17:12Samin Yasar · Tutorial
Clicky Just Changed How We Use Computers Forever! (Tutorial)
A 17-minute hands-on tour of Clicky, the Mac menu-bar agent that clicks around your computer for you while you stay the director.
May 11thThe argument in one line.
Claude Fable's looping, self-testing endurance turns every multi-step job into a responsibility you hand off instead of a task you babysit.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude and want to move beyond one-shot prompts into multi-hour autonomous runs.
- You run a service business or solo agency and want to automate research-heavy client outreach at scale.
- You spend money on SaaS subscriptions, marketing asset tools, or video editors that AI agents can now partially replace.
- You have investing hunches but lack the data-science skills to back-test them against real price history.
SKIP IF…
- You want prompt engineering fundamentals — this assumes you already know the basics.
- You want an objective model comparison — this is enthusiast demo content.
The full version, fast.
The core argument is managerial: stop giving Claude tasks and start giving it responsibilities with a definition of done. The video demonstrates eight applications including building full apps with /goal, connecting bank accounts to auto-cancel subscriptions, running a Higgsfield marketing agent that QA-reviews its own output, spinning up four parallel sub-agents that debate a Tesla trading hypothesis, generating interactive HTML visualizations, editing a video intro using the Hyperframes MCP, and using Clay to find and pitch 10 personalized leads. In every case the workflow is: set a goal with a verifiable finish condition, then walk away.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 01:00
01 · Hook and thesis
Model introduced; core thesis stated — move from tasks to responsibilities.
01:00 – 03:02
02 · How Anthropic uses Fable 5
Embedded Anthropic engineer clip: treat Claude as thought partner, give goals plus verification methods.
03:02 – 05:38
03 · Use Case 01 — One-shot apps with /goal
Full Minecraft game built from a single /goal prompt; demonstrates looping until feature-complete.
05:38 – 09:02
04 · Use Case 02 — Claude as Bookkeeper
EraContext connector links bank accounts; Claude generates pie chart and subscription table from one prompt.
09:02 – 12:32
05 · Use Case 03 — Subscription Killer
Claude for Chrome connector navigates LeadShark billing page and cancels subscription autonomously.
12:32 – 20:15
06 · Use Case 04 — Images and Video Marketing Agent
Higgsfield MCP generates ad images and video ads; Fable auto-approves/rejects on visual QA.
20:15 – 28:06
07 · Use Case 05 — Backtest Your Stock Hunches
Four parallel sub-agents run Tesla price data, back-test, politician trades, and web scraping; agents debate; interactive HTML report produced.
28:06 – 28:12
08 · Use Case 06 — HTML Visualizations on Demand
Claude Code generates branded interactive graphics from natural language; used directly as production assets.
28:12 – 35:24
09 · Use Case 07 — Agentic Video Editor
Hyperframes MCP; Fable edits the video intro live — adds tweet overlays, use-case teaser clips, chapter title cards.
35:24 – 37:54
10 · Use Case 08 — Client Outreach on Autopilot
Clay MCP finds 10 YouTube creators; Fable spins a researcher per contact, drafts a personalized 3-sentence pitch, checker agent verifies accuracy.
Lines worth screenshotting.
- The shift from task-giver to responsibility-assigner is the entire unlock — once you write a goal with a pass condition, you stop supervising and start delegating.
- /goal in Claude Code loops indefinitely until every specified condition passes — the agent does not stop at 5 PM.
- Fable's vision capability acts as a built-in QA layer: it catches its own bad outputs and routes them to a rejected folder before you ever see them.
- Parallel sub-agents that debate each other's findings produce a stronger final answer than a single agent running sequentially.
- Connecting a bank account via EraContext turns Claude into a bookkeeper that can also navigate real browser UIs to cancel subscriptions.
- Clay already has a native MCP, meaning verified contact data flows directly into Claude without scraping.
- Hyperframes lets Claude Code edit video non-destructively via a compositing canvas — natural language replaces a timeline edit.
- The presenter runs Claude on a Mac Mini 24/7 specifically to host long-running agent jobs.
- A checker agent that verifies each drafted outreach message is the difference between a personalized pitch and a generic template.
- HTML artifacts generated by Claude can be used directly as production assets in a video.
Eight jobs you can hand to an agent and walk away from.
WHAT TO LEARN
The model's endurance is the feature — once you write a goal with a verifiable finish condition, supervision becomes optional.
- Describing what 'done' means is more important than describing the task itself; /goal loops until that condition passes, not until it gets tired.
- Bank connectors plus browser control collapse two separate jobs — auditing expenses and canceling accounts — into a single delegated responsibility.
- A visual QA agent that routes outputs to approved/rejected folders removes the need to review every generated asset yourself.
