15:57Mansel Scheffel · Tutorial
Claude Cowork Projects Just Changed Everything (AI OS)
A 16-minute screen-share tour of how to build a four-department AI operating system inside Claude Cowork Projects — no IDE required.
March 22ndThe argument in one line.
Claude's new Dashboards turn chat artifacts into query-backed BI tools, but the refresh only becomes powerful once you route your tools' data through a central Supabase database instead of reading one connector live.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude for work and want dashboards that refresh against live data instead of freezing the moment they're generated.
- You're a solo operator or small business owner who wants one source of truth across tools like Stripe, HubSpot, Calendly and QuickBooks.
- You're comfortable setting up a free Supabase project and connecting it to Claude through a connector and plugin.
SKIP IF…
- You want live dashboards without touching any database setup, this still requires configuring Supabase and a connector.
- You need production-grade data governance, the demo runs on mock data as a proof of concept, not a hardened business system.
The full version, fast.
Claude's new Dashboards tab turns artifacts into query-backed BI tools: each chart keeps the saved query behind it, so refresh re-runs the analysis against live data instead of showing a frozen snapshot. A single connector, like Gmail, can build a quick dashboard directly from chat, but the real power comes from routing everything through a central Supabase database first, the same pattern larger companies use with a data warehouse. The video demos the free Business Rescue skill, which connects tools like Stripe, HubSpot, Calendly and QuickBooks, stores every fact with one owner, forecasts cash flow, and renders a stakeholder-ready dashboard of ranked findings. The close shows scheduling the Supabase sync so the dashboard never goes stale, and reusing the same database pattern to monitor Claude Code's own OpenTelemetry usage.
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 – 00:29
01 · Claude just launched Dashboards
Cold open stating the premise: point Claude at your business data and it builds a live dashboard with the query behind every chart.
00:29 – 02:34
02 · What's different from artifact dashboards
Compares old static HTML artifacts to the new dashboard type with a four-point diagram: snapshots vs saved queries, manual re-runs vs refresh, hidden vs visible query logic, rewriting vs refining.
02:34 – 04:40
03 · Your first dashboard from one app
Builds a quick email-activity dashboard straight from a single Gmail connector by asking Claude a plain-language question, showing received/sent volume and weekly charts.
04:40 – 06:32
04 · Why you want a central database
Explains why a single-connector dashboard can't track history, then walks through routing data into a central Supabase database via a cloud receiver (Edge Function) before building real dashboards.
06:32 – 10:07
05 · Business Rescue: a live dashboard for your business
Runs the free Business Rescue skill end to end: loads the skill, builds a mock customer-truth dataset in Supabase, and renders a stakeholder dashboard with cash-flow findings and a ranked ledger of problems.
10:07 – 12:41
06 · Explore it, question it, keep it fresh
Shows the dashboard's built-in comment/chat function for drilling into individual findings, inspecting the backend SQL behind a chart, and scheduling the Supabase sync so the data doesn't go stale.
12:41 – 13:29
07 · Monitor your AI with OpenTelemetry
Closes by pointing the same Supabase-plus-dashboard pattern at OpenTelemetry logs to build a monitoring dashboard for Claude Code's own usage, cost and reliability.
Lines worth screenshotting.
- Claude's new Dashboards keep the query behind every chart, so hitting refresh re-runs the live data instead of showing a frozen snapshot like old HTML artifacts did.
- A dashboard built from a single connector, like Gmail, only reads live each time it's opened, it isn't stored anywhere, so you can't track how a metric changed over time.
- Routing data through a central Supabase database before building dashboards mirrors how larger companies run a data warehouse: pull everything into one place once, then report on it many ways.
- The Business Rescue skill builds a customer-truth model where every fact has one owner, and records that disagree across connected systems get flagged instead of silently trusted.
- A free Supabase account caps storage at 500MB, enough for a proof of concept, but real ongoing business data usually pushes you to the roughly $25/month tier.
