My AI Team Took 8 Hours. Claude Sonnet 5.5 Took 15 Minutes.
A founder pits his fully-loaded Claude Opus 5.5 team against a blank-folder Claude Sonnet 5.5, and watches eight hours of guardrails lose to fifteen minutes of nothing.
Posted
yesterday
Duration
Format
Demo
educational
Views
6.4K
39 likes
57 · 43
Big Idea
The argument in one line.
A Claude Opus 5.5 team working inside a year-old project folder full of guardrails, guidelines, and sub-agents produced five rounds of broken, ugly UI concepts in eight hours, while a fresh Claude Sonnet 5.5 instance with zero context and one reference image produced a more delightful interface in ten to fifteen minutes, showing that accumulated guardrails can restrict creative output even as they protect code and security work.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You run a multi-agent Claude Code setup with persistent memory, SOPs, and sub-agents, and want to know when that structure helps versus hurts.
You're using AI to redesign a UI or product experience and keep getting safe, boring, or broken output despite detailed instructions.
You're deciding whether to hand a creative brief to your fully-configured AI system or to a blank instance with minimal context.
SKIP IF…
You're looking for a finished UI kit or design system to copy. None of the interfaces shown here shipped.
You want technical detail on how the multi-agent orchestration itself is built. The video doesn't cover that architecture.
TL;DR
The full version, fast.
An AI productivity coach needed a simpler onboarding UI for his course platform's five-course ICO methodology. He gave his established Claude Opus 5.5 team, a year of guardrails, guidelines, and sub-agents, a full day and five complexity levels to redesign it: eight hours and millions of tokens later, every version was broken, ugly, or overcomplicated. He then gave Claude Sonnet 5.5 an empty folder, zero context, and one reference image. It produced a more delightful, tactile interface in ten to fifteen minutes. The lesson: heavy guardrails and multi-agent structure protect code and security work but can box a creative task into timid, self-limiting output. Control over what your AI system stores and restricts matters more than which model you use.
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.
States the video's thesis that simplification applies to AI systems too, and previews a real example of an AI team going off the rails.
00:44 – 02:40
02 · A complete but overwhelming membership
Describes the existing course platform: fully built courses with videos and slides, but so much content that newcomers can't see the big picture quickly.
02:40 – 03:38
03 · Distilling the ICO methodology into five courses
Explains how the team condensed personal, team, and automation/AI levels of the ICO methodology into five courses and 240 self-assessment questions, and decided the current interface needed a simpler onboarding path.
03:38 – 05:28
04 · Briefing the established AI team
Sets up the eight-hour session inside a year-old project folder with full context, persistent memory, the 'inkline' design system, and a five-level complexity spec from pure minimalism to 3D and GSAP animation.
05:28 – 10:30
05 · Level 1 through 5: the Opus 5.5 team's results
Walks through each of the five requested complexity levels from the guarded Claude Opus 5.5 team: broken fonts, overlapping boxes, a mobile-breaking layout, and an overly childish, still-broken gamified version at the top end.
10:30 – 14:43
06 · Starting over with Sonnet 5.5 in an empty folder
Gives Claude Sonnet 5.5 a blank folder, zero project context, and a single reference image, and gets an interactive, tactile interface with sound feedback in ten to fifteen minutes, then asks for three more variations.
14:43 – 16:49
07 · Taking Sonnet's result back to the AI team
Shows the guarded Opus 5.5 team's attempt to match Sonnet's result while still working from full project context; the output looks derivative and unremarkable next to Sonnet's version.
16:49 – 19:02
08 · Removing sub-agents and going 'unhinged'
Tells the AI team to stop launching sub-agents and work in a single session, then invites it to go wild with three more ideas; the results, a stamped tracker and a lit-up tower, are buggy and look cheap.
19:02 – 21:31
09 · The real content test: database access, no style guide
Sets a harder brief for Claude Sonnet 5.5 in a blank folder: pull real lesson content from the database, avoid default AI styling and competitor platforms like Mighty Networks, and build in React and Tailwind CSS.
21:31 – 25:40
10 · Atlas, Metro map, binder, and card game concepts
Reviews four very different navigation concepts Sonnet 5.5 invented unprompted: an Atlas-style map, a Metro-line diagram, a physical binder layout, and a collectible-card game, each interesting but not quite right for shipping.
