A 90-minute Zoom webinar arguing that most AI projects don't fail because the model is weak — they fail because nobody organized the context, and a plain folder of markdown files beats the agent swarm almost every time.
Posted
1 months ago
Duration
Format
Tutorial
educational
Views
26K
837 likes
57 · 43
Big Idea
The argument in one line.
Most AI initiatives fail not because the models are weak but because nobody engineered the context: a plain folder of markdown files, organized like 1970s Unix software, lets one AI model become as many specialized agents as a question requires, without brittle multi-agent infrastructure.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You run an agency, freelance practice, or internal team that has burned real budget on multi-agent pilots that impressed in a demo and then quietly stopped being used.
You want a concrete, low-tooling way to give an AI model durable context about your business, opinions, and workflows without building a RAG stack or agent swarm.
You're evaluating whether a paid 12-week cohort (the Lyceum, $4,750, starting September 12) is worth it for a specific project you already know you want to build.
You're curious about the actual software-engineering lineage (Unix, make, Parnas) behind the current wave of 'AI skills' and folder-based context patterns.
SKIP IF…
You need genuinely high-concurrency, real-time multi-agent infrastructure at enterprise scale — Jake says outright that's not what this talk, or the Lyceum, is for.
You don't have a specific project or goal yet — the webinar (and the cohort) both explicitly turn away people who show up only to 'learn about AI' in the abstract.
You're only interested in a recording to passively watch — over a third of the runtime is live Q&A, chat banter, and sales logistics for the cohort.
TL;DR
The full version, fast.
Jake Van Clief argues that AI projects fail from missing context, not weak models: MIT's Project NANDA found 95% of agentic initiatives show zero measurable return, and Gartner projects 40%+ of agentic projects canceled by 2027. His fix, Interpretable Context Methodology (ICM), organizes context as plain markdown folders one AI reads and navigates, becoming as many specialized agents as a question requires — a pattern he traces to 1970s software engineering (Unix, make, Parnas), not new AI theory. One agency owner who implemented it cut developer/API spend from $10,000/month to $400. The back half pitches his $4,750, 12-week Lyceum cohort (starting September 12) built around the same folder-based 'second brain' methodology, with technical, business, and creator tracks.
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.
Jake waits for the Zoom room to fill past 200 attendees, greeting people by city, before starting the actual talk.
03:40 – 13:09
02 · Who Jake is: the interactive resume
Bio via an interactive 'field map' site (USMC avionics on F-35/F-18s, University of Edinburgh MSc in AI governance, consulting at $2,000/hour, 43,000-person community). Shares the open-source Cartographer's Desk repo used to build it, and states upfront that he'll pitch something at the end.
13:09 – 16:35
03 · ICM's origin and the first case study
ICM (Interpretable Context Methodology) started as a research paper after Jake noticed clients paying premium consulting rates for fragile agent systems. Curtis, a community member, cut his agency's dev/API spend from $10,000/month to $400/month by implementing it.
16:35 – 21:09
04 · The failure numbers
Cites MIT Project NANDA (2025): 95% of agentic/AI initiatives show zero measurable return. Gartner: 40%+ of agentic projects will be canceled by end of 2027; 42% abandoned already, up from 17% the year before.
21:09 – 23:27
05 · Six failure patterns companies hit
Names the recurring pain points he sees in client work: pilot purgatory, context sprawl, token burn, the handoff problem, governance fear, and the quarterly rebuild cycle.
23:27 – 29:10
06 · Features are solutions, not the problem
Argues that n8n, Zapier, cron jobs, agent swarms, vector databases, and RAG stacks are all just branded solutions to old software problems (scheduling, orchestration, retrieval, memory), and explains RAG in plain terms.
29:10 – 34:09
07 · When swarms actually make sense
Concedes agentic swarms are the right call for high-concurrency, high-scale use cases, but says 99% of the time they aren't needed. Traces ICM's lineage to 1970s software engineering: Unix (1978), make (1979), and Parnas's information hiding (1972).
34:09 – 46:42
08 · Live demo: one AI, one folder, the four layers
The library analogy (front desk / catalog / request slip / memory-state) mapped to a live VS Code walkthrough of Jake's own company ICM folder — a fresh AI reads the folder and 'becomes' the agent needed for a given question, without being separately coded. Audience Q&A covers ICM vs OKF and database-tier context (e.g. a 10,000-table Oracle database).
46:42 – 53:11
09 · Augment, don't automate
States the thesis directly: the best AI systems augment human judgment rather than fully replacing it. Cites the 1987 Solow productivity paradox and argues folder-based context survives model upgrades because the model, not your opinions, is what's being replaced.
53:11 – 1:00:03
10 · What would you hand off?
Live audience prompt: attendees post the one workflow they'd hand off tomorrow. Jake reframes it through 'job to be done' (the drill/hole-in-the-wall example) and live-demos an AI drafting his own follow-up emails into Gmail during the call.
1:00:03 – 1:05:46
11 · Pitch: the Lyceum begins
Transitions from free content into the sales pitch, still showing the live email-drafting demo. Introduces the Lyceum: a 12-week cohort program with technical, business, and creator tracks, each student getting a personal 'second brain.'
1:05:46 – 1:11:29
12 · Program structure, cohorts, and prizes
18 hours of live instruction across 9 sessions, certification, guest instructors, and competition prizes (a year of Claude Max, a Mac mini M4 Pro, $50K and $100K of billable consulting time for top finishers). Names who the program is explicitly not for.
1:11:29 – 1:13:47
13 · Price and the placement call
States the price: $4,750 for the full 12 weeks including builds. Program starts September 12. Enrollment goes through a free 20-minute placement call, not a hard sell.
1:13:47 – 1:31:31
14 · Audience Q&A and close
Extended live Q&A: cohort structure, capstone examples (financial-data anonymization pipeline, a 7-person high-revenue company's internal 'second brain', a creator video-tracking tool), certification, guarantee, and time commitment. Closes recommending 1970s-70s software books (The Mythical Man-Month, The Psychology of Computer Programming) over AI-specific content, and repeats his closing line about questions over cheap answers.
Atomic Insights
Lines worth screenshotting.
MIT's Project NANDA (2025) found 95% of agentic and AI initiatives showed zero measurable return.
Gartner projects more than 40% of agentic AI projects will be canceled by the end of 2027, up from 17% abandoned the year before that figure was measured.
An agency owner who rebuilt his operation on files and folders instead of developer tools and API calls dropped spend from $10,000/month to $400/month.
N8n, Zapier, cron jobs, agent swarms, and vector databases are not new inventions — they're branded solutions to decades-old software problems: scheduling, orchestration, retrieval, and memory.
RAG (retrieval-augmented generation) is just one way to answer the question 'how do we get the right data into the model's context window' — folders and files are another valid RAG mechanism.
ICM's folder-of-markdown approach applies three pieces of 1970s software engineering to LLMs: Unix's 'do one thing well' philosophy (1978), make's separation of coordination from execution (1979), and Parnas's information-hiding principle (1972).
One AI model with a well-organized folder can become as many specialized 'agents' as there are questions asked of it, because it only reads what a given question requires (progressive disclosure) — no separate agent build needed per task.
A folder-based context system survives model upgrades better than a fine-tuned or heavily-prompted agent, because the model gets better at executing your process without you having to re-teach it your opinions each time.
The 1987 Solow productivity paradox ('you can see the computer age everywhere except in the productivity statistics') is repeating with AI: the technology is visible everywhere, but the productivity gains lag until organizations learn to augment workflows instead of just automating tasks.
The best AI systems augment a person's judgment rather than replacing it entirely — full automation without a human review step just turns the system into a tool nobody trusts.
Before building an AI solution to any workflow problem, ask the 'job to be done' question: don't build a better drill, ask why the person needs a hole in the wall — the real job might not need automation at all.
Six recurring failure patterns show up across the companies Jake consults for: pilot purgatory (demos that never get adopted), context sprawl (institutional knowledge that's unreadable), token burn (spending on volume instead of efficiency), the handoff problem (one expert user, nobody else can operate it), governance fear (no clear owner if something goes wrong), and the quarterly rebuild cycle.
Takeaway
Fix your context, not your agent count
WHAT TO LEARN
Most AI failures trace back to disorganized context, not weak models, and a plain folder of markdown files usually beats an agent swarm at a fraction of the complexity.
04The failure numbers
MIT's Project NANDA found 95% of agentic AI initiatives showed zero measurable return.
Gartner projects 40%+ of agentic projects will be canceled by 2027, up from 17% abandoned a year earlier.
05Six failure patterns companies hit
The failure pattern is consistent: impressive pilots that never get adopted, context nobody can find again, high token spend for low outcome, and single-point-of-failure operators.
Governance fear and the quarterly rebuild cycle are the least discussed failure modes but among the most costly, since nobody owns the risk or the accumulated knowledge.
06Features are solutions, not the problem
N8n, Zapier, cron, agent swarms, and vector databases are branded solutions to old software problems (scheduling, orchestration, retrieval, memory) — know the underlying problem before adopting the tool.
Retrieval-augmented generation just means getting the right data into a model's context window; a well-organized folder of files is a legitimate RAG mechanism, not a lesser one.
08Live demo: one AI, one folder, the four layers
One well-organized folder of markdown files lets a single AI model act as many specialized agents, reading only what a given question requires instead of running a permanent multi-agent pipeline.
Context stored as plain files survives model upgrades better than heavily fine-tuned or prompt-engineered agents, because the new model just gets better at executing a process you already own.
09Augment, don't automate
The best AI systems augment human judgment instead of fully replacing it; total automation without a human review step turns a system into a tool nobody actually trusts.
Productivity gains from a new technology lag behind its visibility (the 1987 Solow paradox) until an organization changes how it works, not just what tools it uses.
10What would you hand off?
Before automating any workflow, ask what job the person is actually trying to get done — the real need is often simpler than the tool being requested to build it.
Well-scoped automation targets a specific, already-understood job, rather than 'automate everything I do,' which tends to solve problems nobody actually had.
Glossary
Terms worth knowing.
ICM (Interpretable Context Methodology)
Jake Van Clief's research framework for organizing AI context as a navigable folder structure of markdown files instead of building custom agents or complex retrieval pipelines.
OKF (Open Knowledge Format)
A related approach, referenced as using a similar markdown-and-YAML method to ICM, described as more focused on the retrieval side while ICM is focused on process and orchestration.
RAG (retrieval-augmented generation)
A technique where relevant data is retrieved and inserted into an AI model's context window at generation time, instead of retraining the model on that data.
Pilot purgatory
The state where an AI pilot project impresses in a demo, gets some budget, and then quietly stops being adopted or used without ever being formally killed.
Context sprawl
When a company's AI-generated knowledge and processes exist but aren't organized anywhere readable, so nobody can find or reuse what was already built.
Token burn
Spending large amounts of money on AI API usage for comparatively little outcome, often because the underlying workflow wasn't made efficient before automating it.