- Connecting Supabase to Claude needs both the MCP connector, for the live query link, and the Supabase plugin, which carries schema, security and efficiency best-practice skills the connector alone doesn't enforce.
- The demo's headline finding, $142,600 in completed jobs never invoiced, came from cross-referencing field-service completion records against QuickBooks invoices.
- A cloud receiver built as a Supabase Edge Function still depends on a local machine staying on and active to generate the usage data it's listening for.
- Dashboards ship with a built-in comment and chat panel, so you can click a specific finding and ask Claude to explain or fix it without leaving the dashboard.
- Scheduling the Supabase sync, daily or weekly depending on the business, is what keeps a dashboard trustworthy, since it only ever shows data as fresh as the last import.
- The same Supabase-plus-dashboard pattern used for business data can be pointed at OpenTelemetry logs to build a cost and reliability monitor for Claude Code itself.
Why dashboards need a database underneath.
WHAT TO LEARN
Claude's new Dashboards tab turns a chat artifact into a query that stays wired to your data, but the refresh only matters once you're pulling from a real central database, not a single connector.
02What's different from artifact dashboards
- A static HTML artifact freezes the moment Claude generates it; a Claude dashboard keeps the query behind each chart live, so refreshing re-runs the analysis against current data.
- If the underlying numbers change, you no longer have to ask Claude to regenerate the whole page, you just hit refresh and the saved query re-fetches for you.
- Inspecting a dashboard's logic is built in: you can open the query and refresh time behind any chart instead of digging through generated code to check the work.
03Your first dashboard from one app
- You can build a working dashboard from a single connector, like Gmail, just by asking Claude a plain-language question, no database setup required to get something on screen.
- A single-connector dashboard only reads live each time you open it; it isn't stored anywhere, so you can't track how a metric changed over time.
- To see trends and history instead of one live snapshot, the data has to land somewhere persistent, which is the problem a central database solves.
04Why you want a central database
- Centralizing data in Supabase before building dashboards mirrors how larger companies run a data warehouse: pull everything into one place once, then report on it many ways.
- A Supabase free account caps storage at 500MB, enough for a proof of concept; real ongoing business data usually pushes you to the roughly $25/month tier.
- Connecting Supabase to Claude needs both the MCP connector, for the live query link, and the Supabase plugin, which carries schema, security and efficiency best practices the connector alone doesn't enforce.
- A cloud receiver built as a Supabase Edge Function still depends on a local machine staying on and active to generate the usage data it's listening for.
05Business Rescue: a live dashboard for your business
- The Business Rescue skill builds a 'customer truth' layer first: every fact gets one owner across connected systems, and records that disagree get flagged instead of silently trusted.
- It forecasts cash flow directly from connected accounting data like Stripe or Xero, so the projection is grounded in actual numbers, not an estimate.
- The demo's example finding, $142,600 in completed jobs never invoiced, came from cross-referencing field-service completion records against QuickBooks invoices, a concrete pattern for finding money a business is leaving on the table.
- The final deliverable is a rescue report: a summary that prioritizes the business's biggest problems and where to act next, meant to be handed to a stakeholder, not just read by the operator.
06Explore it, question it, keep it fresh
- Every dashboard ships with a comment and chat function, so you can click a specific row or finding and ask Claude to explain or fix just that piece.
- A dashboard is only as current as its last data pull; without a schedule, you're always looking at stale numbers no matter how live the chart looks.
- Scheduling the Supabase sync, daily or weekly depending on how fast the business moves, is what actually keeps a dashboard trustworthy over time.
07Monitor your AI with OpenTelemetry
- The same Supabase-plus-dashboard pattern used for business data can be pointed at OpenTelemetry logs to monitor what your own Claude Code usage is actually costing and where it's failing.
Terms worth knowing.