25:40 – 28:22
11 · Why the existing interface might still win
Concludes that none of the AI-generated concepts beat the platform's current, simpler implementation, and opens up the project's guideline files to show how much accumulated context the established AI team carries.
28:22 – 30:44
12 · The real lesson: it's the setup, not just the model
Argues that heavy guardrails help with code, security, and database work but can over-restrict creative tasks, and that understanding what your own AI system stores is what lets you catch it going off the rails.
Atomic Insights
Lines worth screenshotting.
A Claude Opus 5.5 multi-agent team with a year of accumulated guardrails took eight hours and millions of tokens to produce five UI concepts, and every one shipped broken or ugly.
A fresh Claude Sonnet 5.5 instance with zero project context and one reference image produced a more delightful interface in ten to fifteen minutes.
Sub-agents built to enforce guidelines started limiting each other, citing prior stakeholder opinions as reasons not to try new approaches.
The most minimalistic requested version was also the most broken: fonts jumped around on click.
Running an AI team at a higher reasoning-effort setting didn't fix the underlying problem, because the constraint was accumulated project rules, not the model's reasoning budget.
Asking Claude Sonnet 5.5 to create three vastly different versions from one prompt surfaced ideas, a stamped field passport, a card-collection game, that the guarded team never proposed.
The same Opus 5.5 team, told explicitly to stop launching sub-agents and work in a single session, still produced buggy, cheap-looking output when given the full guarded context.
Negative prompts like 'don't use AI slop' and 'don't look like Mighty Networks' steered a context-free Sonnet 5.5 toward more original visual choices than positive instructions did.
Giving Sonnet 5.5 direct database access to real course content, with no design-system constraints, produced a Metro-map and Atlas-style navigation concept nobody had asked for.
None of the AI-generated interfaces, from either model, were judged good enough to replace the platform's existing, simpler UI.
Heavy guardrails are valuable for code, security, and database work, but the same guardrails can make an AI team too cautious to propose a genuinely new creative direction.
The real lesson wasn't about model choice: understanding what your own AI system stores and restricts is what lets you catch it going off the rails and fix it.
Takeaway
How Guardrails Can Box In AI
CONTEXT VS CREATIVITY
A fully-guarded, context-rich AI team took eight hours to produce broken UI concepts, while a context-free instance produced a better one in minutes, showing that accumulated guardrails can suppress creative risk-taking.
02A complete but overwhelming membership
A course platform can be fully built and functional, with videos, quizzes, and slides, and still fail newcomers if there's no fast way to see the big picture first.
When a product accumulates depth over time, the fix for overwhelmed beginners is usually a simpler on-ramp, not more content.
03Distilling the ICO methodology into five courses
Compressing a large body of material into a personal-to-team-to-automation progression makes a system's shape legible before someone commits to the detail.
240 self-assessment questions across five courses is real, valuable depth, but it's also exactly the kind of scope that needs an onboarding shortcut.
04Briefing the established AI team
Persistent memory, a documented design system, and a year of guidelines all get loaded as context before an AI team even starts a creative brief.
Giving an AI team a graduated spec, like five complexity levels, is a good way to explore a design space, but it doesn't guarantee any level will actually work.
05Level 1 through 5: the Opus 5.5 team's results
Every one of five requested complexity levels came back broken in some way, from jumping fonts at the minimalist end to unclickable elements at the most complex.
Sub-agents built to enforce guidelines started limiting each other by citing prior stakeholder opinions, which narrowed the outcome instead of improving it.
Running the same team at a higher reasoning-effort setting didn't fix the problem, because the constraint was accumulated project rules, not the model's thinking budget.
06Starting over with Sonnet 5.5 in an empty folder
A single reference image and zero project context produced a more delightful, tactile result in ten to fifteen minutes than eight hours of guarded iteration.
Asking for three deliberately different versions from one blank-slate prompt surfaced ideas, like a stamped tracker and a card-collection concept, that the guarded team never proposed.
07Taking Sonnet's result back to the AI team
Handing the guarded AI team the same brief, with its usual full context, produced a derivative, one-to-one copy of a sound effect but none of the original spark.