Progressive disclosure
A design pattern where a system reveals information only as it's needed, rather than loading everything at once — applied here to an AI reading only the folder contents relevant to the current question.
Job to be done
A product framework that asks what underlying outcome a person actually needs, rather than the feature they think they're asking for (e.g. asking why someone needs a hole in the wall, not how to build a better drill).
Solow productivity paradox
Economist Robert Solow's 1987 observation that computers were visible everywhere in the economy except in the productivity statistics, until organizations learned to restructure work around them.
“He was spending $10,000 a month on developer tools and API costs... By implementing the paper, doing this process, he dropped it down to $400.”
Concrete before/after ROI number, the strongest social proof in the talk.→ IG reel cold open↗ Tweet quote
29:43
“It's actually comes from 1970s engineering. I didn't invent anything. I merely applied good software engineering and conceptuals from the 70s to the modern day.”
Contrarian humility that undercuts the hype framing of AI tooling.→ TikTok hook↗ Tweet quote
34:53
“Essentially, it's like walking into a library... One librarian walking through. If you can create that, you can start to map it to features, to things you need to build.”
The core mental model of the whole talk, stated as one clean analogy.→ newsletter pull-quote↗ Tweet quote
46:42
“The goal is to create a system, automations that a person can use, not simply just automate the person.”
Thesis statement in one sentence.→ TikTok hook↗ Tweet quote
53:50
“Instead of sitting there and saying, okay, cool, and spending all your time building a better drill... you need to ask me why I need a hole in the wall.”
Memorable, concrete analogy for scoping any automation project.→ IG reel cold open↗ Tweet quote
1:31:20
“In a world full of cheap answers, questions become valuable.”
His closing line, short enough to be a standalone caption or pull-quote.→ 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
Okay. Everyone should start wandering in here now, hopefully. Should be allowed to join.
You should have access to chat. And if you don't, I should be able to fix that. There it is.
Yep. I see Zeb, you're raising your hand there. Good to go.
Just want to make sure that everyone's able to get in. We'll give it a few minutes here. It's not technically seven o 'clock yet, my time.
Just want to give everyone a moment here. Greetings. Luke Muhammad.
Hello, Jason. Michael Barrett from Denver. Okay, cool.
Denver. Barrett. Dr.
Valerie Johnson. Hello. Hello.
If we do have any doctors or any PhDs, I'd love to, you know, you could drop your thesis work. I always love reading people's thesis work. A lot of my friends are PhDs at Edinburgh or Harvard right now.
And I'm always, I'm always interested in reading more. Always like to see what's out there. We're going to give this another minute or so.
Let people kind of wander in. We're up to 142 people. That is amazing.
MFAs and poetry. Absolutely. Drop that in here, Erica.
I'd love to. Hopefully it lets you drop a link in. I don't know if Zoom might be blocking that just so we don't have too many people spamming, but happy to dive into that.
My mother. She got her master's and everything in writing and religion, actually, and a lot into the kind of old poetry. It's where I learned a lot of wordplay and whatnot.
I did stay in a Holiday Inn. I'm not a PhD, but I stayed in a Holiday Inn Express last night. I love that.
Abdul from Fredericksburg. Hello. Like I said, let me just give it a minute.
Let people start wandering in. We're up to 157 people now. Amazing.
Oakland, California. Cleveland, Hilo. I'm actually going to be in Cleveland here soon.
We have a company that's commissioned us to come go work out with them, not work out. We're not going to the gym, but I might be in Cleveland here soon. Chicago, I was just out there with the company actually working with them.
I love Chicago, Oregon. I know. Oh, Jorge.
For everyone who doesn't know, Jorge is like one of my best friends. I grew up with him. He's probably pissed that I haven't messaged him back.
I've been so busy. I promise I will, Jorge.
Miami, Florida. Amazing. Hello, Jesse.
We got Richmond, Malibu. Yes, yes. New York, New York.
Ah, Beirut. Thank you for taking your time. I know it's late there, so I'm happy to have you.
Rhode Island, San Diego, Uncle Chief. Thank you. Well, all right.
We're giving it a minute there. We're up to almost 170 people. That's amazing.
Wow. Albuquerque, Seattle. Wow, this is so cool.
Yes. Yeah, no, I'll have I'll have meeting notes and everyone's going to be getting an email of slide decks and any links I share. For those of you who are at last week's webinar, which hopefully isn't too many of you, there is a lot of repeated stuff.
Of course, you know, I always all of my lectures are not scripted. I follow the slide deck, but, you know, I'm always kind of reading along Clearwater, Florida. Oh, my gosh.
Yeah, I spent a lot of time up in in St. Augustine. And I am currently right now.
That's where I'm recording from. But OK, cool. Two minutes in.
Yes. No, Jade. Jorge said I have to hit the gym.
He's a so Jorge is a firefighter. He's an absolute legend. He works out monster and has he actually has an amazing Instagram page that he just started where he's going and sharing dives.
Actually, Jorge, if you want to drop your tag into chat, you're more than welcome to. He's doing all sorts of really cool dives off the coast. But anyway.
That's not what a lot of you are here for. Most of, well, I mean, Jorge is amazing. We're almost at 200 people now, which is absolutely amazing.
I'm happy to have you all here. So if at any point you struggle hearing me, you cannot see my screen, please let me know. Just say in the chat.
Oh, we got David McDermott as well. He's actually one of my co -founders for my software company. He's been building things together with us.
Uh, he's an amazing CTO, um, and has done some amazing work, a full stack engineer, all the good jazz. He's one of the weirdos that like C sharp and JavaScript. So he's one of the monsters in the programming world with that.
But, um, anyway, happy to have you here too, David. Um, anyway, so. Let's go ahead and start wandering in.
So you're all here. I appreciate having you a lot. It's still growing.
We'll probably be over 200 as people join in. So that's an amazing feed out. First of all, thank you everyone for showing up.
I really appreciate it. Teaching and lecturing is my passion. It's what I love to do.
And if no one showed up, actually, I wouldn't be that sad. My first. real lecture in front of a big group of actually it was professors was probably about three years ago.
And only six people showed up. I need to find the photo of that so I can share. And we went from that to rooms filled with hundreds of people, things like this.
So again, all of you make this worth it. So I appreciate it. Now.
On to today. So a lot of this is going to be very lecture based. We're obviously going to be answering some questions here or there.
You'll have Matt hopefully in the chat answering questions as well. David, I'm sure would be happy to answer some questions if they're very technical. He's been building a lot of amazing stuff.
But the goal really is to kind of look at. AI differently. But before I do that, for those of you who may be here because you saw my ad or you're just figuring out about this and you signed up, I think what's important is to dive into who I am.
And so an amazing person from my community, from Cleef Notes, which I'll talk about here if you're not already a part of it. made me this really cool interactive resume that I want to share with all of you. And Cole, yes, this is already being recorded.
And like I said, we'll be sending transcripts and all sorts of stuff out after so that you all can have any links I share, which I'm actually about to share a link. But for those of you who don't really know me that well, maybe haven't looked up much at who I am or like, why should I really listen to this guy in the first place?
Just wanted to give you an idea of who I am and what I am. Again, this is a really amazing uh kind of interactive thing i get to scroll my mouse and i'll share a link on how to build this uh in a second here you all can just throw it to your ai and it can pull from the library and help you build out it's a little complicated setup nothing crazy but then you can start iterating really quickly So I started out in the U .S.
Marine Corps. Well, I didn't start there. I was born in Florida.
But I ended up starting my technical career after building computers and being all these different weird things. I ended up joining the U .S. Marine Corps where I worked on the F -35 and F -18s.
While I was doing that, I learned about automation and cryptography and electronics and all of these things. And that's where I actually first really dived into about a decade ago, this kind of idea of... automation, autonomous processes.
I actually saw These F -35s hovering midair and then landing and the systems that were required to switch from jet to hovercraft just being absolutely controlled by the AI. The pilot saying, well, it basically lands itself.
And in many cases, the pilot would eject and the plane would keep flying. There's one case you can look it up. One of the F -35s we lost because it ended up having a pilot eject and it flew for another few miles before finally kind of crash landing in a field.
And I just realized that that was wild. I knew I wanted to understand it better, but I wanted to take a more peaceful route. The military wasn't really where I thought I could be more peaceful for obvious reasons.
So I ended up getting my undergrad degree. I ended up working in political science and data science and all of these things, which led me to the University of Edinburgh. That is where I got my master's of science in AI and governance, specifically called future governance.
I worked with their future. Futures Institute, which is this beautiful institute. If you ever get a chance, look it up.
If you happen to be in Scotland, go dive in there and check it out. From there, I started doing research. I created data pipelines, multi -agentic workflows.
I started testing whether or not models had bias in their data. I was using psychometric scales to push through and test these AI to see what we have to worry about. I got some interesting papers out there.
I worked with the the Neuropolitics Lab. We have another paper that's just finishing its last round of edits, and it's going to be published in a pretty great journal.
So I'll be sharing that soon. But that taught me a lot of like the no's and the bad's. And it let me kind of look into data science and AI.
And I started teaching people and working with the professors and working with companies, so much so that Edinburgh and a couple other schools ended up hiring me and my team that I was putting together to actually work and build out these systems. But more importantly, importantly, know what not to build.
How do you create scale? How do you create ROI? What are the downsides of machine learning, artificial intelligence, building systems that actually are going to be useful for 10 years?
And as soon as we started charging 2000 an hour for some of these sessions, I realized that and then being told that that was undercharged where we were working. I was like, OK, I need to double down on a company in this space because that's good money. But more importantly, that means that we can really go and explore.
This is something that people value. But also, they're paying that much, who are they paying?
What is the value return? Where are they seeing that? And where is the value not being returned?
And well, I actually realized that a lot of people were getting paid the same as me, but they were building out systems that were fragile. And you'll see a lot of my videos on Instagram. I'm constantly talking about how agents are a problem.
And I go into it later, and I will here shortly. But essentially, I wrote a research paper on where you might not need that. complicated infrastructure where agents are a waste of time.
There is a nuance there. Sometimes they are completely useful. There are specific situations.
But this research paper, ICM, which I go into here a little bit later, is called Interpretable Context Methodology. And it's essentially just a really unique way to use folders to get to the same outcome that building these complex RAG and agent architecture would do. It's much more sustainable.
It lasts a lot longer. And honestly, a lot of people in my community haven't really needed to worry. about updating too much because of how these models work.
Obviously, that paper and that work ended up bringing thousands and thousands of people into my community. It also meant that a lot of companies were hiring us to train their employees. I trained well over 1 ,600 professionals in Fortune 500 and corporate places, learned with them, learned from them.
And then, of course, we have now actually need to update this number. I think we're at over 43 ,000 in the community, which is amazing. Just a quick little idea of who I am and maybe why you should should take what I have.
Obviously, everything with a grain of salt, but that's who I am. If any of you want to build your own version of that. Uh, let me go ahead and share that repo.
Uh, one, you can go see what I have been giving as well as, um, you know, download that GitHub repo for yourself. Uh, it's, it's pretty useful. It's really cool.