- Claude Dashboards
- A new Artifacts type in Claude that stays wired to a live query against connected data, instead of freezing as a static HTML snapshot the moment it's generated.
- MCP connector
- A plug-in integration that lets Claude read and write to an external tool, such as Supabase, Gmail or Stripe, directly inside a chat session.
- Supabase
- A hosted Postgres database service with a free tier, used here as the central store that Claude's dashboards query for live data.
- Edge Function
- A small serverless function that runs in Supabase's cloud and listens for incoming data, acting as the 'cloud receiver' for usage logs in this video.
- OpenTelemetry
- An open logging standard that captures model calls, token usage and cost data from Claude Code so it can be piped into a database and charted.
- Business Rescue skill
- A free Claude skill that connects a company's tools, builds a single source-of-truth database, forecasts cash flow, and renders a stakeholder-facing dashboard of ranked findings.
Things they pointed at.
12:51toolOpenTelemetry
Lines you could clip.
00:00
“Claude just launched its new dashboards. You connect your data, ask a question, and it builds a live dashboard behind every query in the chart.”
02:23
“You might think dashboards is a pretty trivial update, but really it's probably one of the best things that the everyday business owner can have for business intelligence.”
05:35
“There is a limit of 500 megs in terms of storage... I think the cheapest tier is about 25 and then you get way more storage.”
10:47
“So this is what happened. The crew did the work and nobody built for it.”
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.
metaphoranalogystory
Claude just launched its new dashboards. You connect your data, ask a question, and it builds a live dashboard behind every query in the chart. If you point it at your business, you're going to get business intelligence, where your cash is, what's overdue, and where you're burning cash.
If you point it at the AI running your business, you're going to get observability, what you're doing, what's failing, and what it's costing you. You need both of these things. In this video, I'm going to show you how to break those things down and how easy it is to set all of this up.
I'll also share a guide with you and a prompt down below. Let's get into it. Okay, so if you head on over to your Claude account now, you should see this shiny new tab called Dashboard.
And inside here is where we get to build these magical new dashboards. You might be thinking at this point, but Mansell, we could already build dashboards inside Artifacts. And there's a very small caveat there in the sense that they were HTML artifacts, meaning they were snapshots in time.
They weren't something that we could query live from a database where you have all of this information captured. And that's a really important thing because once we have all of that data centralized, we can build pretty things that look like this. For instance, if we just minimize this, this is what Anthropic should be giving everyone, which is just observability into our cloud infrastructure.
So by taking all of the information from OpenTelemetry, which is just a backend for metrics, we can then push all of those metrics into a super -based database. and then report on them. And we can learn everything about our environment that Claude, for some reason, does not give to us natively.
So I can see what skills run, how much money I'm spending, if I was on the API plan and actually cared about that. you can then arrange the data in any way that you want to understand what is going on in your environment. I've built something like this before.
I just built it in raw HTML and set up my own listener, but now we can have this inside Claude thanks to this new dashboard functionality. And if I just hit refresh, this would pull the information straight from my Superbase database, meaning it would update every single time I did that. The other more important use case that you could have if you're a business owner is to literally grab all of the data from every system that you have put it inside a super based database, and then you can build a single source of truth for your entire business to see what problems actually happen inside your business across various systems.
I've put out an entire video on how to build just such a thing. So if you want to see how to do that, you can click the link below. But I will be showing you how to set up these dashboards and the central database layer in this video too.
But you can see here when we have this information from all of our systems like Stripe and Fathom and whatever our customers talk to us with, we can build relationships between the data that we have. So we know who hasn't paid us on time then we can set up a system that automatically reaches out to them but it gives you the business owner an understanding of where your business is winning and where it's failing so you know what to do more of and what to fix so you might think dashboards is a pretty trivial update but really it's probably one of the best things that the everyday business owner can have for business intelligence where you can literally just come over here open a chat and say to this thing hey go and build this for me so the easiest way to get started like i said you need to have your connectors set up so you want to come on over to this connectors tab And then you want to add the things where you might want to pull data on or report some kind of dashboard over.
