The argument in one line.
Splitting data analysis from design generation — using CoWork to build the data layer and design brief, then Design to render it — is what makes AI-produced dashboards genuinely agency-grade rather than generic one-shot outputs.
Read if. Skip if.
READ IF YOU ARE…
- You work with financial or ecommerce data and need to present findings to clients, boards, or teams without a dedicated BI team.
- You are a consultant, agency owner, or freelancer who builds client-facing reports and wants to cut production time dramatically.
- You are already using Claude and want to understand how CoWork and Design work together in a real client workflow.
- You want a clear, honest answer on when Fable 5 is worth the 2x cost premium over Opus 4.8.
SKIP IF…
- You need dashboards connected to live databases or real-time APIs — this workflow produces static HTML exports.
- You already have a mature BI stack (Tableau, Looker, Power BI) and have no need for a faster AI prototyping path.
The full version, fast.
The workflow is a deliberate two-stage handoff: Claude CoWork ingests raw CSV files and a brand style guide, structures the data into clean JSON files, and writes a design brief as a README; Claude Design then reads that brief and generates three fully interactive HTML dashboard prototypes in a single pass. The host demonstrates all three prototypes — Bold Data Studio, Dark Command Center, Summit Altitude — plus a standalone board report, all produced without any manual design work. The key insight is that CoWork handles the analytical and structural thinking so Design only has to render, eliminating the prompt iteration that usually degrades output quality.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:54
01 · Hook and Fable 5 benchmarks
Three dashboards in 15 minutes introduced. Fable 5 vs Opus 4.8: 80% vs 69% agentic coding, 78% vs 40% cybersecurity, 38% vs 14% spatial reasoning.
00:54 – 01:44
02 · Workflow overview
15 min manual work plus 15-20 min wait. Three dashboard prototypes plus a full board report. Dataset: ecommerce sales, cashflow, and expenses CSVs plus brand style guide.
01:44 – 03:07
03 · Claude CoWork prompt
Exact prompt shown: data exploration, analyze, and data-validation skill invocations, Big Four analyst persona, output structured for Claude Design, board accuracy requirement, ask-if-questions close.
03:07 – 04:23
04 · CoWork execution and output folders
CoWork produces analysis folder (cash flow, channel performance, discount analysis JSON files) and design folder (brand design tokens, README for Claude Design).
04:23 – 05:46
05 · Prototype selection and Design upload
Three prototypes proposed: Earthy Editorial, Dark Command Center, Bold Data Studio. Editorial rejected and swapped for Summit Altitude. Files uploaded to Claude Design with Fable 5 active.
05:46 – 07:58
06 · Dashboard showcase
All three prototypes demonstrated: Bold Data Studio (Profitable on paper, bleeding cash), Dark Command Center (neon/dark aesthetic), Summit Altitude (clean topographic theme). Each fully interactive with month-range filters.
07:58 – 09:12
07 · Board report
Separate CoWork prompt generates a board report. Same handoff pattern: CoWork writes prompt, Design renders standalone HTML report with executive summary, findings, and action steps.
09:12 – 10:29
08 · Cost context and Design edit tools
Fable 5 is 2x cost of Opus 4.8. Recommendation: reserve for high-stakes outputs. Design edit tools: chat tweaks, markup click-to-edit, font and color panels.
10:29 – 11:07
09 · Export and handoff to Claude Code
Export options include share link, PDF, PowerPoint, standalone HTML, project archive. Send to Claude Code hands off full project context to terminal for backend work.
11:07 – 12:06
10 · CTA and community pitch
AI Impact School community on Skool, Infinite Brain OS system, price increase June 15. Subscribe and comment CTA.
Lines worth screenshotting.
- The key to clean AI-generated dashboards is not prompting Design better — it is having CoWork write the design brief as a file so Design only needs to execute.
- Asking Claude to produce a prompt for the next tool in the chain eliminates manual re-prompting and preserves analytical context across tool boundaries.
- Three genuinely different prototype themes produce more useful options than three variations of the same aesthetic.
- The ask-if-questions close on a CoWork prompt surfaces missing context before a long-running task begins rather than after it fails.
- Fable 5 costs 2x Opus 4.8 — the practical rule is to reserve it for client deliverables and board presentations, not routine iteration.
- Attaching a brand style guide to the analysis pass rather than the design pass lets CoWork extract tokens and embed them in the brief so Design inherits them automatically.
- Claude Design's Send to Claude Code export hands off the full project context to a terminal agent, making prototype-to-production a one-click operation.