More context and more history with a task doesn't automatically produce a better creative result than a model starting fresh.
08Removing sub-agents and going 'unhinged'
Explicitly telling an AI team to stop launching sub-agents and work in a single session changed the process but not the quality of the output.
Encouraging a guarded AI team to 'go wild' produced ideas that were interesting in concept but still buggy and cheap-looking in execution.
09The real content test: database access, no style guide
Explicit negative prompts, like 'don't use AI slop' and 'don't look like Mighty Networks,' did more to steer output than positive style instructions did.
Giving a context-free model direct access to real data, the actual course content, let it design around real material instead of guessing at placeholder content.
10Atlas, Metro map, binder, and card game concepts
An unconstrained brief produced navigation metaphors, an atlas, a metro map, a binder, a card game, that nobody on the team had thought to ask for.
Novel and visually interesting isn't the same as usable: several of the concepts were engaging to explore but didn't clarify the underlying productivity system.
A card-collection-game structure can represent how lessons interconnect, but gamification alone doesn't guarantee it teaches the underlying skill.
11Why the existing interface might still win
None of the AI-generated concepts, from either model, were judged strong enough to replace the platform's current, simpler interface.
A year of accumulated guideline documents can be so extensive that no single AI agent loads all of it, which is exactly why sub-agents exist to scope it.
12The real lesson: it's the setup, not just the model
Heavy guardrails are genuinely valuable for code, security, and database work, where restricting an AI's options prevents real damage.
Understanding what your own AI system stores and restricts is what lets you notice it going off the rails and correct it quickly.
The comparison wasn't really Opus versus Sonnet: it was a heavily-scaffolded process versus an unscaffolded one, and the scaffolding lost on this creative task.
Glossary
Terms worth knowing.
Orchestrator
The lead AI agent in a multi-agent setup that delegates sub-tasks to specialized sub-agents and combines their output into one result.
Sub-agent
A narrower AI agent spawned by an orchestrator to handle one piece of a larger task, sometimes enforcing a specific guideline or checking another sub-agent's work.
Guardrails
Written rules, guidelines, and prior stakeholder feedback stored in a project folder that constrain what an AI agent is allowed to produce.
AI slop
Generic, default-looking AI output, such as a model's standard style sheet or layout, that appears when no specific design instruction is given.
x-high effort
A reasoning-effort setting that lets a model spend more computation per response; used here on Claude Opus 5.5 without fixing the underlying design problem.
Resources
Things they pointed at.
05:33toolThree.js
09:52toolGSAP
08:18toolClaude Opus 5.5
11:00toolClaude Sonnet 5.5
11:32toolExcalidraw
20:23toolReact
20:24toolTailwind CSS
20:38productMighty Networks
Quotables
Lines you could clip.
00:00
“Simplification is king and it also comes to AI.”
Tight, quotable thesis that opens the video and doubles as the takeaway.→ TikTok hook↗ Tweet quote
10:30
“After eight hours, millions of tokens burned down the drain.”
Blunt, high-stakes summary line right before the pivot to Sonnet.→ IG reel cold open↗ Tweet quote
12:29
“This reminds me of my time in corporate when we worked on complex projects... you just got stuck in meetings, talking about the same over and over.”
Relatable analogy comparing over-restricted AI agents to corporate meeting paralysis.→ newsletter pull-quote↗ Tweet quote
17:48
“Now that you are unhinged and you can do whatever you want, go really wide and create three more things.”
Memorable, slightly absurd prompt phrasing that captures the 'let the model loose' moment.→ TikTok hook↗ Tweet quote
28:42
“It's not only the model, it's also how you work with these models.”
The video's actual thesis, stated plainly near the end.→ newsletter pull-quote↗ Tweet quote
The Script
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.
17px
metaphoranalogystory
Simplification is king and it also comes to AI. We are all building AI agents to work the way that we want it and we built amazing systems in the past six months. But here are some conclusions I want to share with you that are critical to understand.
when AI goes off rail and why it doesn't work the way that you want. And I have a real life work example here I want to share with you. This is our membership.
