12:15Austin Marchese · Tutorial
Stop Prompting Claude. Start Loop Engineering.
A 12-minute framework for replacing one-shot prompts with self-running loops that verify their own work.
June 19thThe argument in one line.
A loop is just a skill you've already validated by hand, wrapped in a schedule, and the highest-leverage loops are the ones that make your whole AI system self-correct over time rather than the ones that do a single task faster.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude Code or a similar AI coding tool for real work and want a repeatable system instead of one-off prompting.
- You run a content, agency, or product business and want a structured way to turn raw inputs (Slack, email, calls, competitor research) into usable output automatically.
- You're comfortable with the idea of scheduled/automated AI tasks running in the background without a human triggering each one.
- You want a vocabulary for organizing the automations you already have (or plan to build) into a coherent system rather than a pile of disconnected scripts.
SKIP IF…
- You're brand new to Claude Code and haven't yet built a single working skill — this assumes skill-creation is already familiar.
- You want copy-paste automations rather than a framework; the prompts shown are templates you still have to adapt and test.
The full version, fast.
The video presents eight named 'Claude Code loops,' each built the same way: create a skill, run it manually to confirm it works, then wrap it in a scheduled routine. The loops are grouped into three buckets — ingest (data ingestion, external alpha farming, internal alpha farming), build (optimization, code-build, improve-system), and compound (ecosystem monitoring, North Star). Two concepts cut across all of them: approval gates at any 'critical call checkpoint' where a wrong turn wastes time, and a three-bucket change classification (auto-approve, needs sign-off, needs more context) used to keep automated changes safe. The improve-system loop and North Star loop are framed as the highest-leverage: one makes the system self-correct, the other checks whether all the automation is still pointed at the actual goal.
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 – 01:49
01 · Loop 1: Data Ingestion
Connect data sources (Slack, Gmail, Granola), build a skill that aggregates + filters noise, then schedule it as a local routine.
01:49 – 05:03
02 · Loop 2: External Alpha Farming
Define what alpha you're farming, identify sources (creators/newsletters/forums), and validate for level-two (non-obvious) insight before ingesting.
05:03 – 05:45
03 · Sponsor: Nexos AI
Mid-roll sponsor read for an all-in-one AI model router / no-code agent builder.
05:45 – 07:11
04 · Loop 3: Internal Alpha Farming
Mine your own data for recurring patterns and gaps; output must be an action list, not a report; includes building a reusable plan-verification skill.
07:11 – 09:34
05 · Loop 4: Optimization Loop
Pick a quantifiable metric, run/measure/propose/apply/remeasure in a loop; feedback doesn't need to be instant — 24-hour cycles are fine.
09:34 – 12:11
06 · Loop 5: Code-Build Loop + two core concepts
Six-step vibe-coding loop (goal extraction → plan → human sign-off → build → review → verify); introduces approval gates and Anthropic's dynamic workflows.
12:11 – 13:47
07 · Loop 6: Improve-System Loop
The creator's favorite — run twice weekly, analyzes sessions, proposes changes bucketed into auto-approve, needs-sign-off, and needs-more-context.
13:47 – 15:40
08 · Loop 7: Ecosystem Monitoring Loop
A loop that manages other loops: surfaces composability across loops, health-checks run logs, and auto-registers any new '[name]-loop' skill.
15:40 – 16:51
09 · Loop 8: North Star Loop
Locks in stated goals, analyzes trajectory from session history, extrapolates forward, and proposes direction changes if drifting.
16:51 – 17:11
10 · Recap + sign-off
Recaps the three buckets (ingest/build/compound) and points to two related videos.
Lines worth screenshotting.
- A loop is defined as a skill that already works manually, then wrapped in a schedule — skill first, automation second, never the reverse.
- Alpha farming splits into external (scraping outside sources for non-obvious insight) and internal (mining your own data for recurring patterns and gaps).
- Level one AI analysis restates the obvious framework; level two analysis is the non-obvious, contrarian insight that most people skip — only level two should get ingested into a system.