For most of you, you'll have several systems that have this stuff in, but really it could even just be something as simple as an Excel sheet. Once you've connected everything inside here that you want to report on, you just head on over to artifacts and then click on this dashboard thing and it will blatantly ask you, what should this dashboard show?
Make me a dashboard on the number of emails that I've received and the number that I've sent and show me some behavior around that. So you could do something as trivial as this and then it would just pull from a live session. The thing that you need to understand here is that it is just reading from Gmail.
So we're not stashing anything inside a database yet. meaning that you wouldn't be able to report on that data over time, track changes and things like that. So that's always why I recommend having this database route because it actually mimics what larger businesses like the ones I work with actually do.
They have something called a data lake or a data warehouse and they query across all of their company's data whether they need it for security software, business intelligence, so they can make smarter decisions, whatever. That's the path we're ultimately going to take but I did just want to start off here to show you how easily you can use this to create yourself a dashboard.
to show anything to stakeholders, investors, friends, whatever. So you can see here after about five minutes, this thing told us that it's used to Gmail as the sources, obviously, because we told it to do that. So we just click continue.
And we now have a shiny dashboard with absolutely nothing showing yet because it's still loading. A few moments later, we have our dashboard giving us the intelligence on how we work with our email and whatever else we might want there. If we wanted to add something, you can obviously just talk to this.
But I gave this thing freedom to do whatever the hell it wanted. And it just came up with this, but it already told us a few things. Your notes to self are counted separately because it knows that I just use this thing as a memo box for myself because I'm always checking my email.
And then it will present the data to us in a way that's meaningful so that we... can potentially make other decisions on this. But this is obviously very trivial compared to what we can do with it.
So if we want to build some really smart stuff, what we need to do is have this middle layer database over here. For me, I'm going to be using Superbase. You can use whatever works for you in the cloud.
superbase does have some great functionality in the sense that we can also have an edge function which is really just a listener that sits in our cloud waiting for stuff to be centered in my case that's those open telemetry logs that send all of our data from claude usage up into our database over here via this edge function and don't worry if all of this sounds quite complicated i obviously didn't do any of this manually claude can manipulate this entire layer and do this entire process for us because i've already built the skills for you guys the final part of our solution here will be this connector that connects between Superbase and our Claude dashboard.
And that just gives us the information there that we need. Once we have all of our data stored here, though, we're pretty much . So one of the first things that you want to do for this process is to obviously sign up for a free Superbase account.
There is a limit of... 500 megs in terms of storage so depending on how much data you're going to be bringing up into super base and how long you're keeping it for that will illustrate whether you can actually get away with the free plan or whether you need to upgrade to a paid tier i think the cheapest tier is about 25 and then you get way more storage so that will come down to your business needs but certainly for a poc you don't need to buy everything straight out from there i want you to head on over to the connectors tab inside customize again this time we're going to get two super base things we want both the connector which establishes that mcp connection And then we also want the plugin.
Plugin is really good to have because it's got some skills inside it that Claude can use to align with best practices when building a database. That's around the structure, the efficiency of it, and also the security. which obviously if you're bringing in all your company data is the most important thing.
There is also super base CLI, but that's not going to work for our dashboard because the dashboard relies on this connector in order to query the database. So for those of you who love efficiency and try to avoid MCP, it's not going to work in this case out the box. The next part of this is even easier because now with the skill that I have that we'll take a look at in just a second, all we need to do is say.
Can you use my business rescue skill? Store all of the information in a super -based database and then present it as a shiny dashboard that looks really pretty and I can show to my stakeholders. All we need to do is hit enter then.
I have also added a clause to use mock data in this case because I don't want to have to blur out the entire screen, in which case I'd have nothing to show to you guys. But while this is running, why don't we come on over and have a look at what the skill is actually doing. In the sense of how the skill works, it's actually pretty simple.
