ~7:30, harder close at ~10:30 + final ask at ~11:30. Three hits total, well-spaced.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