- An optimization loop doesn't require instant feedback — a conversion-rate loop that measures once every 24 hours and iterates over weeks still counts as a working loop.
- Any step in a build loop that is a 'critical call checkpoint' — where a wrong decision invalidates every step after it — needs a human approval gate; not every loop needs one.
- A three-bucket system (auto-approve low-risk changes, require sign-off on high-stakes changes, flag anything the AI can't decide) is the proposed middle ground for letting a system self-improve without letting quality drift.
- An ecosystem-monitoring loop that manages other loops can save tokens rather than burn them, by finding duplicated logic across loops and pulling it into one shared composable skill.
- The naming convention '[name]-loop' lets a monitoring loop auto-discover every new loop added to the system without manual registration.
- A North Star loop reads session history and recent activity, then extrapolates forward: if nothing changes, where does this trajectory actually land in six months.
- Every loop in the video is exposed to the user first as a slash-command skill they run by hand, and only promoted to a background 'local routine' after it's been manually proven.
Automate a skill only after it already works by hand.
SYSTEM DESIGN
The most transferable idea here isn't any single automation — it's the discipline of proving a skill manually before scheduling it, and classifying every automated change by risk before letting it run unsupervised.
01Loop 1: Data Ingestion
- Connect the data sources you actually work in (chat, email, call transcripts) before trying to build any downstream automation.
- Have the ingestion step actively strip contextual noise rather than storing everything it finds.
02Loop 2: External Alpha Farming
- Name specifically what edge you're looking for before picking sources — vague research goals produce vague results.
- Curate for non-obvious, contrarian insight and discard information everyone already has access to.
04Loop 3: Internal Alpha Farming
- Mine your own historical data for recurring patterns and gaps before assuming you need more external research.
- Make the output of any analysis loop an action list, not just another dashboard or report.
05Loop 4: Optimization Loop
- Pick a genuinely quantifiable metric to optimize against — vague quality judgments don't work as loop targets.
- Accept slower feedback cycles (daily or longer) when the metric requires it, rather than assuming a loop must run in real time.
06Loop 5: Code-Build Loop
- Separate planning from execution — extract the goal and get a plan signed off before any building starts.
- Add a human checkpoint specifically at the step where a wrong call would waste the most downstream work.
07Loop 6: Improve-System Loop
- Run a recurring self-review of your own system rather than assuming it stays effective by default.
- Log every automated change to a visible file so you can see exactly what your system has altered about itself over time.
08Loop 7: Ecosystem Monitoring Loop
- As you add more automations, actively look for duplicated logic across them and consolidate it — otherwise fixes only ever apply to one copy.
- Track which automations are actually succeeding so you can turn off ones that quietly aren't working.
09Loop 8: North Star Loop
- Regularly compare your day-to-day activity against your actual stated goals, not just against how busy you feel.
- Ask what happens if nothing changes — extrapolating your current trajectory is often the clearest way to see you need to adjust.
Terms worth knowing.
- Claude Loop
- A skill that runs continuously or on a recurring schedule until a task is complete, as opposed to a one-off prompt run manually each time.
- Alpha Farming
- Systematically hunting for information that gives an output an edge — either from external sources (competitors, creators, forums) or internal data (what's already recurring in your own system).
- Level Two Analysis
- Non-obvious, contrarian insight that goes beyond the basic framework everyone already knows — the bar an insight has to clear before it's worth storing in a system.
- Skill-Driven Loop Creation
- The practice of building every automation first as a manually-run skill, confirming it works, and only then wrapping it into an automatic recurring process.
- Critical Call Checkpoint
- A decision point in a workflow where getting it wrong invalidates all the following steps, marking where a human approval gate belongs.
- Dynamic Workflows
- A set of patterns (documented by Anthropic) for dividing a goal among multiple AI agents to complete a task, rather than running it through a single agent.
- North Star Loop
- A recurring check that reads your recent activity and data, compares it against your stated goals, and extrapolates where your current trajectory leads if nothing changes.
Things they pointed at.
Lines you could clip.
01:00
“A loop is something that runs continuously or in a specific schedule until a task is complete.”