And in our membership, we share our courses, right? And we build this ourselves where our members have a lot of things that they can do, including a complete social media platform and so on. We are happy with everything that there is and also the content.
It's fine. We have our courses. These courses are having questions.
They have a video. They have a magic slide. Everything is perfectly functional and we like it the way that it is.
However, due to the sheer amount of information that we provide to the members how to set up a productivity system end -to -end, we now realized we need to give people an easier way to kickstart their journey and understand quickly the big picture of the productivity system. This is where we are now researching how we best can provide this to our members.
And of course, there we use our AI systems to see what they can come up with to simplify the UI UX for the user. So for example, in our case, tends always to be very wordy and does a lot of words and information that is maybe distracting and things like this. This is something that we wanted to address.
However, we wanted also to see how far we can move AI forward to do something that perfectly represents what we are looking for. So that's where we had the meeting. And there we thought about all the courses and all the content that we have in these courses.
And this is what we came up with, that the people who want to learn the ICO methodology to build a productivity system, at least on a personal level, these are the core concepts and workflows they need to know all in one box. And then these are just five lessons that people need to watch in order to get started. Then here is the rest of these courses that really dive much deeper in dozens of lessons in the nitty gritty, right?
But if you really want to kickstart, this would be it. Then we split our ICO methodology from a personal level, which includes also private life, personal work in a company and so on into a team level, which was the project management like a pro. And that's what we get here.
ICO for teams basics again. what to focus on and how to dive deeper. And then the I go for automation and AI, which is the whole course about process automation, process mapping and things like this to build your AI system.
And this is also where we include our. AI course, our AI courses to move on next. Okay, all this together is the iCore journey.
It is five courses people go through. And if we go back to what it looks right now, currently, they just go through these five courses, which is quite a lot because each course has four questions people can answer to understand how healthy their productivity system is end to end, summing up to 240 questions. And this is actual work people put into if they really want to max out the productivity system.
However, newcomers obviously they are overwhelmed they just want to get started quickly and that's why we thought about this good this is the context we had a whole session we have even transcripts where Paco and I our co -founder we're talking about this concluding these things how we bring in our ICO framework for those who are no new here welcome but I'm not going into much detail about how this works but we have it inside the community where you can lay out your tools on these frameworks to understand how your productivity is system works end -to -end on a tool level okay now i let ai work on interfaces that we can use to replace this interface that we currently have with something more unique something more minimalistic easy to understand easy to digest and to follow and this is what it came up with after guys after and i show you the real session here in in vs code this was part of the session and this is where today at 12 o 'clock lunchtime i started the session and it ran eight hours burning through tokens and building this out and i just want to quickly explain to you what my guardrails were first of all i was running this in my live folder so it had the full context about me how i work what i like the
the persistent memory that we built with our systems too. This also includes the ICO methodology itself, but also access to our platform, all the context, everything is there. It also includes the design system because this is the inkline design system that we developed uniquely for our platform that everything has this style and also, you know, the font styling and this accent color and the handwriting and things like this.
This is the inkline design system. So obviously when it was working on this, it tried to stick to these guardrails.
In addition, I mentioned that the team should go to a bootcamp and brainstorm about it, why we are not happy with what they came up with. I'm not even showing what it showed in the first place. And they should really work on this to find something that ticks all the boxes.
And the main guardrails I gave it was five levels of implementation of the interface. The first one is reductionism. highly minimalistic all the way up to level five which would be three gs and animations and all this okay but within these guard drives the thing is i looked at this then midway i stopped after eight hours and burning all this there was a lot of back and forth between the main agent larry in this case the orchestrator was delegating the work to the sub -agents and all the sub -agents there were agents in there that were limiting the other agents and say we cannot do this because of look at these guardrails and look off the opinion from paco from this session and look at the opinion from me in this session and they worked according to plan but they boxed themselves in so narrowly that this outcome is actually shocking after all this time and token investment running opus 5 .5 on x high effort, this is what I got.
This should represent this idea of the quick understanding of an end overview of your personal productivity system. It's this okay, it's the very minimalistic level one approach. And it's even broken when I click here, look at this how the fonts are jumping around.