Again, big shout out to Danny Trio, uh, for building this out. Uh, it's such a good idea. He used a lot of my folder methodology to help build it.
Uh, and, uh, if you ever want to build one for yourself, it's a lot nicer than a resume, right? At least that's what I think. So go ahead and take that link.
That's all yours. And again, we'll throw these into an email to share with you as well as other links. And that's also my paper too.
If you are interested in it, you can throw it to your AI if you don't feel like reading it. But if you're one of the book nerds that like to dive into it, I have a whole research paper that people are diving into. So you can go ahead and check that out there.
Again, you don't have to copy them now. We'll send them an email after as well. So the goal tonight is to talk about ICM.
Now, some of you may already know what it is, but we're going to dive in a little deeper. Again, if you were at the last webinar, it's kind of the same concepts here, but it's always nice to double down. I'm going to talk about why pilots are failing.
So I've worked with a lot of companies and whether you're part of an agency, whether you're inside of a company, whether you're running a company, I see some people getting huge ROI, but I see some people wasting thousands, if not millions of dollars. And so I want to kind of show you what that looks like. to show you how I like to build from a high level.
Obviously, there's only so much I can teach you in a short amount of time, but I've tried to condense it all to be really just hitting heavy and where you can do it as well as where you fit in. Because a lot of people are so focused on automating, they ignore the most important part, which is augmentation. The goal is to create a system, automations that a person can use, not simply just automate the person.
Because if you automate everything, well, it's just become a tool, not just simply a company within reason. We'll dive into what that is. And then we'll talk about what's possible there, where people are making money as well as other things.
And this is where I want to be super transparent. I do want to sell something at the end of this, all right? Okay?
It's going to be free stuff all the way through. I'm going to be dropping links. I hate it when people are like, come to my free webinar, but also buy this thing.
I want to be very upfront. I have products. I have processes.
I have things that I do want to sell at the end. Some people in here might already be signed up for it. So just if you're sitting there, oh, he's probably trying to sell me something.
Yes, I 100 % am. You do not have to buy. This is completely optional.
But if you learn a lot in here, It is an extension of those learnings and dives in. You should feel obligated to at least book a call with us and learn more about it, I hope.
Again, this is what allows me for being able to do this, to do this at scale, to do this full time. And honestly, people have been telling us we're not charging enough for it. So that's, I guess, a good sign.
I don't know. But yeah, I just wanted to be very transparent. And thank you, James.
I really do. I don't want you to be blindsided. However.
That's the last time I'm mentioning it. I'll mention it again when I'm about to sell again. But from here till I mention it again, it is free stuff.
There's no weird funnel stuff here. This is good value. So just putting that there for you.
So again, as I mentioned, this started out as a research paper. It was called Interpretable Context Methodology. And essentially I was trying to find a way.
For AI, for these models to be able to interpret our context, what we needed, because I kept hitting all these problems. We had all these different ways to be able to build AI. We're going to go deep into what that is here in a moment.
But I just want to explain the traction because what ended up happening is that people dove in. They ended up using this paper, giving it to their AI, restructuring the way that their data pipelines, the way they work with things worked. research teams, enterprise.
We had a CTO who we had a call with of a large private equity firm say, it's like, I read your paper twice. Absolutely love it. Diving in more.
And it's not right. It's one of those things where it's like you have papers, you have all these people online telling you about features and the next clog thing. I want to lead with outcomes.
People used my free open source paper and got. outcomes out of it, right? We had one company, Curtis, he talks about it in our community deeper, and he's also an amazing guy, lots of teaching in his own agency.
He was spending $10 ,000 a month on developer tools and API costs with AI and these things just for one of his teams. By implementing the paper, doing this process, he dropped it down to $400, literally saving $9 ,000 a month, literally over $100 ,000 a year. just by implementing that process, the data structure.
And I really, really think there's some gold in that that you would all love. The other side of this is I see these limits, these models. They're brilliant.
They're doing so much. But realistically, a lot of these local models, even some of these other things, they're good, but their limit, their problem is the context. And even if we have all the best rag pipelines and data structures, the problem is your opinion, your workflow, your context.
Because in a world where we are entering more automation, best becomes relative. The best type of politician, the best type of art, the best type of program is relative for where you are. And so productionizing your opinion, which I'll go to in a moment here, became the value stack.
These models are important and they are getting better, but there's a better place for you to sit so you don't feel like you're rushing, which is the other thing. You shouldn't need to feel like you're rushed. You shouldn't need to feel like you're falling behind.
Your goal is to feel like. AI is another tool. It's Excel.
It's Microsoft. It's whatever it is in your stack that you're providing value in. If you don't look at it that way, you have not understood the layer of work, the layer of abstraction, as in how things are working that you should be sitting at.
And the people who didn't understand it gave us these numbers, right? So there's an MIT project, Project Nanda, that went through and found 95 % of agentic and AI initiatives had zero measurable return.
Zero, like lost money or had no return, right? 40 % of agentic projects are being canceled by the end of 2027. That's from Gartner, right?
Almost half of agentic projects are not being pushed forward through. The other side of this, 42 % of these projects and initiatives are being abandoned entirely.
And that was only 17 % the year before, right? This is important numbers. Understanding why they failed, but also understanding, well, what about the 5 % that succeeded and did get a return?
What about the 60 % that didn't cancel? What about the 60 % that are not getting, or 58 % in this case, that are not getting canceled? right?
That is important, right? And Gio and Augie, I'm so glad this is your second time. I hope, again, this is a lot of the same stuff as last time.
We're trying to speak to other groups and whatnot here. But as always, I don't keep a script. So there's always little nuggets that are here and added in.
So I'm glad you're here. Happy to have you, Gio. So the problem, though, is that the ones that failed did not see the pattern underneath.
They did not look at what the software was like. They did not look at structure. They did not look at the context.
They looked at the surface. They looked at the shiny objects, the new tools, the new frameworks, the new things that change weekly. Ah, Gemini's better this week.
OpenAI's better that week. Claude, if you're doing your workflows right, your data's right. It doesn't really matter which model you use within reason.
And I prove that in a lot of my work. I barely bounce between models. I just choose one that I know works well for what I need and upgrade it if I feel like we need the newer model.
And it should be easy to upgrade. It should just be a simple button click. It shouldn't be this insane thing that you're reaching for the next newsletter, in my opinion.
And that's what we really want to do. Again, the research paper, the link is earlier in there, but it's called ICM, Interpretable Context Methodology. I also, and I'll dive into it here in a bit, I do have a GitHub repo skill that is more up to date and is about how to orchestrate AI systems and processes and organize your own folder setup, right?
Which I'll talk about again later, but it just allows you to ask your AI to help you build these workspaces that my research paper talks about. So essentially what you want to understand and what I'm trying to do, we want to understand the systems of software, not just AI, not just the business, but the entire system. We want to be able to orchestrate.
processes, outcome, and instructions. Currently, lots of people use agents to do that, but here's what we get problems from. First, a lot of companies that I'm working with and I see out there, and again, this is important for you to know whether you're going to be a freelancer, whether you already operate an agency, if you're in a company or you're looking to start one.
These are the pain points that you can solve for and people are actively paying for, but let me explain what they are. Pilot purgatory, right? It's this idea that essentially you start pilots, you're making multiple co -pilot chats or Microsoft agents or multi -agentic workflows, but they end up piloting and it sounds great.
They spend some money on it and then it just kind of stops being adopted and stops being used as much. And it kind of just goes into this loop. They're just demos, which are impressive, but the demos don't actually go anywhere.
The other side of this is lots of companies are building good AI systems, good agentic processes, But the knowledge they're generating, the processes, they don't really go anywhere. There's no way to really read them all.
And companies are starting to build out processes that organize context, right? Different teams, different areas. I've built a few myself.
But those are an issue too. The other side of this, token burn. It's getting to the point where people are just burning through tokens and they're spending so much money for so little outcome on the hope and the plan and the marketing that it does work.
Now, there is some situations where you can throw money at it. You can throw agents at it and you will get a good outcome. But why do you want to continuously do that?
Why not look at pricing and predicting and building a proper company so that when prices do go up, and they will, you can actually have a system for it and understand that, right? Lots of people is and thank you, Dylan. A lot of people who've used my paper, my research have actually blown away how much token burn, not just because it's making the AI more efficient.
It's making your distance to the outcome shorter. You're making the necessity of needing the AI to do so much less while still keeping the speed and efficiency. And so we're just changing the way of thought and structure so you get to your outcome faster or more efficiently.
The other side is you have one person who's an expert in it all, and then they hand that over to someone else. And they don't know how to use it. Or it's like, oh, I could do it faster myself.
And it just dies right there. They go on holiday. No one can ask questions, right?
There's no responsibility, which technically falls into the governance, right? Who's responsible for it? If something does go wrong, if I put in PII, if they're stealing our data, if I'm using legal, there's so many security issues.
There's so many problems. And that is a real. problem area that's going to exponentially increase as more laws come out to structure them and go away with them.
The final one that I see all the time with companies is they go through, they learn the first five problem sets and then they go, well, we didn't do it right. Let's rebuild. And they do it again and they do it again and they do it again.
And that's really just something that I see constantly. The other side of it is what I chatted about. Everyone's talking about features.
not the reason for those features, right? N8n, Zapier, Crone, Hooks, Agentic Loops, OpenClaw, Agent Swarms, Vector Databases, Ragstacks, Memory Plugins, Chatbot UIs, all sorts of different labels and names, Second Brains, Kaparthi Wiki, LLM Wikis. All those are, are different people's solutions for a series of problems.
N8n, Zapier. Cron jobs, scheduling, hooks, those are all names for different solutions built by people to solve the problem of automation, specifically around scheduling and triggers. These are software fundamentals that have been around for decades.
These are just newer tools or different company solutions to the problem of scheduling and triggers. And now, before I go through the rest of this list, you're like, well, why, Jake, why are you telling me all this? I came here to learn specifically AI and this stuff.
If you want to use AI well, you have to know the why behind the how. The how of software, of processes, of scheduling, of automation tells you what is useful to you and what you are lacking in, what you should spend time learning. Because right now, everyone's telling you everything you need to learn.
What you really want to do is simply get a baseline of why things are the way. And see where you can dive in and learn better, right? Exactly.
Treating a symptom, not the problem. And Jorge, yeah, dude, me and you will get lunch or a beer after and we can dive in deeper, my friend. No, you got to come up to St.
Augustine. So anyway, the how, the why, the history, all that good jazz. Open claw, agent swarms, agentic loops.
Anytime you hear those words. Those are solutions. Those are people's bespoke solutions.
And the same way ICM is my own solution for orchestration, deciding what does what and when, who connects to who. The vector databases, rag stacks, second brains, big query, all of that is just context delivery. How can we get a data?
sent to the model? How can we make sure that the model is seeing all the right data? Where can we store that data so that our orchestration and automations, again, whatever you want to name those as features, can move that to the model, right?
Open knowledge format. Exactly. OKF actually uses a similar method to my paper using markdown files and YAML things.