02:50
“Level two analysis is the non-obvious information that brings an output from good to great.”
12:25
“Don't bring me a problem, bring me a solution.”
16:00
“If the other seven loops are instruments, this loop is the compass.”
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.
metaphoranalogy
00:00Loop number one is a data ingestion loop. Before we cover how to create this loop and why it's so valuable, what is an actual clawed loop? A loop is something that runs continuously or in a specific schedule until a task is complete.
00:10And to see it in action, loop one is the data ingestion loop. With AI, at the end of the day, your data is your mode. And if you wanna differentiate yourself and get a better output, you have to take data aggregation and ingestion very seriously.
00:22To set this up, we first connect our data sources. If you click the plus in the desktop app and then select connectors, you can then select the specific resources that you interact with. For me, it's Slack, Gmail, and then Granola for call transcripts.
00:33Once you establish those connections, it's time to actually create a skill that aggregates the data for you, validates that it's not just contextual noise, and then ingests it into the system. On screen, you can see a prompt that I used to create a skill called a data ingestion loop. This loop will go to Slack for messages, Gmail for emails, Granola for call transcripts, read the last twenty four hours, and remove all of the fluff and only store the valuable information.
00:55For every loop we create, I'm going through a process called skill driven loop creation, where every loop we create starts as a skill first that we manually run, confirm it works, and then we create a process to automatically run it. At this point, let's assume that we've tested it and it's now time to actually make this automated.
01:11To do this, I use local routines, which I get to by clicking routines plus local routine, and then I fill in the information where the instructions is I just use the skill that I already created. In this case, I set the skill to run Mondays at 8AM.
01:24And for every loop we cover here, you can just create a routine if you wanna run it automatically in the background. Now, that's just the first of eight Claude code loops that we'll go through in today's video that will help you build faster and more efficiently. And for all eight of these loops, I bucket them into three groups.
01:37The first is loops that help you ingest data get better outputs. The next are loops that help you build faster. And the last is a series of loops that help you create a system that automatically improves over time.
01:47Loop number two is external alpha farming. So what is an alpha farming loop? This is a loop that goes, looks for information that's valuable, and then pulls it into your system.
01:57And by sourcing the right information, you're able to get better outputs. Now, are three parts to this loop and so here's a prompt to create it, but I'm gonna go through each step and why it's important. The first part of it is you wanna establish what alpha you're actually farming.
02:08And for this, you wanna be very specific. So if you're a content creator, the alpha could be storytelling patterns or recent news topics. If you're building a product, it could be about conversion funnel benchmarks and optimization tactics.
02:19If you're a sales rep, it could be objection handling patterns from reps who are closing your exact customer. The second part of this is identify the sources of alpha. This is where you could use AI to help you create a list of options, but ultimately, this is where your human taste comes in.
02:33The question you have to ask yourself is who online is creating valuable insight about the topics that you're working on? So YouTube creators, newsletter creators, Reddit forums, where is their valuable information online that can produce a consistent stream of up to date information. And the third part is that you validate the source has level two analysis.
02:50Most people skip this part, but this is where it becomes an actual loop. So level one analysis is where AI just gives you basic stuff. Right?
02:57The basic framework, something that doesn't really differentiate the output. Level two analysis is the non obvious information that brings an output from good to great. And you wanna curate it so only level two information is ingested into your system.
03:09Here on screen, you can see me running the alpha farming loop, and you can see that it's actually going and fetching different resources. One thing I wanna call out is that I do use an MCP called Firecrawl that helps me get better information across the Internet. Now before we get to the next loop, which will focus on internal alpha farming, you may find that a lot of what we're covering feels a bit technical, which brings us to today's video sponsor, nexus dot a I.
03:29This is what I call an AI platform, which is a category of tools I love. And the reason for this is that it removes all of the technical skills that are normally needed to set up AI agent systems, and it can save you hundreds of dollars each month because you don't have to pay for five different AI tools. There are three specific features that I love.
03:44The first is that every flagship model is included in a single Nexus login. So Chatty Boutique, Claude, Gemini Grok, 200 plus other models can all be accessed directly through Nexus. And this is critical because there's really no single best AI model.