- A fictional brand dataset is a legitimate way to demo client workflows publicly without exposing real client data — the workflow is identical.
- Interactive month-range filters in a static HTML export give stakeholders exploration capability without any backend infrastructure.
- Financial narrative headlines like Profitable on paper, bleeding cash and Plug the leak. Then build. came from the AI — the host did not write them.
The two-stage handoff that makes AI dashboards work.
WHAT TO LEARN
Splitting analysis from rendering — letting one agent build the data layer and brief, then passing that brief as a file to a design agent — is what separates clean AI output from iterative prompt mess.
- Let the analysis agent write the design brief as a file; the design tool reads and executes it without manual re-prompting, preserving the full analytical context across tool boundaries.
- Attaching a brand style guide to the analysis pass rather than the design pass means the AI extracts the tokens and embeds them in the brief automatically — Design inherits the brand without a separate instruction.
- Requesting three prototypes with genuinely different themes forces visual divergence; editorial, dark command center, and topographic produced options that were actually different, not just recolors.
- The ask-if-questions close on a long-running prompt is a deliberate technique to surface missing context before the agent begins, not a politeness habit.
- Reserve Fable 5 for high-stakes deliverables — client presentations, board reports — and use a lighter model for iteration and routine tasks to stay cost-efficient.
- Claude Design's Send to Claude Code export is the bridge from prototype to production: it hands off the full project context so a Code agent can continue backend work without re-explaining the design.
Terms worth knowing.
- Claude CoWork
- A Claude interface mode that can read and write directly to local folders, enabling agentic workflows that produce and consume files rather than just chat responses.
- Claude Design
- A Claude interface mode for generating visual artifacts — dashboards, web pages, presentations — with live preview and iterative editing via chat or markup tools.
- Fable 5 (Claude Fable 5)
- A Mythos-class Claude model one tier above the Opus series, with substantially higher scores on agentic coding, frontier code, spatial reasoning, and cybersecurity benchmarks.
- Brand design tokens
- A structured file (usually JSON) that encodes a brand's color palette, typography, spacing, and other visual rules in a format that design tools and AI models can consume programmatically.
- Agentic coding benchmark
- A standardized test measuring how well a model can complete multi-step software tasks autonomously, including tool use, file manipulation, and error recovery.
Things they pointed at.
Lines you could clip.
00:00
“This is one of the three dashboards that I just made in about fifteen minutes using Claude's brand new Fable five.”
05:31
“Profitable on paper, bleeding cash.”
06:34
“Plug the leak. Then build.”
09:13
“It's two x the cost of Opus 4.8. I would probably keep Fable five usage to the minimum, only for hard or very important tasks.”
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.
This is one of the three dashboards that I just made in about fifteen minutes using Claude's brand new Fable five. And today, I'm gonna show you exactly how you can do the same because analyzing loads of data and creating these beautiful advanced AI dashboards is one of the main ways that I personally use AI for some of my clients, and it just got a massive upgrade.
Because Claude just released Fable five, which is a brand new model one step above the Opus models. It's basically a Mythos model, which has been highly anticipated for weeks now, and you can see some of the benchmarks right here.
Compared to Opus 4.8, it has 80% in agentic coding compared to 69%, and with Frontier code, it has 29% compared to 13.
Spatial reasoning is 38% compared to 14%. Cybersecurity is 78% compared to 40%.
So this is by far the biggest upgrade to an AI model that we've seen in months, and it's a pretty big deal. And it's important to be up to date with these things.
Let me show you the absolute best way of analyzing data and creating beautiful AI dashboards right now with Fable five. So just to show you an example, this is one of the three dashboards that I just had made.
Basically, three different prototypes of the same data. So let me show you the exact workflow I used to generate this result.
It basically just took me fifteen minutes of manual work and maybe another fifteen, twenty minutes of waiting for Claude to execute all of this. And it made this beautiful dashboard and two other dashboards plus a full board report that I will show you very soon.
But I basically had this folder with a lot of ecommerce data over the past six months or from October to March. So we have this massive sales CSV file, cash flow CSV file, and expense CSV file, plus I also attached the brand style guide for the brand I was working with.
In this case, it's a made up brand because I don't wanna reveal any actual data from any brands, but you get the idea. And the way I always like to start with doing these sorts of workflows is in Claude CoWork.