I love OKF and I'm writing a new paper that shows. how you can kind of use those together. And that GitHub repo that I shared also describes a way to put it at the bottom of some of your projects.
But anyway, the last two are memory, state, and interface. So technically speaking, the vector databases, the RAG stacks, memory plugins, those can also be part of memory and state. But the idea is what is the ground truth?
How do we have stuff that we can verify against? we know is real, we know is proper. And the final is the UI.
How do we talk to any of this stuff? How do we route all of that together? So essentially one concept that I talked about here, and I hear all the time, RAG this, RAG that.
RAG came out of paper about six years ago, four years, six years ago? It's almost six now, I think, in a paper. And it was just explaining, hey, You know how we're training all these models and we're training them all this data.
Well, that's expensive. And sometimes they forget stuff. What really is useful is if we put stuff into its context window as it's generating, it works pretty well.
So what if we were to retrieve things so that we could augment its generation? Retrieval. augmented generation.
And there is thousands of ways to do that, right? Vector search. These are really crazy mathematic databases that allow you to search through terms.
That can be a RAG function. SQL, BigQuery. That's what OKF, open knowledge format, if you Google that.
And again, I'll be sending a term sheet out after this if any of this is confusing for anyone. That's how you select data. That can be used as RAG for an AI model.
Folders and files. My method is a form of rag. This is how you organize context ideas in a simple folder, a zip file.
The AI walks through it and gets the information, retrieves the information to change its generations. And the biggest one, APIs and MCPs. That's what a lot of people are using for this agentic process.
This is how we get the data to it and where it needs to be. Jake, the difference between ICM and OKF. I don't want to go too deep into this, but this is an important process I am going to talk about here in a few slides.
They're technically two sides of the same coin. And so I'll talk about that in a moment. But that is a good question.
And I have a slide that dives into my concept. So if you don't mind being patient, I'll explain it. And Tony, absolutely.
I'm so happy that you're getting all this and it's making more sense. That's the whole idea. So the swarm, right?
This is what most people have been building. I've built a few of these in the past. This is all those processes.
And Kevin, sorry, I meant to answer this earlier. Yes, September 12th. And we'll be going more about the Lyceum at the end here.
That's the thing that I'm selling. But I'll talk about all the details and everything with that later. I just want to focus on the free stuff, the good stuff here too.
But yeah, people will build, right? They'll use Langchain or Langgraph. or their own custom Python builds, or any of these different things to create a swarm, right?
A whole bunch of agents that all get context at different times and do different things. But the problem is, if you ask why Agent 7 did what it did... how much it costs, who approved it.
A lot of the times it's hard to track all of that. A lot of the times it's where it is. Yes, Jake, this is being recorded and we also have more webinars in the future as well.
So go ahead, hit up your work call. You're more than welcome to shoot right back as well and you'll still get the links and everything and the information.
Swarms can win, right? This idea of agentic architecture, there is niche cases where it is wildly useful, right? I've built a couple of those data pipelines myself.
It's when you need really specific, high concern, concurrency, millions of people using one type of process, then it kind of makes sense to use that type of stuff. But 99 % of the time. You really don't need to.
And so that's where my ICM, this folder methodology, if you've heard of skills in Claude, they're using this methodology comes in, but it's not new. It's actually comes from 1970s engineering. I didn't invent anything.
I merely applied good software engineering and conceptuals from the 70s. to the modern day. You have this idea behind Unix, right?
And these ideas of rules, right? Each program does one thing well. Folders of Markdown, technically are those rules applied, right?
Plain text is a universal interface. And that's been the case. Think about it.
If you were to get rid of files and folders on your computer, how much of your workflows and things would disappear? I just said, hey, we don't have folders. They don't exist anymore.
That's a problem for a lot of people. Like that's a hugely useful technical feature and we take it for granted, but that had to be developed, created, standardized, right? And you have all sorts of amazing different papers to look into that.
So I just wanted to kind of explain why that's important there. And David, thank you for answering that question. ICM is a process.
OKF is more focused on retrieval. Such a simple, and this is why a software engineer can describe things simply. Thank you, David.
Orchestration, orchestrating AI actually can be done with just a file system. You don't need all this complex software and AI and Landgraf stuff. You do in certain situations, but if you don't know why you need it.
you probably don't need it. The folders can structure hierarchy, who, where, what needs to be read. One agent can wander through those folders and become a million agents based on a question.
Let me kind of show a quick example of what I mean there. So this is my personal little. ICM I built for my company during our fundraise.
It has all our metrics. It has who we are, different evidence. I even have pitch decks in here.
But the overall thing is saying my CloudMD simply says, hey, this is what this is. This is where you need to go. And here's what the folder structure looks like.
So I can take any AI, not just Cloud, if this was an agent's MD or anything else. And this is a zippable file that I can share with anyone else. I could open up Codex.
I could open up Claude. In this case, I could open up a local model. This is not an agent.
This isn't something I coded. This is a fresh AI reading a fresh zip file. Hey, what is our slide deck on the fundraise talking about and who manages it?
So this would usually require a bunch of things. What is our slide deck? Traditionally, would have had to be context around it.
I would have had to give it a persona. Where is that slide deck? What is it doing?
On our fundraise, what does it mean to be fundraising? Where is it there? How are we doing it?
And who manages it? We'd need a whole other separate file place for that. And then I need to create an agent that has all of that information embedded into it.
Or when I ask a question, the agent... retrieves the information and becomes the agent that I would have built. So it goes through and it reads, oh, well, you have a seed deck storyboard.
Let me go read through that. Oh, the seed deck storyboard talks about check and tax timings. Maybe I need to read that and memory as well.
And so it's going through and as it navigates the folder. As it goes through, it is learning what it needs to be and immediately is linking me to those files, telling me what those are, how it is, who manage it, who I am, who the CRO is. Again, not an agent.
This is a fresh AI becoming an agent because of a folder. Something that used to take all sorts of crazy harnesses and, well, you still use one agent and one harness, obviously, but you don't need multiple agents, multiple processes. It immediately is diving into it, right?
It is progressive disclosure is an interesting way to look at it. Yes, exactly. And then it's not reading everything all at once.
It's only reading what it needs to. And then when it edits something, when it creates something, when it moves something, it actually adds it to the folder structure. So the skill becomes a process.
The folder structure becomes the architecture, the orchestration. And if it makes sub agents, well, guess what? They immediately read the files and become the same thing.
That can be confusing to some people, so I broke it down with an analogy. Essentially, it's like walking into a library. And this is not just for ICM and folder structures.
This is how I think everyone should be imagining building with AI in their company, in their institution, where they are. First, you need a front desk. that you can go and ask questions or try to get information from.
Then you need some sort of catalog, right? Good old Dewey Decimal System. I don't know how many people are young in this chat and don't use that anymore, but this idea of being able to know where things are, what it is, why it's there.
It's not the whole book. It's not the whole bookshelf. It's just a description of what should be there.
Then the request slip. Hey. I want to be able to get that book, download that book, bring that book here.
Let's read through it. Maybe we need to go to another library or go online and pull that from somewhere else. The model itself has the weights to understand.
As it reads each of those, as the librarian walks through the library, it becomes more knowledgeable. We're using the power of the LLM to generalize and then specifying it by. the files we give, right?
One librarian walking through. If you can create that, you can start to map it to features, to things you need to build. So you might want a front desk, which would be a UI.
Perhaps it's just a chat with Claude. In this case, my front desk is just the extension and VS code attached to my folder. For you, maybe you've downloaded Claude or OpenAI or AntiGravity.
onto your uh computer and so in that case this is your ui just the good old chat perhaps you built a custom one yourself that just connects to your ai your one agent and it's just a nice place to talk some sort of front end that is the front desk the catalog would be something like this folder an icm it describes the steps the processes who owns what and then directs you onto how to make request slips.
So that's where scripts could be. That's where, hey, go access my Google Drive, or hey, this is how you access this data point. Some people are...
adding big query and SQLs at the end of that to have really big data sense and things like this. And that allows you to really clean it up. And exactly, Jason, you can use open router, open code.
You can use Kimmy K3. You can use Claude. It doesn't matter.
The data structure and context underneath stays the same. It stays yours. So if Claude goes down.
just hop over to the other one. It will work the same because we are ignoring the agentic infrastructure that people are building and only relying on the most important parts, the ability to read and walk through the library, right? That becomes amazing.
Now, if you start a new session, which is a good question here, Dustin, and why a lot of people are able to save so much money as it's wandering through, if it's just asking queries, I can keep the session running. But if I say, hey, can you add a new file talking about this? It adds it in.
I can close that session. That file's now there, that walkthrough, that process. Opening up a new session, boom, the memory's there.
It doesn't need to remember everything else. And if you do, you just write it down into a file or add it to the database that's already connected to it. So you don't need to have these huge, massive context sessions open.
Or if you do, you can keep them saved and local. Right. You can keep it much more clean.
And that's the ideas as the agent or your single model with a good harness reads everything. It's simply becoming the agent you need. You get it done.
You drop it. You move on to the next part. Important thing here.
It also keeps you in control. It lets you be able to sit there and ask the process. Right.
Imagine if your company had a brain and you could ask, hey, what did we talk about in that meeting? What who did what? Right now, AI has to connect to all these different things.
It might get it wrong. It might not be able to see it. But if you can map the structure underneath or at least make a version of it that's simpler in a folder, you can then add the complexity ons after.
Yes, we should have the recording saved. I have to ask Matt about sending it out and how that works. But at the very least, I'm...
right after we're going to have emails on all these ideas, concepts, and things like that. And of course, if you are trying to learn more, uh, shameless plug, um, if you're not already in the community, cleef notes, uh, I have literally hours and hours of free stuff. Like it's free to join.
I have some stuff behind paywalls cause my time is limited, but most of this like long sessions. whole modules, quick starts, videos. I go into the history of Python.
I go into how to build folder structure for 20 minutes straight in a video. Go check it out. It's amazing.
Lots of people are loving it in there. You're more than welcome to. So if you have any other questions, that's also a lot more asynchronous.
And there's a lot of other cool stuff. We do competitions, things like that. So yeah, definitely, definitely go check that.
I just dropped the link into chat or you can just type in cleef notes as well. You said, what is the GitHub repo? So that was the repo here.
This one specifically helps you. You can talk to your AI. You can give this to your AI.
And it's a skill for Claude if you want to use it that way. Or you can mold it into something else. But it helps you build out these workspaces based on your work.
So you can ask it questions. Or it can restructure one because a lot of people out there are already doing this method and restructure it based on how I've seen the highest levels of efficiency, right? Forms .md.
So that's that GitHub repo right there. And it's also nice to read through to see how I see the forms of data structure and context working because this is based off of like actual research. This isn't me saying, hey, AI, go look at all this.
This is like, oh, that one's good. Let me record that, like traditional academic stuff that I love to do. So yeah, go check it out.
It's a really great repo. I think we're at a couple hundred stars. Actually, hold on.
I only posted that last week. Let's see. How are we doing now?