What it's going to be doing over here is it creates a manifest. So it checks all the systems that you've connected via... It looks for all of their identities, their domains, and it looks for blind spots inside those systems that you might not have thought about in your business.
The second part is that it's going to keep all of the evidence inside the system. in this case we're going to be storing that inside superbase it can pull out anything from all of the systems that you connect to this and if you've connected appify which for those of you don't know is just a scraper that can scrape anything online you can even get your google reviews and several other things that people have said about you on other platforms and bring that into your database as well that way you've got all of the internal information about your business and the external information so that you can act from both points for any failures or successes that you might have i do cover setting all that up in the deep dive video that i have particularly around this skill the third thing that this thing is going to do is it's going to do a forecast so if you've connected stripe or xero whatever your accounting software is it's going to give you a cash projection and it's going to do that obviously grounded on your actual information and you can take it one step further like i showed you in the beginning
you can use this as a live thing. So you can talk to Claude and say, oh, but did you go and check this? Or can you calculate X, Y, Z for me?
So you can change things on the fly once we have our baseline and super base. The fourth thing that it does is obviously the most valuable, that is the rescue part. That takes all of the findings, all of the information that it's gathered, and it turns it into something meaningful for you so that the human or the stakeholder, whoever's gonna be looking at this thing, will be able to do something with this data.
not that stakeholders aren't human don't know why i friend my sentence that way but here we are and then finally the last part here is the rescue report and this is just a summary focusing on what your biggest problems are and where you need to steer the business next and so we can see here after 10 minutes this thing is still going it started by loading our skill it checked out the dashboard then it loaded superbase for in my case it started building the mock data for you guys it would be scouring your actual mcp connections for all of this data looking at super base stuff and then ultimately building the tables and adding anything that this dashboard is ultimately going to need there might be some pop -ups for you along the way just asking for permission and things like that but really it's nothing complicated especially if you're doing this as a poc it's not going to cause any damage to your business all we are doing here is reading the information and putting it into super base so as long as you've secured your super base account
two -factor authentication, and used the Claude plugin to build this thing securely, you should be all good. And at this point, if you're sitting there scratching your head thinking this is all too confusing, I do have a community that helps both AI builders, consultants, and business owners build AI operating systems and sell them to clients.
So if that is something you need help with, then you can just click on the link down below and you'll be able to check out my community. After that thing's done, it will again show us that it's using specific connectors for this information. In this case, we're using Superbase.
We're just going to hit continue. probably take a couple more minutes to go and illustrate all of this stuff and then we should be good to go and move on to the next part. And then it just tells me that all the queries ran successfully but we can obviously see that they didn't.
Now this might happen to you because this is obviously in beta at the time of me recording this so all you need to do is you could click on this ask Claude button or we could just say I'm getting an error saying nothing loaded. Can you check this?
The dashboard needs a fix. And there we go. A couple of minutes later and this thing is done.
So now we can make it full screen. We can come and have a look at our interactive dashboard over here. It's given us some tabs up at the top so we can see the coverage of what we have.
you can then see how things compound over time in your business where all of the problems are you can then have a look at your accounting and of course this is telling you about all of the things that it found it's not just reporting metrics that's the whole point with business intelligence is that we can have a little chat function here so if we wanted to see anything more on this all we need to do is put comments in here by clicking this little tab and then we can place a comment and say tell me more about this we could say claude tell me more about this row What's the problem in my business?
Then we'll send it off to Claude. This will open up a separate little chat window where Claude will then get back to you. And there Claude got back to us again, just telling us that this is marked data, obviously, but if it was real data, you would see it outline the problem.
So this is what happened. The crew did the work and nobody built for it. 83 jobs were marked complete in a field service app, blah, blah, blah.