- Parallel sub-agents on different data sources, instructed to argue with each other's findings, produce more trustworthy research than a single sequential search.
- The architecture for outreach and for stock research is identical: gather data in parallel, debate the findings, surface one verified answer.
- Clay's MCP gives you verified contact data without scraping; the value of Fable here is the per-contact researcher agent, not the list itself.
- Natural language video editing via MCP is iterative, not destructive — you direct it back and forth the same way you would brief a human editor.
- HTML as a delivery format for analysis lets the model produce interactive, shareable artifacts instead of flat text summaries.
Terms worth knowing.
- /goal
- A Claude Code slash command that puts the model in a looping execution mode, keeping it working and retrying until the goal pass condition is met.
- EraContext
- A free connector that authenticates Claude to read-only bank data via Plaid, letting it analyze transactions without storing credentials.
- Claude for Chrome
- An MCP connector that gives Claude control of a Chrome browser session, enabling it to navigate real UIs and cancel subscriptions on the user's behalf.
- Higgsfield
- An AI video and image generation platform with a native MCP, allowing Claude to generate ad assets, review them visually, and route approved vs rejected outputs.
- Hyperframes
- A HeyGen product that exposes video editing via MCP; Claude can add overlays and splice clips using natural-language instructions.
- FireCrawl
- A web-scraping service with MCP integration that bypasses bot-blocking on data sources like political trade disclosures.
- Clay
- A sales intelligence platform aggregating verified contact data from 150+ providers; its native Claude MCP lets Fable pull prospect lists directly.
- Multi-agent debate
- An architecture where parallel sub-agents each run a different research task, then challenge each other's conclusions before a final synthesizing agent produces one answer.
Things they pointed at.
Lines you could clip.
00:40
“Stop giving it tasks like it's a tool and start giving it responsibilities like it's on your payroll.”
04:45
“The contractor used to just leave at 5PM regardless. With /goal, he does not leave until the job is done.”
26:03
“I don't want the first answer. I want the answer that survives all the arguments.”
33:10
“I'm editing this video with Fable.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Clot just released the public version of their most powerful model called Fable five. And in this video, I maxed out my $200 Clot plan to show you guys what this model was built to do, how to properly use it, and a few insane use cases that you should be using it for. And, yeah, the benchmarks are insane, but let me show you the practical applications of this new model.
00:19So how is this thing actually meant to be used? Well, Anthropic said it themselves. The longer and more complex the task, the bigger fables lead, which means this model is built for long running agentic work like big analysis jobs, vision tasks, and stuff that takes even hours or days to run without you having to babysit it, which means the way we've been using AI has to change where we have to stop giving it tasks like it's a tool and start giving it .
00:47And that's exactly what I'm about to show you, how to put Fable five to work. Oh, and it's free on paid plans until June 22. But with that out of the way, let me show you how the Anthropic team uses Fable five so you can pick up some really cool tips they shared.
01:04Hey, everyone. It's Tharq from the Cloud Code team here to talk to you about Cloud Fable five, our newest model in Cloud Code. It is an incredible model and has really pushed us at Anthropic to rethink how we work with and prompt Claude.
01:16We used to need to check if Claude is doing the work right, for example, by breaking task into small chunks for it, double checking its output, catching when it stopped early. With Cloud Fable five, I instead increasingly find myself checking if Cloud is doing the right work.
01:33So Fable can run for hours at a time, it tests its own work, and honestly, I often find it produces better code than me. My job is increasingly more about direction and setup, not supervision.
01:45So here are the three ways I've changed how I work with Fable. First, I treat Claude like a thought partner. I give it the context it needs.
01:54Second, I give Cloud goals and ways to verify them. And finally, I'm trying to be more ambitious and ask Cloud to do things I've never tried before. We have two new features that work well for this that I'd encourage you to try out, flash goal and workflows.
02:07Goals help Cloud keep working until it's done, and workflows help Cloud verify its work. So after I have a spec, I might tell Cloud, set a goal to implement the spec fully, then use a workflow to verify each part of the plan and prepare a report on what was implemented and if anything differed. This lets Claude use its capabilities as creatively and thoughtfully as possible while making sure it's building what you want.
02:31Finally, just be more ambitious. Fable is truly an incredible model, and it's pushed me to think outside of the box with my work. For example, I'm editing this video with Fable.
02:41If there's something you assumed LLMs couldn't do,
02:44give it a chance. We really think Fable raises the bar on what is possible. That was really insightful.
02:49But notice how he said he edited his video with Fable, for example. I'll show you how to do that in a little bit. But right now, let's dive in, and I'll show you how to do use case number one.
03:01Alright. So first things first. To make sure you have Fable and you're using it in the best way possible, number one, make sure you have the Claude desktop app downloaded.
03:10And number two, if you don't see Fable right here in this model selector, make sure to quit and then reinstall the application. Right?