Let's see. Yeah, 353 stars. Amazing.
Thank you, guys. That means a lot. I'm glad people are liking that.
So Clayton, what if you're storing context for longer periods of time? Absolutely. So that's where you would want to have some sort of actual stored area.
So there's a couple places that people are doing it. SQL normal databases and at the bottom of markdown files, right? Like let's say in my folder or in my process, in one of my markdown files, let's go to something here.
Let's say we're going in here. We have entities. We have people that are in here.
And let's look at David. Let's say I don't have that much information on him. Simply, I could attach his GitHub repo here.
I could attach a database that I'm constantly updating, like an Excel sheet. I could have a link to an Excel sheet here that's public, a SharePoint, right? It's up to you to decide where you like or where your data is already stored.
And all you have to do is just add a little process here. And that's actually where Google, OKF, so Google has obviously loved some of my research at their lab there. That's where they've been like, well, hey, we got this big old BigQuery database that's storing things and connects to your Google and Gmail, why don't we use Markdown Files and YAML, which I don't know if that looks familiar to read my paper.
It's everything you're seeing. Well, not there. That's HTML.
Everything you're seeing here, right? To simply describe where or what type of database is being used. Here's the link for it.
Here's what you might want to ask about it. And so that's how you're able to create huge connectivity to something more moldable. Because not everything should be in Markdown files.
That's just kind of one of those processes there. But really good question. Another Markdown thing.
Yeah, exactly. Right. It's annoying because it's like, oh, look at this crazy new stuff.
And it's just Markdown and Unix methodology from the 70s. That's why I'm trying not to hype it too much. It's more of a description of like, it really does work.
Thank you. Thank you for getting me to 360, Jared. More stars.
And feel free to fork it and make changes to it or argue with it. I'm here to build it and make it stuff better. And again, so that becomes all you need is one good model with one good harness.
And it can become as many agents as you can. And I promise you, the reason I came to this isn't because I thought it was the best way. It's because I kept making agents and being annoyed that I wasn't using them or it just was off slightly.
And I had to, it was like, okay, well, then I'm going to need 10 ,000 agents. What if I made an agent that made more agents? And then I was like, wait, why don't we skip all of that and just put all the data and context into a way where one agent can read based on my question.
And that's the simplicity of it. That's why it works, is it molds to your intent. It puts you at the center of control and skips all the hubbub of what you need to do to get to the other absurdities.
And again, as you asked, at the bottom of that library, instead of just markdown, it could connect and have instructions. Hey, if he asks a question about this data, if you want to go get update, just go to this Google Drive, go to this Google Sheet, go check out this big query. It'll go catch it, bring it into its context window, and then wander through.
Bruce, I don't understand GitHub. So I will say, I usually don't share this because we're working on it. It's in alpha.
We're going to be probably launching it this week. Actually, David told me to launch it. It's ready now.
It's kind of like a better... I don't want to compare it to GitHub. We're not trying to replace GitHub.
But the idea is like, you can just sign in, upload your folders, and essentially be able to kind of just... have your chats, right? Like if I want to go ahead and upload that ICM that you're seeing right here, or like one of my folders, I can just hit Okay, let's go ahead and let's do this one, right?
And let's throw it in. And then I can just chat with it like I would with AI. That's kind of fun.
Hopefully, we'll be releasing that to the community soon. But I don't want to spend too much time talking on that. We were doing it because a lot of people didn't like how GitHub or BigQuery was working with them.
And so we're making a deployment layer. David, actually, David, if you can share your YouTube, he does dev blogs on it. Uh, which is really fun.
They're very, if you're not into the technical stuff, you might not like it. Uh, I think we've gotten a couple hundred views on those videos, uh, because they're meant to be explanatory. Some of them are really just meant for me.
Um, but that's the kind of idea. That's the other thing too, is, uh, trying to get away where we made it very secure. David has been building it all custom himself and C sharp and having control of the back ends, obviously using up some Azure.
So there is Microsoft in there, but trying to make it clean, but. That's off topic. We don't want to go too much into that, but do keep an eye out for it.
Hopefully it'll be ready for me. I need to make a video on it and get it launched to alpha people. But again, this all comes back around to one thing.
The best software in the world, the best products in the world do not just simply automate. Automation is important. They augment.
They do something for the human that allows us to do. more. That is the best type of software and products in the world.
Right now, people seem to be so focused on features and AI things that simply automate. They don't think about the next step, which is augmentation. So I like to see, and in my paper, a lot of the times A person spending so much time in direction sending, in intent.
If you can build it and make it easy for people to share their intent and what they're doing, that's amazing. The machine should carry all that judgment and translation work. It should translate all of that into some sort of artifacts, processes, areas.
And then finally, the human should have an easy way to come back in and review it all. With all these agentic systems, it's super challenging because I can't always see what it's doing, where it's doing. Here, I can see everything.
I can read everything. And I don't got to switch through a whole bunch of code. Yes, there's some markdown in here, but I could render this in Obsidian instead or in some, oh, that makes it seem way more confusing.
I shouldn't do that. But like something like this, right? And I'm like, oh, okay.
It's really easy to go see what's going on here. That makes it 10 times easier. And then also, guess what?
If AI breaks down tomorrow, we still have a whole human readable process. We can still follow the pipeline.
We still have SOPs. It is resistant to problems, and this is kind of continuing, right? How do I give database tier context?
For example, Oracle Database with 10 ,000 tables and 100 ,000 golems. I'm so sorry. Fresh builds with integrations.
One of the biggest things for understanding SQL scripts and being able to actually pull and extract out of databases, there isn't one for Oracle yet. I can imagine Oracle is going to be very quickly behind. I'm going to share that.
And again, this might be too technical for some people, but this is a good question that I want to hit on, Justin.
This they are looking at ways they actually have a GitHub repo, the Google people to go through these databases and organize the markdown files to then. So it's not that you are organizing the database as much, but you are organizing queries for the AI to be able to navigate the database much faster, right? Like, hey, if you have that big Oracle database, it just says, hey, by the way, this is where orders are.
This is where you should find it in that Oracle database. Again, this is. Google query, but same concept here.
Why not organize it from a description standpoint? So that way, you don't have to spend as much time ripping it apart and refactoring the whole database. Still a challenge.
I'm sorry that you got that onto you, Justin. That's That's just hard work.
But hey, job security. So that's nice. There's a lot of agents and processes and tools out there.
But again, I don't want to go too much down the rabbit hole on that. I got to stay on topic. I'm sure Matt's already yelling at me or thinking about it.
So this also goes into a concept here, right? This augmentation, this idea of humans using this stuff, the fact that all of you are even in this call to begin with, right? It's because there's these problems.
We've noticed if we go back to that first slide that I talked about, projects failing, it's because they weren't seeing the productivity. They weren't seeing the process, right? That is the same exact thing that happened with computers in the 80s.
When we started building computers and selling them and everyone's like, we got to have these. We actually saw a drop. production.
And there's something called the solo paradox, which is essentially this idea that we have these data on productivity, but there was seemingly no productivity anywhere because we didn't understood how it fit into something. We didn't understood how our workflows, what augmentation meant. Once we understood that, and after solo said, you can see the computer age everywhere, but in the productivity statistics, after we understood the augmentation, how we give a computer to someone to do more of their job, better job, rather than just replace them, that's when we see the surge, the productivity, the value increase.
And that quote could be the same. We see the AI age everywhere, but in the productivity statistics. Companies are learning how to change workflows, work with these systems because they work in a different way.
They automate things slightly differently than how we ever had before. But it is understanding where we sit in it, where it unlocks it. So one argument or comment that I see a lot, um, and also chat to, to answer your question about skill opt.
I love skill opt. I think that's really cool. That one's really technical.
So I don't want to confuse the non -technical people in here, but I do think that's a cool paper and I'm actually building something that, uh, pretty much, uh, automates the process of like refining your markdown files, but I won't go into that right now. why, how do we survive these updates? Why does the folder structure survive the next model update, right?
New model drops, Opus drops, Fable drops, all of these things. I don't care. I have had the same folder structure since Sonnet 3 .5.
Obviously, it's gotten a little better, more refined. But when the model updates, my context, my opinion, my files, the way I like to do things, the thing that's not being trained for in these models is the useful thing. And so that model simply just gets better at executing my process.
Further, folders are very easy to condense into each other. If I delete a couple of markdown files because I created a tool or a skill, and now the model does it naturally. I just remove that folder and just make my instructions cleaner.
It actually means every update means less tokens. It means that my system works better with these model updates. You can swap between models.
Obviously, you need to make sure the model can read, write, make bash commands, has a little bit of a harness. But it's that beautiful process there, right? You are controlling your data.
You know the process. And that is going to be something that is challenging to automate at all times. You're building that.
And again, you don't have to trust me. I hope you all have been seeing in chat if you're new to all of this and seeing all of this. This work has done a lot of work for a lot of companies.
I built out inside of things. Obviously, Google themselves is taking on this methodology because they really like it. Anthropic themselves are building around this methodology with skills and their project files and all of these things.
I'm not teaching. my way. I'm teaching a standard that I'm seeing other people doing, and I'm formalizing it.
I'm making it digestible. I'm taking the time to help standardize, right? There is easy work and hard work though, right?
The paper, the repos, all of that stuff is there, right? I can help you with the structures, but what I can't help you with is your problem, your process, your loop, what you think best is, what your opinions are. You need to figure those out or at least be able to ask yourself those questions.
Decide what is valuable. What I kind of mean by that is this concept of, hey, well, if we're sitting here, and I was to ask you to help build me a better drill because I don't like how it's putting holes in the wall.
Well, instead of sitting there and saying, okay, cool, and spending all your time building a better drill, finding features, researching batteries, you need to ask me why I need a hole in the wall. You should ask yourself why you need a hole in the wall. And then you can then find out, well, I need to put a screw in it so I can hold a painting.
oh, well, you don't just need a hole in the wall. You need to hold a painting. Well, here's some 3M tape so you don't even need to worry about the drill at all.
You could be optimizing the wrong thing. And that's where automations become a problem. You could become automating something that never needed to be automated in the first place, right?
This is called job to be done. It's a really cool framework that I very much like. And you want to be very careful, right?
The biggest thing, if you just want to ask yourself a question or you ask your clients a question, What would you hand off, right? What is one workflow you would hand off tomorrow and you would do it because you don't want to deal with it anymore, but it is super important for you?
Well, just by thinking about that, you can look at it. And I am kind of curious. Let's see, can you describe in one or two sentences a workflow that you think you shouldn't be doing anymore or you could make better with AI?
I want you to place that in chat, not just for me, but I want the other, 300 people in here, 284 people, to see that a lot of us might want to offload something simple. We might all have similar things we want to offload, right?
In one or two sentences, just describe, hey, well, you know, I, for some reason, every day log in and do this, right? Jason, manually moving and renaming videos and adding bumpers to the end of each video, right? That is a workflow.
It is a process. How Jason likes to add it in there. The way an AI might not, when Jason says this is important, the AI doesn't know what is important to him.