And this is what I cannot excuse them. Because it is broken this it's just a few lines and it is just a few words it's a very basic basic thing and here it understands okay there's the icon for life icon for teams the things that i explained but okay there's no visual representation and i would even say it doesn't look too bad at this from a stylistic point of view as i was asking for the most minimalistic approach and yet i mean the centered things here it's still stylish nonsense.
And I already had the feeling the reason is that the team was too restricted by all the guidelines and SOPs and things like that, that the team is following inside my folder. And therefore, obviously, that's the reason why it's burning so many tokens. Let's go to level two.
Okay, it thought a bit to represent this in a better structure. And I can click on this. But, but guys, I don't know.
This is, first of all, nothing special at all. And it is not a quality that we usually ship to our members. Okay.
That's useless on mobile. It looks like this. it was completely broken again.
And this is so simple that it shows that it overcomplicated things even on a simple level. And I will show you in a moment what Sonnet 5 .5 created out of the box without any context at all. Okay, so then here, this is level three and I mean, we are now with Opus 5 .5.
I created crazy things with Opus 5 .5. That's why I was so confident that it will get the point to what we need. And yet it didn't.
And, you know, look at this overlaps here on the box. Really bad design choices. And how the heck should...
people gain clarity from this thing and then here's the icon for teams and yeah maybe some people would ship this it won't be us nah nonsense and really bad bad design level four level four it went then too far and still it looks like crap here's a broken image here I can click on this the text is disappearing and shows up here but how the heck man it looks so cheap and we don't we are not into this style at all so I don't know I have to click start in order to watch this then and then far too complex too confusing now the lamp is moving in so it seems that maybe I stopped it halfway in and that could have fixed along the way I don't know.
We will never know because I'm not burning any more tokens on this. As this is, see, it's not clickable. This is not it.
Okay. And it's also not the style that we are following with our incline approach. And this is level five where I thought, you blow my mind.
Here I said, you can use 3GS, GASP, all the things where people creating amazing things. And I did too. And it came up with this.
Okay. I can click through the different items of iCore for life, but it's too far gamified. It's too childish.
When I go iCore for teams, man, and this should represent professional business team and the productivity, how make team communication, how to manage ideas, how to get things done, project management, complex projects, and people come here and see this. What the heck? This is not representing and my AI team should know, based on all the context that it has available, that this is certainly not the approach we are going for.
And again, half of it is broken. I cannot click it. This was it.
After eight hours, millions of tokens burned down the drain. Then I thought, you know what? Let's use Sonnet 5 .5 on a complete new folder.
empty folder without any context in there and let's let it create something from scratch and the only thing that i gave it here in the initial prompt was this image what i showed you initially the idea of the ico for life basics to give people the understanding what are the key concepts and workflows that they need to understand in order to build up their productivity system.
And heck, even this would work in some way or form, but we still think it could look a lot better. That's something I quickly sketched out in my Excalibur version here that I customized to use the incline branding here. And this is the result from Sonnet 5 .5.
Yes, it is not really useful. but it is visually impressive. And it even sticks for whatever reason, maybe because it's clawed.
It's also having orange in there. So it used the clawed colors, but well, it looked like this. I can go through this.
I like the idea going through this and it updates here. I can click here to see the different things in here. Okay.
It's a gamification. You hear the sounds and I love the feedback. That's something I asked for.
I want to have it. pleasing to click on these buttons and you know look at this when i click done okay there's some there's some feedback happening there's something going on and i can click on this you see there was an empty box it is represented here this was one prompt 10 to 15 minutes it took to create this compared, I'm just comparing this to eight hours of overthinking.
And this reminds me of my time in corporate when we worked on complex projects and things like that. And you just got stuck in meetings, talking about the same over and over having So many different opinions left and right where everybody needed to be pleased.
And that's exactly what happened to my AI team and the different agents talking to each other. They overthought. They're worried that they cannot please me anymore because I also introduced feedback loops to say, okay, you know what my goal is.
Ensure that my reaction will be full of joy instead of meh, right? These are the things that I gave it. And this was the result.
from this to this and this and this okay but but this was not at all because i said to sonnet in these 15 minutes create me three vastly different versions and he created also this where it actually used a diagram and then represents it this way and yeah it is like from a little game looking and but the thing is my team also created something like this but it's one to one converted it into an interface here.