03:16Okay. After that, you wanna go right here and then expand this, and you'll see Fable five for your toughest challenges included until June 22. Okay?
03:25So let's hit that and let's try it out. And that's all you need to do. Alright?
03:30And in co work, you'll see it as well. So right here, if you look at the model selector, you can see you can select Fable. So those are the two things you can use it for right now.
03:43Alright. So first things first, I wanna show you a simple slash goal command and the lens at which that fable will go to get your tasks done. Right?
03:51So I'm gonna enter a very simple prompt slash goal, create me a complete Minecraft game. There's a lot of stuff here, but let me but let me play this and then come back. This is gonna take a while.
04:02And notice how I used Fable Max. Anyways, I'll come back when this is done.
04:08So very simply, let me open this up and let me play with this.
04:14Right? This is pretty cool. It's actually insane how well it's done, uh, and the things I can do in here.
04:21Yeah. I just wanted to show you slash goal and how the extents at which you can get Fable to run.
04:27And, you know, that was a very simple use case to portray how good Fable is with slash goal. Right now, the fact that, you know, when we use slash goals, these agents don't quit until they're done.
04:38So it used to be when you're using other models, like the contractor would just leave at 5PM regardless. But with slash goal, he does not leave until the job is done. They basically keep looping and looping and looping, and that's one of the biggest things Fable was built to do.
04:54So think of any long running tasks that you may have, something like running overnight, like optimizing your code base or some other things that I will show you later in the video. But, you know, the implementation of it is really easy.
05:07You just go to Cloud Code, you describe the app and what working means, and you put your definition with slash goals and you walk away. It just takes a long time and you let it keep working in the background. That's why I put mine in my Mac mini and I keep my Clawd on all the time working as it needs to.
05:24By way, if you wanted the board that I'm using for this and all the use cases and the text guides, whatnot, uh, in the link below in my community, if you go to YouTube video resources and the first one, Cloud Fable Board, you should be able to access all the resources that I'm giving out in this video. Alright?
05:44So this use case actually just helped me save over $1,500 a month that I was spending because I realized my spending is getting out of hand. And how I do that is by getting clawed to cut all the unnecessary subscriptions that I don't need.
05:58And honestly, do you know what you're actually paying for every month? Even if you're spending on too many subscriptions like right now, sometimes you just forget or are too lazy to cancel things that you don't need.
06:09Like, look. I have so many random subscriptions that I'm paying for that I'm gonna show you how we can cancel. So the next use case I'm gonna be showing you is one, how we can get Claude to be our bookkeeper, and number two, how we can go out and get Claude to use our computer and automatically cancel subscriptions that we don't need.
06:27And also, you'll learn how to build these cool graphs, look at your expenses, talk about it, all those stuff. Alright? So let's dive in.
06:33Alright. So step one, to get Clot to be our bookkeeper, we wanna make sure we have we wanna make sure it has access to our banking information.
06:42And the way we do that is by going to this plus button and then going to connectors and then go to manage connectors. Okay?
06:49After you're in manage connectors, let's look for a new connector by hitting this plus button and then look for browse connections. Now you're gonna look for finance and look for any of your bank accounts that you may have.
07:06A lot of people use Mercury. A lot of people use Zoho Books. There are QuickBooks.
07:13A lot of these things. Right? And if you're and if your bank account, let's say, is not in here, like you wanna connect your Amex account or something like that, what you wanna use is this thing called EraContext.
07:23Alright? So if you go here, EraContext, and you can allow it access right here.
07:30Make sure you're logged in first, of course, and then you should be good to go. Okay.
07:35Now just to show you. So error context, what you wanna do is once you log in, and it's free, that's why I'm using it right now, you go to connect your accounts, and then you can connect any of your accounts.
07:48Right? Like, connected two of my accounts. And to connect an account, it's really easy.
07:51You just go to connect more. And then you, you know, you find your banking institution and then you just log in with that. It's played, so it's pretty secure.
08:00Uh, that's how everything is done. But after you have that connector enabled, then if you go back to Claude and you see error context right here, make sure you have given it all the approvals that you need to.
08:12Now what happens now what happened is basically, you have Claude and you have your bank, and now Claude can access your bank through this Era context thing.
08:24Alright? Okay. So once that's done, all you need to do is right here, go here, and then go to connectors, Make sure you have Era context enabled and just ask.
08:37Hey. Can you break down my finances for the last month? Okay?
08:44And then I did that's the exact same thing I did to get to here. And that's how I got this graph. Okay?
08:53And you can ask it for any graph or something like that. Like, if you look all all I asked is show me make me a table of what I'm paying for and make a pie chart. Really simple.
09:02Okay. Now that that's done and you have listed out all the different subscriptions you have, what you can do right now is you can ask Claude to go and cancel that.