That's where you would find a markdown file describing how, where, what does he like? When should it be in there? What are the type of videos, right?
Profit and loss analysts, okay? Profit and loss for who? What's important about profit and loss?
OCR batches of files, right? What is the OCR for? Why do we need to OCR it?
Are we trying to get information out of those files to put it into a database? Are we trying to give it to people? Are we trying to templatize it?
Is it a workflow for someone else? They're paying us to OCR, so we don't care what they're doing with it, right? These are the questions.
The why behind the goal is just as important behind the goal, right? In a world full of cheap answers, questions become valuable. Let's see what we got here.
Bumpers, man, absolutely. Uh, collateral perfection line filling. Ooh, wow.
Uh, pulling in your latest email, helping me summarize. Ooh, yeah, that's, that's, that's a big one that actually can be done even without some of the folder structure. Uh, right now, actually, as I'm talking, I have, um, one of my AI going through and, uh, for our Lyceum, we're building second brains for everyone, which I'll chat about.
And I'm having it draft emails based on what I wrote. for each person and adding it to my Gmail as we're talking so that after this call, I can go through and verify that they're all good and then start sending that to the people who've already signed up. And as you can see, while I've been sitting here, it's already created.
Yeah, it's created almost all of them. So now I get to go in and verify, all right, is it sending the right one to the right person? Do I like it?
And it's in the template that I like with the information that I want, the way I want, with the links to my Google forms and processes, all with one AI, not multiple agents, not multiple crazy N8N connections. This is one AI. One chat and a folder.
I made zero automations for this. I just connected it to my Google Drive because those integrations, those MCPs, they already exist in harnesses, right? I'm going to share how I did that as well, probably in the next day or two.
But that's just, you know, those types of concepts are amazing. N8n, so when I say you don't necessarily need N8n, it doesn't mean it's not useful. It means a lot of the times you're building a solution with N8n.
that you don't need the complexity of NAN for, right? NAN is a solution for managing triggers, processes. If it is making you money and working, great.
But think about, is that what you need? Is it too complex? Can you do it on yourself?
What is your outcome? Are you trying to make money? Are you just trying to make your life simple and faster?
Was the time you put into building that NAN too complicated and didn't give you that workflow? If you're getting value out of it, that's great. But at the end of the day, most things...
do not require complexity to solve, and we solve them complexly, right? Um, thank you everyone for, for kind of diving into this. Yeah.
Python fast API, but maybe you could replace Python and fast API with something even else, just a simple integration or a query because Claude or open code already has a connector to your Google drive. So you don't need to make a Python connector to your database or something like that. Again, you'll, if you know you need to make it because there's no other option, then you'll know.
But if you're like, oh, well, why am I building this software? Do I have to build it? You always want to ask yourself that question.
And again, we chatted with different people. We talked about Curtis at the beginning who saved money. We have Mark in our community.
He ended up being able to actually get promoted and won People Choice Award just by using these systems and processes. We had another individual who was trying to sell these processes and ideas and ended up getting even more promotions inside of his company. So even outside of starting your own company, making money on your own, other companies are seeing value in people using these systems.
And these are people in our community actually getting it. Again, I'm outcome -based, not feature -based. Now, how we did it, you can go read about it in the communities.
We have a win section. You're more than welcome to dive into it. But that's something important there.
And again, the intro I showed you earlier today, this very slide deck was built using this system. It makes it much easier to navigate, to control, to be able to give something for you, right? So this is the part of the webinar.
Right? We're hitting the top of the hour here. I pushed it well over the original time, but I hope that was a solid amount of information.
I am about to sell something or tell about what I am selling, and it's called the Lyceum. So again, we're going to be emailing all sorts of links and everything to people, but I hope you stick around at least to listen and you've learned a lot because I'm still going to teach a little bit here. So the Lyceum is my solution to the fact that we have 40 ,000 people and hundreds of them every day are saying, Jake, we'd love to spend more time together.
We'd love more live sessions. And we're doing high teas on the weekend, but they're kind of. very good lectures.
They're just meet. We sit in there, we talk, we chat, we solve problems, we ask questions, but we want a curriculum, right? I'm working with these companies and building these long programs that end in actual outcomes.
I've been working with universities and building curriculums so that students actually build and learn. So I wanted to be able to build something similar myself. And I get it, courses, all that good jazz.
But I'm taking this from a course of college perspective of building a business perspective. And it's going to be 12 long weeks live with instructors. And everyone's going to have a working build process at the end of it.
But there's different cohorts. And each of you is bringing your own project, your work, your goal. to it.
You need to have a goal to even get in. We've been turning away people who don't know what they want to build specifically so that we can focus on the process. Now, my pedagogy, my process behind it is we're going to have three cohorts, not just simply to keep people who are more advanced or less advanced in technical stuff together, but because of what our goals are.
It's the same curriculum. across all of them. I am teaching roughly the same ideas.
I'm lecturing. I'm creating breakout sessions. We're doing work with other instructors and people that I'm hiring in.
But the time we spend on each part of the curriculum, whether it's security, SOC, all sorts of compliance issues, data sensitivity, or if it's business, ROI, who do I hire? I want to, I'm starting a new business or I already have a business and we're trying to build out a better product inside of our business. Or I'm trying to build systems that Allow me to share my work.
I already have a business. I've got some technical stuff down, but I need more attention, right? I've been able to succeed pretty well in all of these.
And I have friends who've succeeded even more that we wanted to teach about these processes around the concept of AI, around the concept of ICM. What does it look like to be technical, to be business oriented, or to be a creator in a world where AI and these processes exist? How do you implement it?
And so the idea there as well is if we're going to be talking about AI, we should be implementing it. So I'm every single person who joins, I am and already have been building a second brain for. We interview you, we have you do questions, and we're literally going through and taking a template of each person so that I can go through and I understand what your pain points were from the beginning, what your goals are.
where you are working, what you're working on with. So at any moment, I can build you a bespoke slide deck to take home with you in week three. Or if I hop on a call with you or someone else does, they can query it real quick and we can know we're not wasting time.
Oh, what are you doing? Oh, right. That's what you were on that student.
And then we can add your learning, your process into a second brain that is handed back to you at the end. So you have all of your learnings, all of your girls to mark the process of it. This allows us to track better.
This allows us to build more bespoke situations for you. This allows us to verify what we're actually building with you at scale. Obviously, I can't have that many people in, but it allows me to have a lot more, right?
This allows me to give some attention or at least the outcome that usually one -on -one would have been. Now, good question. The data, if we are not giving this to OpenAI or anyone else and...
I will not use Claude or OpenAI or anything on your data or your persona if you do not want me to. I have local models set up for it. This data I am not using to train a new model or anything like this.
The main things I'm going to be using this for is to connect people together who have built similar connective processes. right? I want to be able to share if all of you built really cool stuff and I see overlap where your company, your process, your things should work together.
You should be knowing that you should be working with each other. If someone in the technical cohort is building something that solves a problem of someone in the creator cohort, let's link you two together. This allows us to create context and connection between all the students, all the processes, and then you get to have it.
Now, obviously GDPR, these different things, certain areas where you're coming from, I have to delete certain amounts of data after 30 days, 60 days, 90 days, certain data I can't keep at all, right? I'm managing all that. We're working through that.
But the goal is to give you that data and that package so you can hold on to it. And of course, if you don't want us to hold on to it or you're worried about the privacy things, book a call. We can talk about it.
We're working with NDAs and processes, but I'll dive deeper into that in a second. every seat is getting its own program, right? We're giving you a series of answers.
We're giving you your own brain. We're giving you your own track inside of the cohort and your own capstone. but you're all gonna be coming together as a class.
We're all gonna be learning together. We're all gonna be working together if we need to. You don't have to share what you're working on, but I promise you sharing your work, listening, hearing from others, that is the best way to move forward.
I promise. It's not just me looking at it. Having peers look at your work is a great way to keep yourself accountable.
We want you to leave with a build. not a consulting minor, not a nice slide deck that tells you, well, this is what the next six weeks should look like. We want you at the end of 12 weeks to either have an MVP, clients booked, or some sort of great account and process set up.
And if you do not... We are guaranteeing you that we will work with you until it is finished. If you show up to all the classes, I am not letting you leave without something that has gotten you either the money back that you spent, your goals built on it, or at least some sort of research, right?
One of my ideas is actually having the technical group deploy a real system, measure it in the last few weeks. The business cohort being able to repurpose a product or process and get someone on retainer in the last few weeks. I can't guarantee it, obviously.
I'd love to guarantee it. But at the end of the day, you have to put in the work too. I need you to be there.
If you do the work, you will leave with it working. I can guarantee that. Whatever that process is, whatever you're trying to build, I will make sure that that is there.
And again, the way it looks, we're going to have 18 hours of live actual. Instructions, lectures, processes across nine main sessions. I'm bringing in instructors from Harvard, from the University of Edinburgh, from businesses that I work with, from artists that I respect.
I'm bringing in, or we already have a private Discord where you can all meet each other before the class even starts. You're able to collaborate and work for it. We're giving you all of the recordings, not just from your cohort, but the other cohorts.
We're giving the curriculum, the workbooks. We're also doing a certification. because we have so many companies looking at us now working with us.
They're asking if we've been certifying people so that they can trust that they can work with them or hire them. And so, yeah, Trent, that we are doing a certification. I don't really talk about that much because certifications only mean as much as the attention you're getting.
Luckily, we're doing millions of views a week or a month. So that helps us get more attention and also the process of it. But we did want to offer that.
We're also giving everyone lifetime VIP to our community. access to cleef notes. And of course, we're trying to connect you all after being in that alumni seat, right?
This is an actual, right? These are things that I pulled from college, that I pulled from businesses, and I wanted to put into a solid, fun program that I would enjoy to do and go through it. Also, I like competition.
And one thing that I think college needs more of is competition. And I think it should have monetary value to it. So we are going to do some prizes as well, right?
Cohort third place on each of the cohorts is going to get a year of Claude Max. Just here you go. And we're going to be judging the distance you did from where you started.
to where you ended your progress, your chains. And we will be giving you kind of rubrics and stuff for that and what that looks like. The other side of it, we're going to give you a second place is going to get a Mac four or Mac mini four M four pro and a year of Claude max.
Plus we're going to be talking about your projects and sharing them to our group. First place of each cohort is going to sit down with my enterprise company, Aduba, and we're going to sit there and do $50 ,000 worth of our billable time to make sure that you have an extra build onto it, a whole process for it. The grand champion, we're going to spend $100 ,000 of our own time and effort into building out your progress, your process, case study, doing my research and things into it, because that is...
Awesome work. And that's honestly what some universities do do in competitions and things like that, right? That's what we want to get there for.
Now, here's the thing. This is who it is not for. I understand times and things like that.
But if you're just only going through recordings, you're only looking through recordings, go to my free stuff. Go to my YouTube videos. Don't pay for this.
We want you involved. We want you part of it. If you're not willing to put in some work and actually try to get this going every week, right?