03:57Different ones are better at different things, and Nexus AI will automatically route each prompt to the best one. The second feature is that there's zero data retention. This means that your data is never used to train their models.
04:08And that alone is one of the big reasons why I can even consider using this tool for client work. And the third feature is their no code AI agent building. At the end of day, if you can use the Internet, you can use NexSys to build AI agents.
04:19You just describe what you want and NexSys will build the agent for you. And whether it's for one time jobs like, I need to do deep research on this topic like right now, or for repetitive tasks like turning your notes into a weekly report automatically, NexSys will handle all of this. For me, I have a lot of products in my ecosystem, my YouTube, my newsletter, build partners, so it's hard to keep track of all of this.
04:37So what I can do is create a weekly report that analyzes my entire marketing funnel and connect it to my Slack to send my entire team a summary of website visits, conversions, and what we need to work on. This is a task that I'd normally just never do because I just don't have the time. So to check out Nexus, click the first link in the description where you'll get 50% And that 50% discount is for a limited time because when I was working with them, I had to stress the importance of getting a large discount for everyone watching.
05:00So go check that out. Now getting to loop number three, which is internal alpha farming loop. If loop number two was all about hunting for external alpha farming, loop number three hunts for internal alpha.
05:11This will look across all the data that's already in your system and surface what's recurring, where the gaps are, and what you should be doing about it. To set this up, it's broken down into three parts. The first part is you wanna establish what internal alpha you're farming.
05:23If you're a content creator, which video concepts actually landed and performed? If you're building a product, what features do people keep requesting? If you're a sales rep, what questions or rebuttals does everyone have and should you build that into the service?
05:35This is the exact kind of pattern that humans will miss, but a loop doesn't. And the insights from this can lead to changes that actually move the needle. The second part is that you wanna point it at the right datasets.
05:45In loop number one, we set up data ingestion, so this builds on exactly this. But now, you actually have to ask, do I have the necessary data for internal alpha farming loop to actually work? And if the answer is no, this identifies gaps in your data pipeline, and you have to resolve that.
05:59And part three is the output is an action list, not a report. Essentially, every internal dashboard tells you all the information about what happened, but none of them actually tell you what to do about it. And one of the things that I tell my team is .
06:13And that's exactly what this loop is designed to do. It creates an implementation plan that passes our plan verification skill to make sure it hits our requirements for any sort of delivery plan. To create this loop, first, we need to create that slash plan verification skill, which is a utility skill that we'll use whenever we wanna verify a plan.
06:31Here's a prompt to actually create that skill, which will interview you about what tools you're currently using and what you're open to and not open to implementing. The key here is that any implementation plan that this creates has to actually be a viable option. So this skill is used to keep AI honest about any plan that it actually proposes.
06:47Then you use this prompt to create an internal alpha farming loop skill, which will create the loop, identify recurring patterns, and suggest places to improve. This loop will help you identify what you have to build next based on the data you're ingesting. As this loop runs, it'll look at the data that you're ingesting and suggest places for improvement.
07:04Now, we've covered ways to ingest data and how to actually analyze it, but the next three loops are all about improving the process in which you build. Loop number four is an optimization loop. This is a process of optimizing a system to approach a specific metric or quantifiable result.
07:19For example, does something pass level two analysis? Right? That's hard to actually quantify.
07:23But what a proper optimization loop does is it looks at something that is an objective metric. An example could be how quickly does a website load. So what this loop will do is it'll pick a specific goal.
07:33It'll run a loop, measure the result. If it doesn't actually fit the goal, it'll propose a solution, it'll apply it, remeasure it, and continue this loop until it gets to that final goal. On screen, you'll see a bunch of areas where an optimization loop makes sense, which is typically for more technical delivery mechanisms because these are super quantifiable.
07:52When I first started thinking about loops, I thought that all of the feedback back and forth had to happen almost instantly. Make a change, measure result, etcetera. But that doesn't necessarily have to be the case.