09:13So let me show you how to do that.
09:17Alright. Step number two, go through your subscriptions and note down which ones you don't want.
09:22Right? So for example, just to make it fast, let's say I don't want this lead chart thing. Right?
09:27So what I'm gonna say is let's say that's the only subscription that I don't need. Next, what I wanna do is make sure and right here, I go to connectors, and I have Claude for Chrome enabled.
09:42Okay? So Claude for Chrome is letting you letting Claude use your Chrome browser to actually get stuff done.
09:51And how that works is because Fable is so good at computer vision and looking at things and watch this clip right here.
10:00Claude Claude Fable was actually able to complete the entire Pokemon game just by looking at what it needs to do, doing all that thought process and actually getting it and going end to end in the game.
10:15That is incredible. Right? So what I'm trying to say is that relates to this task because you have so many different subscriptions.
10:23And in the same way that all these things look different, your browser Claude is going to use your browser with Fable to go through all the different UIs and then click and find and cancel those subscriptions. Right?
10:36So after you have Claude for Chrome enabled, so all I did was in Cowork, go to Chrome and go all I said is open Chrome, go to LeadShark, and cancel it. And you can see this is really cool.
10:50Uh, I can even see a little preview of Clot in Chrome. I can see what's going on, what it's capturing, and Fable is just doing its thing. Alright.
11:00So notice this. It's saying stuff like you're logged in. I can see your dashboard, pro plus plan.
11:06It's doing a bunch of stuff. And the nice part is that, you know, you can put in a list of different things that you want canceled, and it'll just go through and keep doing it for you rather than you having to do it.
11:19And that's what I mean by long running task and giving it responsibilities. This thing is just running on my Mac Mini, and this is gonna keep doing this for, you know, a long time and then cancel a bunch of subscriptions that I don't need while I can go be doing work somewhere else.
11:35And you can see, uh, this is actually kinda cool. It's showing, uh, the cancel subscription button, and it's going through and then doing it for me. Okay.
11:43So I I just verified, uh, just for you guys, but I just saw it's done. You can see this little orange, um, background just means and this cursor means Claude was working with it.
11:55But, essentially, I just said, tell me the steps and our outcome. They said, hey. It's canceled, meaning the subscription won't renew.
12:02And, you know, that's really cool. You can do this for multiple of your subscriptions.
12:08Just make sure that you have know, you are logged in. And in your Chrome, if you're already logged in, you can just go one by one by one and keep hammering out and canceling the subscriptions really easily.
12:21I love this use case because, you know, it tangibly saves me money all the time. But this is just getting started. Let me show you another use case where we're gonna take Fable to the next level.
12:34Alright. So so the next use case is the one I'm most excited about, and it's using Fable as my marketing agent. And this one is so cool because marketing is like 10 jobs at once.
12:44It's strategy, copy, directing, production, quality control, so much stuff. But the thing is, AI video, it's literally like a slot machine, honestly, where I prompt and then I have to pray and then I have to try again over and over.
12:59The nice part about Fable and why I'm so excited about Fable is the number one reason is that it has eyes. Like, it has really good taste.
13:10So it looks at every image, clip, and catches the bad ones and keeps learning. So that's already a lot of quality control issues that I don't have to do.
13:18And number two is the prompting. So Fable can actually look up and really write very in-depth prompts for these AI videos for any model without me having to know or develop any skills.
13:33So when, let's say, a prompt failed and I don't like it, it will remember that and rewrite itself. And number three, the goal loop. Like we saw with the Minecraft, it's gonna keep going until it passes on our own review, so it's in built quality assurance.
13:49Okay. So with that theory out of the way, let's put this in practice, and I'll show you by making some ads for my own community. And doing that is honestly so damn easy.
13:59So step one, what you wanna do is you wanna I like using this kid thing called Hicksfield. So if you go to hicksfield.ai/mcp, again, the link will be in the description below.
14:10And all you need to do is grab this thing right here, go back right here. You wanna go into connectors, manage connectors.
14:20Okay. What you wanna do is you wanna come here and then add custom connector and then paste this in and call it Higgs field.
14:30Add. And then right now, you can see it's in custom. Okay.
14:33Then what you wanna do is make sure you have everything on always allow, and you refresh tools list, and you should be good to go.
14:42Awesome. So now we should have Hicksfield.
14:48Right here, if I scroll down, I can I can see Hicksfield? Great. So now that I have Hicksfield, now let's actually build a marketing campaign for my community.
14:58And if you guys didn't know, I run this community called school.com/clotclub. And right here, you can see I have this about page, and you can see this light mode, anthropic theme, all the results, and all the stuff.
15:13And I want to give my AI as little direction as possible and let Fable take the wheels, and let's see how it does building a marketing campaign with very little direction. Right? Okay.