I'm sorry. Like effort is tied to success. I can't have you in here.
If you're just chasing prompt tax and just surface level of ideas, we don't want to keep, we're not letting everyone in, right? We've already turned away a lot of people. you're just going to get frustrated because of the other people around you not building like you are.
And if you want to go chase and learn more and do that stuff, I'm not saying that's a bad thing. I'm just saying that's not what this is about. If you don't know what to build, I'm going to be honest with you.
Go into my free stuff. Take your time. Try to figure out what it is that you actually want to work on.
If you have too many ideas and you know there's a lot of good ideas and you don't know which one to choose, that's fine. We can work with that. That's easy.
But at the end of the day, I don't want you to necessarily dive into that. Also, if it's like a really big project and you know your company is going to need a whole enterprise from this is where we actually need multiple agents, concurrencies, all those processes.
This is not for you. You can reach out to our enterprise team. We can try to work with you and build something out.
But this is for those projects that you know you can build, you need to build. And honestly, those agentic pipelines and things like that. going to be worrying about.
We're worrying about outcome, so we don't have to build those things, right?
Now, the other side of this, right? What do you mean? I'm sorry, Timber.
I'm sorry you feel duped. I don't think this is fake. It's me here.
I apologize. A lot of people are enjoying this, but I'm sorry you feel duped. I hope that you get a lot more knowledge out of my other free stuff out there.
But I'm glad you don't feel dupe, Barrett. I appreciate that. The number is 4750, right?
That's what we're charging for all 12 weeks plus all the builds plus everything inside of it, right? That's going to be for, you know, the whole process, everything included from now, you're getting into it and all of those processes in there. Again, 12 weeks, working build when we finish it, everything I talked about before this.
Important note. Because I have published research, because we work with companies, your company, if you are in one, can pay for this. We already have people getting their company to pay for their training, for their educational side.
We have colleges paying for professors for their educational stipending. You don't necessarily need to be technical. You need to understand ICM and the processes.
You at least need to understand all of my free stuff that's in Cleef Notes. And we'll be sharing that with everyone. We'll make sure everyone's up to par before they get in.
That's why we're starting it on September 12th. But we are willing to do data separation. We are willing to look at kind of these processes of payment plans and things like that.
But I highly recommend like, hey, if you don't want to pay for it yourself, you want to get reimbursed. Some companies are already reimbursing for this, right? If you're unsure, you're not sure what cohort you're in, things like that, you can happily book a call.
Matt just dropped a link. You're more than welcome to dive in. You don't have to make the decision now.
If you do want to make the decision, I think Matt might be sharing it. I don't know if you want to share the buy link, but realistically, we don't want to pressure you into anything. We're happy to sit here like this is a real thing.
We're taking our time with it. That's why it's not going into September. It's not starting next week.
We're doing it September 12th because that's how you take real progress to process. You take your. time.
We do have more webinars. Most of them are going to be very similar to this, right? I'm trying to get new people in new processes.
It's important to note, if you are working, you are running a company. We have people who are neuroscientists who are running a company. We're trying to spend most of the sessions, all of the cohorts on the weekends.
So think Saturday. That was where those two -hour sessions are going on. We are doing some take -home work that you're going to have to do during the week asynchronously.
But most of the live learning will be done on the weekend. That is where we're going to be diving into it. I will say, though, if there is any kind of quick questions.
Uh, the idea right now is I am Eastern. I am on this time. We had to limit it.
And again, I can't take everyone. We had to put it down. So there are going to be from Eastern time.
We're going to be running cohorts from about, uh, I think 10 AM all the way until about 6 PM. Matt, please check me on that on Saturdays, right? An hour of live lectures builds based on what we're all doing, as well as an hour of breakout rooms with actual instructors, processes, things like that.
9 and 11 on Saturdays, yeah, different cohorts, processes. Solopreneurs, absolutely. We have a whole bunch of people who either have a product or already have a service, and they're trying to scale it.
And so the first three weeks. We're actually going through and we're spending time asking the right questions, the right processes, what not to do. How can you build it out?
The middle three weeks is lining out the processes, the ideas. And again, this is for every cohort. This is why the curriculum can be the same.
It's the details that change. And then at the end, we're actually using it, right? If you're a solopreneur, my goal for you in the last three weeks is to have you approaching clients, selling a product, selling a process.
If you're inside of a company, you're piloting it within your company. You're piloting the process. If you're a creator, you should have some sort of videos launched and being launched, processes, ideas, artistic styles.
We have this one person, Fernando. Fernandez, he's amazing. He's doing this really cool art project and dining experience.
And he needs to be able to build a system to navigate all the bookings and processes, but he's an artist at heart. So he's going into the creator cohort because he's creating automations for his art and processes, right? There's a lot of amazing things there.
Your professional development is less than 3K. That's okay. Yeah, like we can try to work with the company too because is it by certain months?
There's ways to look into it. Um, yeah, I find that Eastern is best international time for what we're doing as well. Are we offering this again?
Yes. So this is our first major one. This is the one that I'm diving so much effort into.
We are looking at doing more. However, this is going to be we're spending all of our time on this one. We're not offering different ones.
And, you know, we're doing a 12 week one here and then a 12 week one overlapped with that. It's these cohorts and this one. And that's it for right now.
And then we want to do a lot more in the future as well. I don't have any enterprise needs now. I'm freelance, but I love the content core.
Thank you, Sean. I appreciate that. I hope you learned a lot, right?
Again, I hope even if, you know, you're not going to be a part of Lyceum, I hope you all learned a lot from today. That's really good. Should we do should the project we choose do 100 % be for profit?
Oh, no, no, no, no. There's some people coming in who aren't going to make any money off of it. They just see it as a problem.
They're doing it for a nonprofit. They're trying to solve issues with learning and accessibility. And they just have.
happen to have the cash to be able to dive into it and they see the problem important enough. We're very happy. We're not just focused on profit.
Again, that's where you might go into the technical or the creator one. That's why I separated them because some people in the creator one aren't as focused on profit as they are attention or impact. People in the business one, right?
We want to be business minded. Having peers that think like you, not regardless of skill level. is going to be easier for learning than if we had all the same skill level, but different goals, right?
And that's an important pedagogy in learning that a lot of people. I would commit at least two hours a week to live in -person sessions and then another hour or two per week to be able to actually work and build on these things. I find that...
Some people spend 10 hours, 20 hours building stuff out because they're excited for it. But at the very minimum, four to five hours a week to be able to maybe book office hours, things like that can be useful. We do have a test at the end that we're going to have people take based on what they learned.
And there is going to be a pass fail for it. It is going to be based on what you brought in and how it is. My forms of tests are a little different than most because of what I believe a test should be measuring.
It should be measuring where you are to where you finished. So every test, although it's following a rubric and a template, will be bespoke to each person and their process, which is something that I don't think has ever been done before. And we can only do it because of the data system that I've created to monitor everyone.
And I think that's going to be really fun. What if what you need is a full company Jarvis? All right.
So a full company Jarvis, like second brain company brain, you might want to reach out to me or Matt personally. That's where our enterprise work is. And that's usually a much longer process.
There's a lot of people try to skip to the huge company brain, but there's a lot of steps you have to take before you even get there. A lot of traditional data things. Let's see.
I have been corrected. Negative. Sorry.
I thought this was fake. Oh, it's okay, Timber. No, no.
Look, man. Hey, one, call me out all day, my guy. That's what we're here for.
There's a lot of fake BS out there. There's a lot of people. I will say we've looked at the idea of recording the webinar and then posting it so that it continues to answer questions and things like that.
But there's a lot of problems there and things like that. So I appreciate that you feel more confident that it's legit. Um, yeah.
And again, go check out the free stuff. If you're still skeptical, I don't need to sell you. I'll be completely honest.
I don't use sale tactics because I know my stuff works. I've worked with amazing people. I have so much free stuff that people are getting value of people themselves will work together.
You go into the community, go check it out. Um, I it's, it's okay. Timber.
I appreciate you apologizing. That takes a lot. Thank you.
Um, yeah, take your time. Chew on this, uh, the deadline to sign up. So there's not as much of, obviously, like you have to sign up before September 12th.
We're probably giving it about a week or two is what we're deciding for people to sign up beforehand. But we also have a max. So if we hit the max before the deadline, then that's it.
Because I can't have a thousand people in this. Like there's no way I can take a thousand people into this. The quality would drop.
It just wouldn't be where it is. So there is a max per cohort and a max amount of people that we're looking at. But there is some things there.
Scott, the ICM paper really improved workstation efficiency. I'm so glad. Um, and again, it's, it's a method, right?
It's, there's still ways you might notice things to do better. And that's why I want to get people together. We should be working together.
Um, there is quite a few payment plans. I see Matt's kind of, um, I'm sharing those as well. I'm so glad you learned a lot as well.
Uh, sorry, I missed it. I run my own CPA firm and I have so many things I want to work on a bit crazy. You want to share and fully commit.
What is the cohort cadence? So again, three weeks, week off, three weeks, week off. For a total of 12 weeks, three major three -week sprints, two hours minimum live in -person session each week on weekends.
And then after sessions, different take -home assignments and processes that I'll give to you as well. So an extra couple hours there. We are very aware a lot of people coming into this have a job, are running a company, are running two companies.
And so for all of them, everyone we've talked to that's most likely joining and has already joined and is already part of the cohort, they have basically said, look. cohort that time, those weekends are best for us. And so that's what we're sticking to.
Bruce, go enjoy learning GitHub. Absolutely. I'm so excited for it.
Test score and prize winner. I'm glad you like it, Scott. Um, is there a recording?
Uh, we have another webinar going on. Um, we should be, honestly, we should post one of these recordings. I feel like we should do that.
Honestly, I'll talk to Matt. I feel like we should post a recording and then I'll make a whole new type of webinar for next week. Uh, but I can't commit to that cause I'm quite busy for a bit.
Um, but yeah, September 12th, hard, hard, hard start September 12th. We're going to spend the next month or two getting everyone in processing interviews, right? We want to take our time with this.
This isn't a, a quick thing. This is a process. Um, booking consultations.
Uh, yeah, I mean, you, you could, you could book a meeting with us if you have a bigger company or something that you want more involved. Uh, we do have our, our company, a Duba that actually does corporate and, and larger companies. Um, the other side of it too.
Uh, and again, just, just a reminder, let me share with the cohorts and prizes. I guess I could share there, but, um, Technical business, again, business and creator, some people are technical in it.
Again, it's more focused towards goals. Thank you, Matt, for answering a bunch of these questions. Can you give us an example of a good capstone?
Yeah, so one of my favorite capstones that I'm looking at right now, I actually, she won a, she got sponsored, so she didn't have to pay for her seat because someone sponsored for it, is basically the... Problem in the financial industry in this kind of area here and is this idea of anonymization of data. So she's building out and already has most of the pipeline built out where a local model will take financial data and like kind of compliance ready data, anonymize it.
hand the important data to a larger model so that it can do all its stuff. That will hand it back to the local model, and then it'll input it into its data points easily, the context, the ICM. And her goal is to actually sit into the technical one to be able to learn more security loopholes, deal with SOC compliance.