I use Claude CoWork a lot because I think that's all in all, like, the best tool for most people because it can work directly in your folders and it also has a lot of other really cool, uh, features like the live artifacts, etcetera. So I basically just told it to work in that folder with the ecommerce data that I just showed you, and this is the exact prompt that I gave it.
I said in this folder, you'll find a lot of e com data from a client of mine along with our brand style guide. Please use the data exploration skill, analyze skill, data validation skill, and any other relevant skill to analyze this data in detail. Act like a professional senior data analyst with ten years of experience working for the big four consulting companies, figure out the best way to structure the data and findings.
I want all the key data easily readable and a full report with the key findings and improvement areas. I will then give this data to claw design to make it into a visually stunning dashboard. So make sure it's easily digestible and structured for claw design.
The final data, findings, and dashboards will be presented to the board. So make sure everything is accurate, everything is included, and with proper action steps. Ask if you have any questions.
I always like to end with this because then Claude easily can ask you anything that is kind of missing in your prompt. So it asked me a couple of questions, and then it got to work.
And it then created this new analysis folder. So here you have all the different data files. So cash flow, channel performance, uh, channel revenue, customer type, dashboard data, discount analysis.
Basically, all the data easily structured and analyzed for Claw Design to use. It also made a separate design folder where it had the brand design tokens, where it kind of gave claw design some instructions for the actual design part, and it made this read me file with instructions for Claw Design.
I then went ahead and said, you can now make me a prompt for Claw Design to create a complete dashboard with different menus slash drop down bars for the different sections. It should look advanced and as visually impressive and unique as possible. I want three prototypes that are different and unique to each other so I can decide which one I want to use.
It should follow the brand style guide at least briefly, but it doesn't have to follow it 100% so we get some variation in the different prototypes. Ask if you have any questions again.
Then said it wanted to make earthy editorial, dark command center, and bold data studio. I said I didn't really want the editorial one, so please change that out. It then changed that to a summit altitude prototype instead, and it made this full prompt to give to Claude Design.
And it added it into the folder that it made. And what I then did is basically just uploaded all of those files into Claude Design, made sure that it was using Claude fable five, obviously. Same thing goes for co work right here.
If you can't see it, make sure to just update your Claude app, and you should be able to use fable five. I then just told Claude design that I have attached all required data files, read the JSON files and the README MD file, and then proceed with the Claude design prompt. You don't need to say anything else because AI will then just go ahead and read all your files, and it will see the prompt that you have attached as an MD file, and then just follow that prompt.
So that's what it did, and it then went ahead and created these three dashboards. So it created this one, which is the bold data studio dashboard. So as you can see, this is a bit different to the one I showed you earlier.
In my opinion, it has some more character. It's more fun. So depending on who you're showing this data to, there's certainly a use case for using a dashboard that's more like this.
Everything is interactive. Everything works as it should. You can drag to the side which months you want to include in the data.
Obviously, it's visually super appealing and pleasing to the eye. It has all the key data. It then has the menus on the left hand side, so you can go to cash story.
45 k, 21 k left, every dollar of paper profit and more was consumed by refunds and inventory. Kind of tells more of a story than some of the other dashboards, but this is a really, really cool dashboard. You can then see some info on the returns.
One in five orders come back. You can head into sales, strong engine, shrinking basket, another really cool page right here.
Again, interactive, so you can easily change the data. You can then go into costs and marketing. Marketing is ratio locked, and you can see the action plan.
So plug the leak, then build. Priority number one should be stop the cash leak, and it gives you some action steps for how to do that, like kill the welcome 20, which has the worst return rate, rebuild the margin, go in marketing, cap at 16% of revenue, cut the worst CAC platform first, retention flywheel, and some board targets for the next review.
So this is honestly a really, really cool and unique dashboard. If I head into the next one, this is the dark command center that I showed you earlier. Again, a really cool dashboard that I like a lot.
You have menus for the months you want to include up here. Here you have the executive overview. You then have the cache story.
It's obviously the same data as the previous one I just showed you just with a different design. You then have the returns. I really like this dashboard.
It has these neon glowing kind of numbers and charts. Everything, again, is interactive.
You can then head into sales, costs, and the action plan at the very end.
And at the end, you have the summit altitude kind of dashboard. So this is what that looks like. Again, everything obviously works and as interactive.
You can head into the cache story, returns problem, sales performance, cost and marketing, and action plan.
So all three of these dashboards are absolutely beautiful, and I'm 100% certain that any of these dashboards would highly impress any board or any team member or any client that you may have. And as you can tell, it did very much follow the brand style guide that I gave it as well.