15:24So so set a goal to make a full set of ads, images, and videos. What I want you to do is use a workflow to find and scrape things from school.com/claud, and then create all the different image and video ads using the Hicksfield MCP.
15:43All of them, you should do in parallel. The ads should should be about this video I'm making right now, so go look up my script.
15:51And then also make a sixty second video. It should be a motion graphic video like the ads you generate in that paper style that I have. So look through my script and extract all the things and use cases that I'll be showing and use those to make ads and content based on all the use cases I'm showing.
16:11Alright? And then neatly organize it and put it in a folder in a finder so I can look through it and open it in my finder.
16:20Actually, I just noticed something. I said, notice open it in my finder. So I can't use chat.
16:26I need to use code. So I'm gonna go here, hit new session. I just copied that thing, and I'm gonna make a new folder.
16:33I'll call it Claude Club Fable Ads and open that.
16:39And then right here, I'll go ahead and hit enter, and then I'll come back when this is done. Awesome. So it's telling me it's done.
16:47It said it generated a full set of video ads, and it put it inside this fables folder. And anything with a misspelling oh, that's really cool. Again, it has eyes and does that quality reassurance.
16:59It put it in rejected, and the clean ones, it's on approved. Cool. Now it's saying just let me know and browse through and what they want.
17:07Cool. Let's go look at it. Um, just for reference, right here, let's look at the video ads first.
17:13There's approved and rejected. Let's see what it rejected.
17:18Why am I in glasses?
17:20I gave Claude years of stock data. It back tested everything and found patterns and ranges I'd never seen.
17:29Uh, school dot yeah. Okay. Yeah.
17:31I understand. This is really cool that it understood all these little mistakes that it made and automatically rejected them. I'm assuming it's the same thing.
17:40Uh, you see, yeah, all these are pretty bad. Okay. Let me just look at a approved one.
17:45I gave Claude every bank statement I had. It found $214
17:49a month in subscriptions. I forgot existing.
17:55Okay. This one's not bad. It's not the war not the best, but not the worst either.
17:59This one's fine too, I guess. Looks good. Looks like me.
18:02Let me look at these. Those look kinda cool. Oh, these are the product videos.
18:05Okay. Always on right when you need it. One brain across your devices.
18:10Clips cut themselves. Clean takes, fast.
18:14Everything you use, linked. Every workflow, connected. Join claudeclubschool.com.
18:21Okay. This is pretty cool. I like that.
18:23That's actually really cool. Now let's check out the images. Image ads.
18:29Fantastic. Let me open all of these.
18:34Wow. This is really cool. It went ahead and scraped my photo.
18:37I didn't give it that. Uh, there's my link. Oh, this is I love this.
18:43Uh, this is something that's very usable. And you can notice if you've been watching this video, this is something that it literally took my script. It found out what I should be showing and even visualize it and show this is so cool.
18:56These image ads are fantastic. Drop in your bank statements, found these things, cancel them on on autopilot.
19:04Yeah. We just did this all of this stuff right here. The Mac Mini, this is coming later.
19:09The backtest also coming later. So sneak peek for you guys. Uh, all these assets, the slide decks.
19:16Okay. Editing your videos. Okay.
19:19This is very nice. Wow. It did a little drawing of me as well.
19:23Awesome. So you can see just with Higgs field plus Fable, the output you get is so much better and it automatically can approve and reject and do all of this work on your behalf.
19:38It just becomes another reason why everything is so much easier, and now I can just put this stuff on autopilot and keep generating and seeing my ads actually happen on autopilot.
19:52Okay. I really like this. I'm going to come okay.
19:58So I really like this. Now I highly encourage you to go try this out, but now let's actually push this to the next level and use some of the things we've learned and go to the next use case.
20:15So the next use case is getting Fable and its crazy capabilities to help us pick winning strategies when it comes to trading stocks. Let me give you an example. So say you've got an hypothesis and you've been watching Tesla and you're thinking it bottoms out at 360 each time and then bounces up.
20:34Right? So every time it dips to 360, basically, it goes up.
20:39I want to check if I'm right about that. And normally, you kinda just have to vibe it out. So because Fable can do deep research, here's what happens.
20:47It can spin up a bunch of different sub agents that are responsible and do all of this stuff parallelly at the same time. So so for example, one can literally go through all the historical price data. One back tests your hypotheses with all the past data of what Tesla has actually done.
21:08Another one, let's say, looks into politicians and senators and what they're buying and why. And another, let's say, goes through other sources, scrapes that data, and then uses that.
21:20What that means is then the agents get to debate each other with all their different findings and which one's the best, and it can tell you which theory would actually hold.
21:32Right? And it's super effective. So the bottom line is that it helps me do all my research a lot more effectively about which trades I want to make.