See, I'm hovering and I see this flickering of the ink. So it is now really focusing on this design. And I love the feedback that just a haptic feedback that I get from this interface.
And here we are using Sonnet 5 .5 guys. And now look at this final version. It just blew my mind.
Boom. Okay. If it is useful, let's Ignore this for a moment, but just the visual representation of these elements of this basic setup.
And then I have these lessons and look at this.
Just undo it. But if I complete this.
What the heck, guys?
This is so satisfying to use and this is something I could now iterate on. That's not something you just stop now and call it a day. But maybe some elements like this haptic feedback or something like this could be now used to iterate.
And I just want to make the point here that this is based on one image that I gave to explain on one screenshot of this part of our whiteboard to explain what our productivity system is about. And it created this out of it. So now why do I think that this is actually default?
interfaces that ai provides well i went back to my team i told my team then to say i asked my team and i said why the heck is sonnet doing so much better compared to you when you have all the context and well it told me because we've been too restricted we tried to force things into the incline and then the back and forth and the worries that they wouldn't achieve the goal was what was holding them back.
And I said, okay, then forget about it and create something as crazy. And I created this.
I think it's interesting. It looks like a tiny device. It's clearly using the idea of this because it's even using the same sound.
So it copied it in and just did the same thing, okay?
In a very minimalistic way. OK, well, he copy pasted. Now the next one is this.
OK, it's not too bad. I have some cards and I can go through these cards and I can bring these cards over. But I have no idea that I can do this.
I just had to figure that out that I can do this. And now I have this. OK, minimalistic.
But does it blow my mind compared to what Sonnet created? Not at all. Plus, this is on Opus 5 .5 and again using the whole team's context.
I know that using the access to the Teams context, the only thing that is different is that now I said, you shouldn't launch any sub -agents anymore, just do it in one session out of the box. And then I said, okay, well, you start it and then look at this. He created a stamp version where I can watch and then it...
gives me a stamp that I watched it all and then what again it's not even it's it's even buggy let's please please let's have a look again at this stamp version because there it also used stamps and then the overall progress on top man I'm really interested if this actually would work obviously I need to watch the video and all this it's not really useful when it comes to this but just the ideas behind it and that's what I expected some something unusual unexpected to see and not something like this and this is where i said okay now that you are unhinged and you can do whatever you want go really wide and create three more things and he created this oh i can't i can't even even kind of because when you i you know it's it's actually also sad because this costs money creating this costs money and i just realized i can move this around so what does it what does it tell me what what can i look at this but also the design how cheap everything is looking then here the lights what the heck i can okay i can switch on the lights
But again, it's not even functional or anything like that. Again, and then what? And then the tower, where I realized then too that I can move these things down here.
And wow, it's so cheap, I have no words. I went further. and i said okay let's call it a day with my ai team let's stay in this empty folder no context at all i i have the access to our database and i said get all the lessons and chapters that we have written in there and use it to get the full understanding of what we consider a productivity system on a personal level will be end to end and create something create interfaces that represent the full productivity system end to end.
Let's actually go to the prompt. Here's the full prompt, okay? This is where I said, go in, check out the courses, the chapters of the first three courses of the iCore journey, and then create something amazing out of it.
You know, so the idea is to have a very simplistic version, but also it can go very detailed, become really detailed. no matter if you're a beginner, or you're an advanced member, you should always get what you want to look at. And then I said, I also say don't use AI slop or your cloud standards, because I'm aware that there are some default style sheets and things like that, that a cloud is providing, if there are no instructions, and these negative prompts, they really help to stay away from it.
I told it to use react and tailwind CSS, because this is what our platform is also coded at and then we will see what it will come up with but i also said it should not look like the boring interfaces, like the standard course from circum mighty networks school.
I'm so fed up with these things. It's so that's why we've built our own platform. But I don't want to bring into our own platform now features from other existing platforms.
And that's why we will keep working on this. And now I'm really curious because it has the access it downloaded all the content here. That's what I asked it for, for later.
reference, I have now all the sessions in here of the courses that it has access to. And now I'm really curious. This is just running on SONNET 5 .5 on high.