So depending on what brand style guide you attach, it will obviously kind of keep that in mind and make the designs based on that. So those are the main dashboards that I built, but I also told Claude to make a full board report as an MD file, which it did during the first analysis.
So I then just went ahead and said, now make me a prompt for the board report. I just want to make a visually stunning HTML for this report, only one version or prototype needed. Let me know if you have any questions.
It then made that report as well. So for that, I just wanted a simple HTML that I could show to the board.
And I just attached that new prompt to cloud design, and it made this one super easily. So this is just a simple report with the key findings and data and action steps, etcetera, that you can show to anyone you want to show it to, etcetera.
So this has the power of Claude Fable five. It made some beautiful dashboards. It also did a great job at actually analyzing the data and finding key action steps.
But do also keep in mind that Fable five is an expensive model. It's two x the cost of Opus 4.8. So it is a massive upgrade.
It's great. And chances are you won't hit your usage limits if you use it like a normal user because most people are not like heavy AI users. But if you are a heavy AI user, I would probably keep Fable five usage to the minimum as well only for hard or very important tasks.
And then you could easily use Opus 4.8 for more basic tasks so that you use less tokens. But if you have any small changes to these dashboards you want to make, Cloud Design also makes it super easy to make changes.
You can just say it in the chat. You can just say make tweaks right here and describe what you wanna tweak, and it does that for the dashboard you are looking at. You can also markup, and you can just click on a specific thing that you want to change, which is super cool.
So let's say it's just like one number you want to change or you want to change the color of this. You just click markup, click on that, and just tell it what to do. Makes it super easy.
Or you can edit certain parts as well, like the fonts and the background colors, etcetera. So Clone Design is an incredible tool. And then once you are happy with a design after making some different prototypes, you can just click on share and you can share a link or you can export it as a PDF, PowerPoint, standalone HTML, project archive, or my favorite is this send to option where you can just send it to Claude code.
It will hand off the project to your terminal and you can just continue working on, like, the back end or making small changes in Claude code after you have actually made the first prototype of the design. So you just click send, and you get this text you can just copy or you can download it as a zip file and just hand that off to Claude Code.
So right now, using the brand new Fable five model from Claude, which is by far the most powerful AI model on the market right now, this is the absolute best way of creating super advanced and beautiful AI dashboards by combining Claude CoWork and Claude Design. If you have any questions, do let me know down below.
Or if you have any better methods or better prompting that you personally use, let me know down below in the comment section because I always want to learn and hear how other people use it. If you are interested in learning more about this, we do have a full AI Impact School community with full Claude courses. We give away starter repos, and in the next few days, we are also adding our complete infinite brain system.
Instead of a second brain for your AI, we have basically created a full infinite brain operating system that works with the AI model of your choice and just takes your entire AI experience and output to the next level. So you can check out this video to learn more about it, but the link to our school community will be down below.
We are increasing the price on June 15, so keep that in mind. Make sure to like and subscribe to the channel. With that said, I hope you have an amazing rest of your day, and I'll see you guys in the next
The Hook
The bait, then the rug-pull.
Three production-ready financial dashboards, built in fifteen minutes, from raw CSV files — no designer, no developer, no BI specialist. That is the opening claim, and the next twelve minutes walk through every step of exactly how it was done.
Named ideas worth stealing.
01:44model
Two-Stage AI Handoff (CoWork to Design)
- Stage 1: CoWork as analyst
- Stage 2: Design as renderer
CoWork ingests raw data plus brand guide, produces structured JSON and a README design brief. Design reads the brief and renders interactive HTML prototypes. The design prompt lives as a file — no manual re-prompting.
Steal forAny client deliverable workflow where analysis and visual output are separate steps
02:38list
CoWork Prompt Anatomy
- Folder context (what files are in scope)
- Skill invocations (data-exploration, analyze, data-validation)
- Analyst persona (Big Four, 10 years experience)
- Output spec (structured for the next tool)
- Accuracy requirement (board-ready)
- Ask-if-questions close
The six components that make a CoWork data analysis prompt complete and reduce mid-task interruptions.
Steal forAny multi-step data analysis task where output feeds a downstream tool
How they asked for the click.
VERBAL ASK
11:07product
“We do have a full AI Impact School community with full Claude courses. We give away starter repos. We are increasing the price on June 15.”
Soft community pitch after a clean tutorial. Urgency via price increase date. No hard sell during the tutorial content itself.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