21:41And here's exactly why Fable is so good at this task. So it literally tells you the truth even when you're wrong. Basically, we can say we would have made 90% and still lost.
21:53Whereas if we just held the Tesla rather than buying the dips, we would have just gotten more. And the reason Fable is so good at this task is because of this. First, it can run tasks for very long, which means it doesn't just stop after one search like you saw in the Minecraft example.
22:11It gets the data itself. And you've seen how in the Minecraft example, it went out, find all those libraries, and then built its own app.
22:19It can run and write its own code using the same capabilities. Like, when we gave it all my financial data forever, and then it found all the things that I've been spending on. These are all the different capabilities that we can combine together to help us essentially do better stock research.
22:37Okay? And essentially, all of these can debate itself and then lead to a better outcome.
22:43Great. Now let me show you how to implement this in practice. Now to set this up the correct way, we wanna make sure whenever our Claude goes to scrape information, it's not blocked.
22:53Right? So for that, we're gonna be using this thing called fire crawl. And the reason we use that so for example, this is one source of data where we can get politicians' data and whatnot.
23:04But the problem is bots often get blocked when we're trying to scrape this data. So what we're gonna do is come here, go to FireCrawl and sign up, and, uh, you can just use the free account.
23:16That's totally fine. Then what you wanna do is you'll see this API key right here. We're gonna need this in a bit.
23:20Alright? So we're gonna go back to Claude, and we're gonna go plus connectors and then manage connectors.
23:29The plus right here and add custom connectors, and then let's add FirePearl. Then we're going to paste in our MCP URL.
23:37Okay? Now what we're gonna do is after dev, we wanna do slash and then paste in our API key.
23:43Then you're gonna make sure you hit always approve. Great.
23:47Now we're gonna have the fire crawl MCP up and running. Great.
23:52Now I'll show you the prompt we can use to make sure we can get our fable to do all that good research and how to test our hypotheses. Alright? Great.
24:02So after we have FireCrawl MCP connected, then just make sure you have it right here.
24:08And then after that, I'm going to put this on high. And then I have a prompt that I use to make sure I can back test properly. And in the description below, you'll see in my community, uh, inside that same board, I pasted this prompt, and I'm gonna take this and paste it in here with the same hunch.
24:28Alright? So let me just read it to you just so you understand. Basically, I'm saying, I have a hunch, and I want you to either prove it or destroy it.
24:36My hunch is that Tesla bottoms out at 360. Every time it dips near that level, it bounces back. So I'm thinking of buying it.
24:43So this is basically what your hunch is and what you might want to do. Okay? So this is your strategy.
24:49Okay. And then here's what I want you to do. Run a deep research on this and don't stop until the job is done.
24:55So and then I give it a couple things that it needs to do. So first, pull the last two years of real Tesla price data and then back test all that entire hypothesis, like, every time I did that. And then check capital trades.
25:08This is so I can back test it with politicians and what people or the big players are doing. Then I wanna also see what other analysts and hedge funds are doing.
25:18So there's two different datasets, and then they're gonna be arguing against that. Have the agents challenge each other's findings before you give me anything else. I don't want the first answer.
25:26I want the answer that survives all the arguments. Then put together a polished reports document that I can download so my hypothesis is up top. And there you go.
25:35That's about it. So I'm gonna go back to Claude, and then I'm gonna hit go.
25:43And then after this is done, I'll show you what this looks like. But this is how I test if I have an idea to buy something. There are different ways you can, uh, go towards and do.
25:53But if you need help with this, I'll be inside the Cloth Club. Happy to help you guys do anything like this if you guys want to. So if you have a look, it just made this, uh, artifact, and you'll find this in the description below in the community again.
26:04But you can see it's my hypothesis, and it clearly tells me this is a bad hunch.
26:12And with that bad hunch, you can see it also generated a lot of interactive graphics that I can be playing with, seeing what's going on, looking at all these different things, and, you know, make my judgments.
26:25It showed me the bull case versus the bear case, all the different trades, a bunch of stuff.
26:33So, basically, it makes this entire assessment and makes me make my decision really clearly, really fast with an adequate amount of research and data.
26:46Great. So and you can notice and if you've noticed, it's very simple to build out these kind of assets with CloudFable because it's so so good at coding in HTML.
27:00And if you've been watching this video, I've been it's actually how I've made all these assets. So for example, you can see these are some examples of other graphics that I've made that you maybe have been seeing across, and it's as simple as just telling Claude Fable to create me a visualization with HTML.
27:23That's another use case as well. So it's fantastic the way it can visualize it because it has that visual understanding. It can do that looping process, and it makes fantastic graphics.
27:34You can see these cards, and I can I'm using this to edit my video and actually plays these throughout. So I highly recommend you play with that because it's probably one of my favorite use cases, although it's so simple to do.