08:01So for example, let's say you have a website and you wanna try and optimize conversion rate. You could have a loop that runs, checks the conversion rate for day one, makes a tweak, checks it for day two, etcetera. And where this feedback loop is over a course of twenty four hours.
08:14This is entirely fine. And once I kind of realized this, this opened up my brain to the realm of possibilities because initially I thought that this feedback mechanism was way too slow. Now to create an optimization loop, here's a templated prompt that you can use and you can fill in for whatever you're trying to work on.
08:30Now before I get to the next loop, if this is your first video of mine, welcome to the channel. But if this is your second or more, here is our anti slop agreement. All of the stuff that I do in this video, right, the the visuals, the design, the hours of research, this is for humans, not for AI robots.
08:44That's part of the reason why I put all these prompts on screen because it's easier for you to read and screenshot, and it's not for these AI robots to see. So as part of the screaming, all I ask is that you subscribe to this channel to help this content reach more people. Also, every video I give away a clawed max subscription, and this video's winner is the GOAT.
09:00I'm Jasmine. They're building a platform to help people connect faster. Absolute bangers.
09:05To enter the giveaway, comment below with what you're building. And if you already entered, you can enter again by providing an update on whatever you're working on. Bring us to loop number five, which is code build loop.
09:13This loop is specifically if you want to vibe code a product. And everyone watching this video should be vibe coding whether you have a technical background or not. And the key with anything vibe coding is you don't wanna just go and grip and rip things.
09:24You need to plan before you actually do anything. And what this loop will do is your goals will get extracted. It verifies a delivery plan, and then and only then does it go through a loop to complete the tasks.
09:34On screen, you can see the prompt which breaks the task into six parts, which leverages skill chaining to create the actual loop. First, it'll extract or bring in the goal. So this can be through an interview or you can bring in a document you already have.
09:46The second is it'll plan the delivery using Claude's built in plan mode. The third is sign off on the plan. This is where you, the human, comes in.
09:52The fourth is it'll build it. Fifth is it'll review the code. This uses Claude's slash code dash review dash dash fix.
09:59Six, it'll verify the behavior matches the goal. It'll use Claude's slash verify to help with this. Now, went through all six features pretty quickly there, but there are two specific concepts that are critical you understand as you take these loops and start creating your own from scratch.
10:12The first is that there's an approval gate. If you look at the third step, it says sign off on the plan. That's an approval gate.
10:17And we do this at this stage because this is what I call a critical call checkpoint. If at this point you're in the wrong direction, steps four, five, and six are just going to be entirely wrong and you're wasting time and money. A general rule of thumb I have is that for any loop that you create, if there is any critical call checkpoint, then you need to add a human approval gate.
10:35Now to be clear, some of these loops may have none of these, but others may have four or five. It really depends on what you're building. The second thing is dynamic workflows.
10:42This is how Anthropic divides the goal you have amongst AI agents to complete the task. This is how it actually goes and builds whatever you're trying to build. And there are six patterns that Anthropic has documented for these dynamic workflows, which you can see all of them on screen.
10:56But to eighty twenty this whole thing, you don't actually have to worry about which one. It'll figure it out for you. But if you do wanna play around with it, you can look at the screenshot as a good starting point.
11:04Okay. So we've gone through data ingestion as well as how you can build quickly. But loop number six is about creating a system that self improves with the improved system loop.
11:13Of the eight loops that we're covering, this is by far my favorite because it single handedly transformed my business. Twice a week, I run an improved system loop. Tuesday end of day and Friday end of day.
11:22And that makes it so that my entire AI system is self improving. There are two layers to actually get this done. The first layer is you have to build the improved system skill.
11:30This is what actually looks at your system and figures out ways to improve it. So it'll read sessions, it'll find patterns, it'll propose changes, and I've spoken about this a lot on my channel. But if you want to use the exact one I use, I do have a plugin called build partner dot a I, and you can do slash b p improve system, and that's exactly what I use.
11:45Here on my screen, you can see me running this skill manually on my system, and it shows all the ways that I can currently improve Now that I look at it, I will have to do after filming this video. Now, the second layer for this loop is to actually create the loop. And this is wrapping a skill in a self improving system loop.