Currently, there are no sub -agents running and it's still... Oh, and he finished. And let's see what it created.
ICO for life basics. Okay. Well, that's not really...
Aha. But that's interesting.
Single source of truth, workflow, jot things down.
Well, first of all, I don't like the looks at all. The functionality is also not really... Oh, okay, there's a full atlas and a starter route, but this is not mobile friendly.
It created a map so I can actually navigate on this map.
And I can click here on the topics and now I have on this the different, well, not at all. Let's go further. Okay, the metro.
Okay, so it represents this as different stations. That's the express. And this is the full context.
So it has all these things. But now I'm underwhelmed too, again. I mean, it is much more innovative than anything that I could have wished for from Opus 5 .5.
But Probably due to the fact that I didn't give it any directions, I'm getting now something that it just made up and tries to link things like an Atlas, a Metro. And obviously there's no design system in this.
I didn't provide the ink line design branding or anything like that because I really wanted to have it unlocked as much as possible. In general, I like this thing that it understood how to perfectly store a node and then how this is actually...
That's actually interesting.
Okay.
Well, okay. It's complete gamification, obviously, and it's not educational at all. And I still don't get the point.
And that's the whole thing. I said it needs to be intuitive. If I need to think too long to actually get to the point of the whole thing, this is too long.
It is a nice idea, but I will ditch this too. Okay, so if I click these key elements, it actually shows just the things that are relevant for key elements. I mean, I need to look into this more calmly because it might have done here something that's really useful when it comes to connecting the dots of a productivity system.
But that's certainly not it, something that we could deliver to the members. Now the binder. Oh, what the heck?
Okay, what is this? We have the starter deck and we have the full collection. And again, It shows me now all lessons or just a few lessons.
I can click on it. And then it shows me this. It makes this interactive card.
I can click on goals and then, okay, there's a bit of a... Okay. Well, for me, I love gamification, obviously.
But that's actually interesting to see. But does it really have to do with professional productivity? I don't think so.
You can let me know in the comments below if any of these three as a professional working in business and want to manage your life is actually attractive to you to use to learn the ICO methodology and to become more productive in life. I might use this for, I don't know, create a card collection game or something like this.
This is also something not useful. I like this idea. I mean, it's a card collection game.
And I'm collecting the cards which represents different concepts and workflows or lessons for these different things. And here, weekly planning is actually one lesson that we have of a workflow of iCore workflow. And I like the fact that here it considers that the weekly planning is not only one lesson, which is this one, but it actually consists of more than that and how it's interconnected.
And yet it doesn't give me any feeling of. calmness, or clarity, anything at all, where then in the end, if we can perfectly live with our current implementation, that's the straightforward one, you have these lessons, you have them here, we can unfold, we can unfold to see the lessons, we go here, we see the progress of the lessons, we can go here, and here's the overall progress of our lesson.
So no matter what I have seen so far, nothing would be worth replacing this. And I'm sure in the end, Paco and I, we will meet and we will come up with something that we define ourselves and then we will use AI to help us building it maybe.
But what I got so far, I just wanted to point out here how limiting it is if... you have an existing folder with a lot of context, with an AI team that is fully established for over a year, I've kept building guidelines, look at this, just the guidelines alone. And then we have guidelines for our my icon system, just a brand system itself, then a content design.
There are so many, and they are not short things, if you click into it, look at this, they are all really big things. And how this works is AI is not loading in all this as context, obviously.
That's why we have these agents and they only look into specific based on their agent thing into specific guidelines. And that makes it very effective to have a lot of information in one folder for the team and still just picking what is necessary. But the more guardrails and things like that that you bring in and the more checkpoints, as you have seen now in this video, the more restrictive you make.
your ai team too which is a good thing in many situations especially when it comes to code or security implementations or database work or things like this this is very important actually to have in place it sticks to the icore branding and try to make it really minimalistic but then you see there is a moment where it really is stuck and cannot move forward because it has the feeling that it cannot move left and right and burns through a lot of tokens.
And this is the moment when we talk here, that might be the top down approach where we have everything ready and you now create something small. And then you go back to bottom up approach. That's what I did here.