27:49It's just asking Cloud Fable to create these kind of colorful visualizations using HTML, and it builds these beautiful graphics for everything we're gonna be covering in this video.
28:06So now that you know how to make graphics, let me show you one of the most powerful features of Fable. And what it is is basically an agentic video editor. You can literally get Fable to edit your videos because it's so good at long running tasks, and it can input and understand so much.
28:22Right? And doing it is really easy. All you have to do is go to Claude.
28:27Let's start a new chat. Go right here, and let's add a new connector.
28:34Let's add connector. And this is called hyperframes. It's free, and it's by Heijin.
28:39So let's hit connect there. Okay. So now we have access to hyperframes.
28:44Okay. Now, ideally, when you do any video editing, what you want to do is use Claude code.
28:51Right? So in Claude code, you can see it right now.
28:55Let's try FlickFable. And this is for any long running tasks.
29:00So I'm gonna select Fable. And what I'm gonna do right now is I'm going to select and what I'm gonna do is because this video you're watching right now, I'm gonna show you how I'm editing parts of it live.
29:13Right? So for example, this is like the initial part of my video where I just have, uh, an empty screen, and let's say I want some captions and graphics and stuff there.
29:23And there's so many things happening inside the video that I want previewed right here. Okay?
29:29So I want Fable to do that. Okay.
29:32So let's get that done. When I hit plus in connectors, I see there's hyper frames by agent.
29:38Right? So just to make sure, hey. Do you have access to edit videos with hyperframes?
29:46And one more thing I forgot to do is right here, go to my Fable videos and then open that folder up. Alright. So it has access to all of the stuff, and now it's gonna be really simple.
29:56So hey. So I want you to edit the first, I think, around two minutes of the video.
30:05That is the intro where I have that white screen. What I want you to do is place in graphics, go find tweets, onboard them, put them as overlays in the intro so that it looks much nicer.
30:21And things that I'm talking about that's happening throughout the video, I want scenes to be inside the intro so it's teased and people can see the different use cases I've been covering inside the video.
30:36Alright? So can you go ahead and then edit that video?
30:42And I'm gonna keep this simple for now, but right now, just natural language, I told it what I want. And just so you understand, what it is is that there is a blank part in my video.
30:53What I want is everything I'm doing. I want all the use cases over there. Okay.
30:58Right here. And it's just allowing me access, asking for all of the stuff. Oh, one more thing.
31:04Hey. Uh, also, can you make sure, um, as I do the different use cases, I want you to put in chapter titles. So use case one, use case two, and then name what the use case is with a small sound effect, and make sure there's a little chapter pause between each use cases.
31:25Alright? And this is how, you know, you kinda work with this agentic video editor. And with this process, you just keep asking back and forth what it does, and let's see what it can do.
31:37Here's something fun. Uh, while it's running, it actually just went out and found some tweets that are, uh, you know, I've already been talking about this stuff.
31:45So it's to find those overlays and put those on. Really cool.
31:50Okay. So it's that says it's done. So let's have a look.
31:55Claw just released the public version of their most powerful model called Fable five. And in this video, I maxed up $200 flatline to show you guys what this model was supposed to do, how to probably use it, and That's really cool. So do you see it added a bunch of stuff on the white, uh, so it added, like, all the little clips that I had.
32:10This is, like, from later in the video. It parsed them out. Of course, I'm gonna go back and forth a bit more to make it better.
32:19But, you know, this is a good start for agentic video, um, editing. And, you know, it gives me a lot easier time that now I can go ahead and release this video. And people can actually see something rather than just see my face talking.
32:36So the next use case. So everything I've shown you so far, these are skills. You can edit videos.
32:41You can build visualizations. We made detailed reports. But the real question is how do we make money with this?
32:48To and to make money, we need an offer. So let's say we can help coaches make better videos. We'll do their visualizations, and we'll help them do some editing.
32:56The exact stuff we just did. Right? So how do we find those creators?
33:01And this is really where everyone gets stuck because to sell a service, you have to find the right people. Then you have to research every single one, like what they make, what they're missing, where you can help them, And then we wanna make sure we can prove it to them and send them a short direct message where how we can actually help them.
33:20Right? Not just some generic template. And then you have to do this hundreds and hundreds and hundreds of times, and it's really hard to do manually.
33:28So let me show you how we're gonna do that. And it's with the service called Clay, and the biggest and the biggest sales teams in the world are using this thing.
33:37When what Clay does is basically it finds people and gets their active verified emails, their socials, their company information, and all the things we need to make sure we can send them a good message. And, apparently, it's pulling from, like, a 150 data providers. And the reason I'm showing you this is be if I go to connectors I don't even have it connected.
33:57But yeah. So it's because there Claude literally built an MCP into clay, and it makes things so so easy.