11:59But within this prompt, I log every change that it makes to a change log dot m d file. The reason is I wanna see what is actually happening in my system. And within those changes, it actually buckets it into three categories.
12:11The first bucket is auto approve. These are all of the low risk things that it can just automatically aren't really up for debate to improve the system.
12:18This is like the self improving component. Bucket two is need sign off. These are higher stakes stuff like skill edits or new skill candidates or structural changes.
12:26This is anything where the wrong choice could actually degrade output quality. These changes get written to a review file as a checkbox list. For each of the suggestions, you can approve, reject, or approve, and don't ask me again.
12:37This is the way to tweak this system over time. And the third bucket is more context needed. This is stuff that the loop can't decide on its own.
12:44Let's say, for example, you mentioned someone three times and it can't tell if this is a new client or a one off relationship. This will go in that same review file from bucket two. This three bucket system is what I found is an effective middle ground to make an automated system that improves without you, makes it easy for you to label improvements, and also keeps you as the tastemaker so that your system doesn't slowly degrade in quality from what you actually want it to produce.
13:05Loop number seven is ecosystem monitoring loop. Whenever you create a loop, more operational debt is added to your system. And this ecosystem monitoring loop is the loop that manages your other loops.
13:15 type Now before you say this is ridiculous, we're just gonna be burning through tokens, there are some key features why this can actually save you tokens while running a more efficient system. On screen, you'll see a prompt to create the actual loop, but there are three important features to call out.
13:30The first feature is it surfaces composability opportunities across loops. As you build more loops, you'll unknowingly write the same logic twice. For example, let's say you have two loops that require fetching data from Slack.
13:41There's a chance that both of these contain the same logic. So the ecosystem monitoring loop scans your loop library, finds repeated logic, and suggests pulling it out into a composable skill that every loop can call.
13:53That's exactly what we did with a plan verification skill back in loop number three. We built it once and then any other loop can use it. The more loops you create and the more logic that's shared between them, the more likely you're gonna fall into a whack a mole trap.
14:04This is where you'll see a problem in one skill, you'll fix it, and then it'll pop up in another place. If the logics were used across skills, you wanna abstract that into a single skill that gets called by the other loops. And to all the programmers out there, yes, it's like you're creating a reusable function.
14:19Feature number two is the health check across the stack. You can cross reference every loop's output to see what's running successfully. But for this to work, every loop has to log its results properly.
14:28So we'll create a right run log utility skill that writes everything to this folder. On screen, you can see a prompt that will create the specific skill and simultaneously enhance any existing loops that you have. We do this so that every loop writes to the same location.
14:41And if you update this specific write log skill, it'll update across every other loop that you have running. And by having an effective way to monitor loops, you're going to keep a close eye on everything that's running and you'll quickly notice what is and isn't working. And as a result, you'll be able to turn off what isn't working saving you tokens.
14:57I can guarantee there are probably millions of people right now running loops that don't know it and they're just burning tokens. This helps you avoid that. And the third feature is it dynamically updates, which removes operational debt.
15:08Every loop that we've created in this video has the same naming convention. It's name dash loop. And on each run of this ecosystem monitoring loop, it scans the skills for any skill that has this name.
15:19And as a result, this ecosystem monitoring loop will automatically include any additional loops that you start running. This means that there's no operational debt to actually maintain this monitoring loop because it self corrects. And remember, for every loop that I'm covering this video, if you wanna run them continuously or on a schedule, you can run routines directly through clawed desktop app.
15:37Now the past seven loops that I covered are game changers for productivity, but this last one could make the biggest difference. Loop number eight is your North Star loop.
15:44If . This loop will monitor your activity and make sure you're actually pushing towards your goals. Simply put, this is asking, is everything that I'm doing pointed at my actual goal?
15:55And there are four parts to this loop, and you can see the prompt on screen. Part one is you lock in the north star. This is similar to the goal extraction we did earlier, but you have to establish, like, what are your actual goals?