Again, also explaining in the ICO methodology. And I started an empty folder and I see what I can get out of it. And then I might start iterating.
Maybe I use this as a starting point for my other AI team to say, OK, look, there was something along with it that you can use. the code is already created I can point it now simply to this folder and say look at this how it was built and now apply the incline design system onto it this being said nothing that I shared in this video will make it into our membership you can be sure about this but I thought I want to make this video to explain it's not only the model it's also how you work with these models and also a bit contradictory to the other videos where we promote these guidelines, the SOPs and how the AI team is working with each other and how they're launching the sub agents and multitasking and even cross session work that they do.
All amazing. But if you have no idea what happens under the hood and that something is off and you kind of point your finger on it, this is the moment where the things start crumbling. And this is the moment where the better you understand what you actually build for yourself, the more control you will keep about this.
That's why when you download skills, when you download plugins or even fully made AI systems, or even these chatbots that come out like Crockpot and all these things, they have something like I had here a setup under the hood to make them work according to these rules that they have there and they decide what they will store for you and so on and that's with our approach we have full control over what the AI is accessing and how it's behaving and that's why we stick to this method until we see that there's really an AI coming out that does it as good as our local folder approach but you have seen here I was able to very quickly realize, man, there's something going off rail and I was able to point my finger at it and now I can keep working on it without any issues.
And I thought I'd share this work in progress with you because a productivity system is actually pretty complex. The ICO methodology itself is already simplifying the complexity. Now all we are working on is simplifying the accessibility to these ICO lessons to make it even simpler to grasp and understand.
And well. I just wanted to share this with you. If you had some similar experience with your AI team, let me know in the comments below.
Let's chat about it. And if you liked the video, give it a thumbs up, share it with your friends. And if you haven't already, subscribe to the channel so I can catch you up in the next one.
The Hook
The bait, then the rug-pull.
The pitch is right there in the title: an established Claude Opus 5.5 team spent eight hours failing, and a blank Claude Sonnet 5.5 instance won in fifteen minutes. What follows is the founder's own screen recording of that comparison, guardrails and all.
Frameworks
Named ideas worth stealing.
05:28list
Five Levels of UI Implementation
Level 1 — reductionism / highly minimalistic
Level 2 — light structure
Level 3 — richer visual detail
Level 4 — heavy detail, near-complex
Level 5 — 3D and GSAP animation
A five-rung complexity dial the host gave his AI team to explore, from a bare-bones layout up to a fully animated 3D interface.
Steal forany brief where you want an AI system to explore a design space by dialing one variable, like visual complexity, instead of guessing at a single output.
02:40model
ICO Methodology Structure
ICO for Life (personal level)
ICO for Teams (team / project-management level)
ICO for Automation & AI (process automation and mapping)
The three levels the host's productivity framework operates at, compressed into five onboarding courses with 240 self-assessment questions across them.
Steal forstructuring any multi-level curriculum or onboarding flow into a personal-to-team-to-systems progression.
CTA Breakdown
How they asked for the click.
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
With memory switched off in both apps, the same prompt run three times against the same folder shows Claude following the rules and ChatGPT ignoring them, until it gets pushed.
A 34-minute tutorial that open-sources Tom Solid's myPKA scaffold — one markdown folder that any agent CLI initializes itself into and turns into a 4-agent PKM team.
A 35-minute live demo of a local browser app built with Claude that replaces Notion, Sunsama, Heptabase, and every health tracking subscription — all from one markdown folder.
00:34
00:57
01:43
02:06
02:52
03:15
03:38
04:02
04:25
05:11
05:34
05:57
06:20
06:43
07:29
07:52
08:15
08:38
08:52
09:14
09:44
10:10
10:33
10:56
11:19
12:06
12:29
12:52
13:19
13:38
14:05
14:13
14:37
15:01
15:33
15:49
16:52
17:04
17:18
17:45
18:14
19:24
19:47
20:10
20:33
20:56
21:13
21:53
22:02
22:28
22:51
23:37
24:00
24:23
24:46
25:09
25:32
26:06
26:18
26:41
27:05
27:28
28:14
28:37
29:00
29:23
29:46
30:34
23:24
34:33
35:38
09:09
09:02
21:38