34:06And here, if I can just, like and right here, let's say I connect my workspace and hit authorize. All you need is an account. You're connected to Clay in two steps.
34:15Right? And that's why this pairing is so good because Clay can bring all the data that's already verified, and I don't have to scrape anything or check if the email is real. And Fable will help me with the endurance because Clay can now hand us, like, let's say, a 100 perfect contacts, but someone still has to research them and write each message.
34:35Right? And that's a hundreds of tasks back and forth, and that's what exactly Fable would be good at. Where old models would probably just quit or hallucinate after 40, Fable should be able to do all of them because the longer jobs, the better it is.
34:50That's literally what it's built for. Right? So let's put this in action.
34:53Let me show you how it's done. So we just connected Claude and Clay. Now let me put in a prompt to make sure we can get this done.
35:01So first, again, I'm going to look for if I have Clay enabled. Clay. Clay.
35:06Clay. Clay. Okay.
35:07I don't. So I have to hit connect right here, then choose a workspace. I'll choose a workspace and then authorize.
35:13Fantastic. Okay. Now I should be connected to Clay.
35:16And now let me check if right here okay. Clay is good to go.
35:21Now what I'm gonna do is let's say we are selling that service. Right? So can you use Clay to find, let's say, 10 YouTube creators in my niche?
35:31Let's say there are around 10 to 200 k subs, and for each spin up a researcher agent, go through their recent videos, find their content, and get their verified email. Then draft a three sentence message, reference one of their top videos, name the gap, offer one specific fix.
35:50I'll make sure all are researched and drafted reflects their real work. Okay?
35:55The checker agent should verify that. And, yeah, that's it. So now what should happen, we just learned now what's gonna happen is we just learned a lot of stuff that Fable can do.
36:06So with the stocks, the way we did deep research, we're using that ability here. And then we're also, like, in the Minecraft that it and the video editing that it ran for hours, it's gonna use those two ability here to make sure it does all the research, spins up multiple agents, and finds me and makes me a good lead list so I can start sending these messages and doing outreach on autopilot.
36:29And honestly, this makes outreach so much easier that you wanna make sure you have really good skills to back up whatever you're claiming. Alright? I'll show you how to do that in a little bit as well.
36:38But after this is done, I'll come back. This is pretty cool. Look at this little animation where it's, like, looking for all these creators, uh, right here.
36:47You can see it's doing so much more research than if I had to just prompt it regularly. But, yeah, I'll see how it goes.
36:55And I can show you this. I'm gonna hide some stuff just because I don't know how to reveal people's personal information, but this is my buddy, Mark. It found him.
37:02I was looking at it, and it got everything right about what he does. And you can see inside Claude, I'm interacting with it. I know he has a school community.
37:09He has this, uh, all the stuff. And it's really cool. It it knows exactly everything about him.
37:16Now it even drafted up the messages, and it makes it really easy for me to research people and find good contact to actually help with my outreach.
37:26Okay. So that's just one use case of play.
37:30But what I wanna make sure you know, but make sure whenever it draft the messages, review anything before you send it out. But if you want to learn more information about how to start a one person AI company, take all this knowledge, package it up, and sell it, make sure to watch this video next where I'll show you how to do that end to end with Claude.
The Hook
The bait, then the rug-pull.
When Anthropic shipped its most capable model, one creator maxed out his 200-dollar plan to find out what it was actually built for — and the answer was less about chat and more about running a small AI company on your behalf.
Named ideas worth stealing.
03:40model
/goal loop
- Describe the app
- Define what working means
- Run /goal
- Walk away
Claude Code loops indefinitely, running its own tests after each iteration, until all pass conditions are satisfied.
Steal forany multi-file coding project, content batch, or process with a verifiable finish condition
21:00model
Multi-agent debate architecture
- Spin parallel agents on different data sources
- Each agent develops its own finding
- Agents challenge each other's conclusions
- Synthesizing agent surfaces the answer that survives all arguments
Used for the Tesla stock back-test; four sub-agents running concurrently produce a stronger self-corrected output.
Steal forany research task where one data source is insufficient or bias is a risk
16:25model
Approve/Reject QA pipeline
- Generate all assets in parallel
- Fable visually inspects each output
- Route to approved/ or rejected/ folder
- Human reviews only the approved set
Fable's vision capability acts as built-in quality control; misspellings and off-brand frames are caught automatically.
Steal forany bulk asset generation workflow — ads, thumbnails, slide decks
How they asked for the click.
VERBAL ASK
37:14next-video
“If you want to learn more about how to start a one person AI company, take all this knowledge, package it up, and sell it, make sure to watch this video next.”
Points to a thumbnail on-screen while verbalizing the value prop.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
AFFILIATECommission earned if you click.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