16:05Hit 100 k subs, land 12 new clients, ship four paid products, get a promotion, sign four new clients. Like, what are these goals that you wanna be pushing towards? Part two is it'll analyze your trajectory.
16:14This will read your clawed session history, the data you're ingesting, the loop results, everything that you've been working on, and see where you're going. Part three is it summarizes it. This is about forward extrapolation, essentially saying if nothing changes, here's where you're going to land.
16:28For me, this is usually the biggest kick in the ass because it's like, if I don't change anything, then I'm gonna be here in six months. The fourth part is proposed direction changes. If drift's detected, it'll surface what's pulling you in that direction.
16:40It'll detect this drift and then propose things that you should change instead. This is the type of loop that frankly everybody needs to start using. So those are the eight loops that you need to build bucketed into three different groups.
16:50Loops that help you ingest data, build faster, and create a system that compounds and can be maintained over time. Now, if you like this video, you'll love this video where I dive deep into how you can set up Clawd to be a self improving system. It builds on a lot of the topics I covered here.
17:03And if you pair what I cover in that video with the loops that I walked through today, you're gonna be on a whole another level. So go check that video out and I'll see you over there. Peace.
The Hook
The bait, then the rug-pull.
The video opens mid-explanation on loop one rather than a cold-open tease — the hook is structural: eight numbered loops, delivered as a countdown, each with its own on-screen title card.
Named ideas worth stealing.
01:30list
Three Loop Buckets
- Ingest (loops 1-3)
- Build (loops 4-6)
- Compound (loops 7-8)
All eight loops are organized into three purposes: getting better inputs, building faster, and creating a system that improves and stays aligned over time.
Steal fororganizing any personal or team AI-automation stack so new automations have an obvious home
01:00concept
Skill-Driven Loop Creation
Every loop starts as a skill that's manually run and confirmed to work before it's ever automated into a schedule.
Steal forde-risking any new automation before it runs unsupervised
02:50concept
Level 1 vs Level 2 Analysis
- Level 1: the basic framework, obvious, doesn't differentiate
- Level 2: non-obvious, contrarian insight worth ingesting
A filter for deciding which scraped/researched information is actually worth storing in a system.
Steal forany research or content-curation pipeline that risks drowning in generic AI output
00:00list
8-Loop System
- 1. Data Ingestion Loop
- 2. External Alpha Farming Loop
- 3. Internal Alpha Farming Loop
- 4. Optimization Loop
- 5. Code-Build Loop
- 6. Improve-System Loop
- 7. Ecosystem Monitoring Loop
- 8. North Star Loop
The full named taxonomy of loops covered in the video.
Steal fora checklist to audit which loop types you're missing in your own AI workflow
12:20model
Three-Bucket Change Classification
- Auto-approve: low-risk (duplicates, dead links)
- Needs sign-off: higher-stakes (skill edits, new skills, structural changes)
- More context needed: ambiguous cases the loop can't decide
How the improve-system loop classifies every change it proposes so automation can't silently degrade quality.
Steal forany self-modifying or self-improving automation that touches production systems
10:40concept
Two Critical Concepts for Loop-Building
- Approval gates at critical call checkpoints
- Dynamic workflows (Anthropic's agent-division patterns)
The two ideas the creator says are essential to understand before building your own loops from scratch.
Steal fordesigning any multi-step AI workflow that includes irreversible or expensive steps
How they asked for the click.
VERBAL ASK
05:05product
“To check out Nexus, click the first link in the description where you'll get 50% off.”
Standard mid-roll sponsor read with a scripted platform pitch (model routing, zero data retention, no-code agent building) and a personal-use example before the discount ask.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
- Loop engineering foundation ↗
- Self-improving system deep dive ↗
- Stop prompting, start agentic engineering ↗
- Free 5-day AI playbook (what I used to build a $25M+ startup) ↗
- Use BuildPartner to build 10x faster with Claude Code (try free) ↗
- Want to build a SaaS product without hiring a CTO? ↗
- Firecrawl (MCP plugin for web scraping) ↗
- Instagram ↗
- Youtube ↗
Visual structure at a glance.