We're bringing in compliance experts and people like that to chat as well. And so that's kind of a capstone. Her goal is to have a fully deployable system that is...
As close as possible. Obviously, I'm not a trained lawyer. You do need a lawyer to go through things, right?
But as close as possible to being safe, compliant as we can get it. And so that capstone is take my technical project and make it there. Another business capstone is this one company is essentially sitting there and they have...
all this sorts of data and processes, but there's only about seven of them in the company, but they're doing a really high return. They're doing a quite a few million a year. And so they just wanted a really simple, like they didn't care about all that fancy stuff.
They, their data, they're not in a really data sensitive place. So they just wanted to be able to build out a really good second brain of the company, something where all of them, the CEO, the CTO, a couple of other people could just query it and ask around and refine it. And so they're building that out.
And their goal by the end is to have that fully deployed. tested with everyone. And then maybe to white label the process to other people.
Another one for creators is there's a one guy who's trying to actually deploy his creator software. that helps creators build more videos, track videos, track processes. And Danny, the guy who built that amazing thing at the beginning that I shared with you, he actually is building a version of his own.
So, right, it's kind of built around an outcome and a goal and something they've already been working with or want to work with more and refine. That's the kind of goal there. Of course, we're learning the whole way as well.
Yeah, you're welcome. I'm absolutely cash money. I'm here for it.
I'm calling them out. It was the test. Absolutely.
Thanks, Jake. I need a Jarvis. Yeah, I like that people are calling it Jarvis.
A second brain, ICM, context layer. Again, that's just a memory bank and a way in which to navigate it, right? I was explaining it earlier.
I think it really can help. Foundry and Copilot have some options and I've built versions of a second brain for companies inside of it, but it's certainly much more limited. than if you have like Claude Enterprise or a local model.
Let's see. You should take 2 ,000 people. It would be good cash.
And I would love to take all that cash. I'm sure Matt would love that too. But again, I'm also here for quality too.
When are you intending to hold these sessions? Days or times? Yeah.
So one, we're going to be sending all of that stuff out and talking about it, but it's going to be mostly on Saturdays. Eastern time around 9 to 11 is when the main cohorts and sessions are going to be going on. Again, Eastern time, US time.
Jason, I'm so glad you have it. I'm so glad you got a lot of stuff out of there. I know it's been a long day.
I'm so glad that you thought that hour was like 20 minutes or less. I'm glad you think this is as good as a college class, if not better. That makes me happy.
Teaching is my passion. You said September 12th, first cohort start. No.
So all the cohorts, business, technical and creator all start on the 12th. And then we're staggering them so that I can go and be in each one. Right.
So each of those cohorts are all running at the same time. And then we're sharing the videos into the discord of each cohort. So if you're in the business cohort, you can go watch the technical and creator.
Um. Trent, I've been searching for these perspectives for so long. This information is important.
I'm so glad you think so. It seems impossible to find normal people. I would argue I'm very far from normal.
I'm a little bit of an odd one, but I'm glad that it's refreshing and a breath of fresh air for you. I think there's quite a... Nate B.
Jones is a really interesting one. He's a bit more into agents, but I think he has good reasons for using those agents. That's why I'm not against what he teaches, and he has a lot of information out there.
Honestly, though, if you really want... to learn a lot about AI, stop learning about AI and pick up like a software fundamental book or like a book from the 70s. Like, let me see.
I've got a couple right here. Hopefully the green screen doesn't stop it. Yeah.
So this is The Psychology of Computer Programming by Weinberg. I think this came out in like 72 or something like that. This is a really great book on why people do programming.
Another good one, I recommended it on my LinkedIn is the mythical man month. And it's talking about like adding more people to a software project doesn't make it more efficient. So go read some older books, go read software fundamentals, because that'll tell you because AI is software.
It's a tool, it interacts with the computer. Just because an AI is building software working with it doesn't mean the fundamentals go away. So go check that stuff out.
Oh my gosh, there's so many messages to catch up on here. It's going to be the longest August of my life. That's awesome.
Could you share the payment plans again, please? Yeah, Matt should be sharing some of them again. Book a call and we can get you set up on all of those.
It might be with me. We're doing round robin on who gets the call. So I can't guarantee you I will be in the call.
It might be Matt, might be someone else that's with us, one of our amazing team members, but just want to put that out there. Jake, I don't know if you're still here. One last question.
How fast do we need to commit? To avoid it filling up. I mean, we have over 60 people already.
Or maybe even more. We might be already over 100. I'll have to ask Matt.
And that's just after two weeks. And that was faster than I was expecting. So it is filling up pretty quick.
But again, we're holding calls, right? Call me a bad salesman. I don't want to put FOMO into you.
Take your time. Really think about it. I mean, if you got the cash, you come in.
I promise I'll deliver for you. You can buy it right now. Please like take your time and learn.
I am not one of those guys that likes manipulating into this. Cause again, I know it's good. I don't need to sell you.
It's amazing. It's going to be amazing. I'm going to have a good time, right?
It's all of those things. The books are mythical man month by Fred Frederick Brooks. And another one is the psychology.
of Computer Programming by Gerald Weinberg. If you go on my LinkedIn, if you look me on LinkedIn, I think I made a post recently with links to those books and two more. You should be able to find it there.
Let's see. I see Matt's answering a bunch of questions. So thank you for that.
Got your call booked. Oh, I'm so excited. Thank you.
Wow, we already still have 170 people in here. Thank you, everyone, for sticking around an hour and a half. And I haven't scared everyone away.
That makes me happy. I will say. I think I should end it now.
It's been an hour and a half. It's quite a while. I haven't eaten dinner and it's 830 and I'm a little hungry.
Also, my family probably would love to see me because I've been working all day. So I'd like to say hi to them before they go to bed. Either way, thank you for everyone who has come out.
Thank you for the people from so many people from all around the world, all around the states. I appreciate your time. I hope you learned something.
I hope you're interested in joining the Lyceum. But if not, you know, thank you for calling me out, Timber. If you're still here, I'm glad that I was able to catch that one and show you I'm a real person.
Yeah, by all means, please have a wonderful night. Don't worry, Don. You know, we'll catch up.
We're here. We're doing it for it. And for those of you who missed out, again, we're doing another webinar next week and we'll be sharing a lot more information on this all over the place.
We'll be emailing everyone who's a part of this as well. bunch of catch -up things and slides and stuff like that. So we'll see you there.
I hope you got some value out of today. I hope you understand what's important. And until next time, enjoy yourself.
Have a wonderful day. And don't fall for the hype. Just learn, take your time, find the problems, and build.
And always remember, in a world full of cheap answers, questions become valuable. Thank you, everyone. Have a good day.
Think I ended it? Possibly.
The Hook
The bait, then the rug-pull.
Before any slide appears, Jake Van Clief spends several minutes just watching attendees pour in from Denver, Beirut, Rhode Island, and Oakland, greeting each city by name until the count passes 200. It's a long warm-up for a talk whose actual claim is blunt: the reason most AI projects fail isn't the model, it's that nobody organized the context — and the fix is older than the internet.
Frameworks
Named ideas worth stealing.
13:09model
ICM — Interpretable Context Methodology
Organize AI context as a folder hierarchy of markdown files (with an entry-point CLAUDE.md-style file) instead of building agents, RAG pipelines, or fine-tuning. Any capable model reads and navigates the folder, effectively 'becoming' the agent a given question requires.
Steal forAny AI-assisted workflow where the bottleneck is giving the model durable, reusable context about your business, opinions, or process — not building custom agent infrastructure.
34:53model
The Four Layers (library analogy)
Front desk (the UI/chat)
Catalog (the folder/index describing what's where)
Request slip (scripts/instructions for how to fetch more data)
Memory & state (ground truth — databases, linked files)
Maps AI system components to the parts of a physical library: a front desk you talk to, a catalog describing what exists and where, a request slip for pulling in more data, and a memory/state layer that's the verified ground truth.
Steal forExplaining AI system architecture to non-technical stakeholders without jargon.
21:09list
Six failure patterns
Pilot purgatory
Context sprawl
Token burn
The handoff problem
Governance fear
The quarterly rebuild cycle
The six recurring reasons agentic AI initiatives stall or die inside companies Jake consults for.
Steal forA diagnostic checklist before greenlighting a new internal AI project.
53:11concept
Job to be done
Before automating a workflow, ask what underlying outcome the person actually needs — illustrated with 'don't build a better drill, ask why they need a hole in the wall.'
Steal forScoping any automation or AI project request before writing code.
29:10concept
1970s engineering lineage
Unix (1978) — each program does one thing well, plain text as universal interface
make (1979) — separates coordination from execution
Parnas (1972) — information hiding, decompose by what each part hides
Jake's framing that ICM doesn't invent anything new — it applies fifty-year-old software engineering principles to LLM context management.
Steal forGrounding a 'new' AI pattern in established engineering principles when pitching it to skeptical technical stakeholders.
CTA Breakdown
How they asked for the click.
VERBAL ASK
1:11:29product
“The number is 4750... That's what we're charging for all 12 weeks plus all the builds plus everything inside of it.”
Delayed and telegraphed pitch — flagged at 13 minutes in ('towards the end I'm going to sell you something'), reiterated once more before the reveal, then delivered as a plain price statement without urgency tactics; explicitly frames the free 20-minute placement call as the next step rather than an immediate buy button.
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.
A mastermind member's question about landing clients turns into Jake Van Clief's longest, most honest answer on how to go from zero to paid, and why buyers really open their wallets.
A creator feeds a public archive of 2,000+ Alex Hormozi coaching transcripts into a Claude Code skill that interviews any business and spits out an AI-written constraint diagnosis — then uses the free download as the top of a paid-consulting funnel.
A solo creator-coach draws her identity-change framework straight onto the glass behind her — helmets, a thermostat, a devotion-vs-discipline split — arguing results never outrun self-image.
A live coaching session where a business advisor rebuilds a client's AI service pricing at a whiteboard, using backward math and a three-tier offer ladder.
A business coach who built 400,000 followers solo lays out the three renewable habits — mining comments, scheduled unplugged thinking, and refusing to hide her real story — that keep her posting four times a day without ever running dry.
01:42
02:51
03:40
05:24
06:17
08:34
09:43
10:54
11:33
13:09
14:17
16:35
19:24
19:42
21:09
24:35
25:44
26:52
28:01
29:10
31:53
32:03
34:09
34:53
36:12
37:42
39:12
40:45
41:23
42:20
44:13
44:36
47:16
48:37
49:45
50:54
52:03
53:11
55:28
56:22
57:22
58:54
60:03
61:12
63:42
64:38
65:46
66:55
68:03
69:12
71:29
73:47
74:55
76:04
77:13
79:30
80:38
81:47
82:56
84:04
85:13
86:22
87:30
88:39
89:48
90:56
16:25
39:00
13:04
13:45
09:35
16:04