Why it happened, why it's the biggest row and the fix for this. So the great thing about this is that you can actually solve all of your problems line by line if you wanted to. Of course, if you wanted to do the whole thing, then just open up a chat and talk about what your biggest problems are as a whole.
If you wanted to see where the data is coming from, we just need to leave comment mode and you can flip back across here. this would give you some information about the backend database if you cared at all to see where the information was stored inside your database you could then click on one of those rows and you can see the query that we're using inside sql the transform that it's using for the data and the result over here then there is one more thing that we need to do before we dip off though and we need to understand that the data that it's reading now is only as recent as the data that we have pulled into superbase meaning if we don't do this on a schedule you're always going to be reading this old information So what we want to do is just come down here and say, can you make sure that this updates my SuperBitAce database once a day?
You do you. You'd have to figure that out as part of setting this up on your architecture journey to figure out how often you need these results. But really for a lot of people, one day or even once a week might be perfectly fine.
After we hit enter, this will just go ahead and build a scheduled task for us. If you wanted to do it manually, of course, you could just head on over to scheduled task and create a new one by clicking set up manually. put in some custom instructions if you wanted it to do something specific, set up the frequency and the permissions.
But honestly, I would just do it in the same chat as the one that I was having with Claude over here. It will walk you through an interactive process for setting this up. And of course, you can just steer this in any direction that you want.
In terms of setting this thing up so that you can have a command center to monitor your AI operating system itself using OpenTelemetry, that is super easy to do. It would be the exact same process you've already set up SuperBase. then you can just use the prompt that i'm going to link down below and that will go ahead and just set up the exact same dashboard that i showed you at the beginning of this where we were looking at my live claude environment just copy and paste it into a new chat you can obviously change any color themes and get claude to build whatever it is that makes sense to you when you're looking at something like this and that is pretty much it for this video if you want a deeper dive i do have the data centralization video linked down below along with skills and whatever else you need Leave some comments if you do have any questions, or you can check out my community if you're an AI builder or consultant or a business owner trying to navigate this crazy world of AI.
See you guys in the next video. Thanks for watching.
The Hook
The bait, then the rug-pull.
Claude's new Dashboards tab promises something artifacts never could: a chart that stays wired to a live query instead of freezing the moment it's generated. The video walks through what changed, then builds a real one, a Supabase-backed business dashboard that catches $142,600 in unbilled work before the cash runs out.
Named ideas worth stealing.
00:45model
HTML Artifacts vs Dashboards
- Numbers in the page -> Saved chart queries
- Data changes -> Open or refresh
- Check the working -> Query + refresh time
- Change the question -> Refine the dashboard
A four-point diagram contrasting a static HTML artifact against a query-backed Claude dashboard: snapshots vs saved queries, manual re-runs vs one-click refresh, code inspection vs visible query/refresh time, and rewriting the artifact vs refining it through chat.
Steal forexplaining to a client why a 'live dashboard' claim is different from a one-off chart
04:43model
Give The Usage A Home
- Claude Code (usage/activity records)
- Cloud Receiver (Supabase Edge Function)
- Database (selected fields, daily summaries)
- Claude Dashboard (open or refresh to query)
- Local Inventory Job (skills and health checks)
An architecture diagram showing local Claude Code usage logs flowing through a Supabase Edge Function cloud receiver into a database, which a Claude dashboard then queries; a local inventory job runs daily snapshots, and the whole pipeline still depends on a running machine to generate activity.
Steal forarchitecture diagram for any 'local tool to cloud database to live dashboard' build
How they asked for the click.
VERBAL ASK
09:17link
“If you're sitting there scratching your head thinking this is all too confusing, I do have a community that helps both AI builders, consultants, and business owners build AI operating systems and sell them to clients. So if that is something you need help with, then you can just click on the link down below.”
A soft, single mid-roll pitch framed as confusion-relief rather than a hard sell, placed right after the most technically dense section (building the mock dataset), so it lands as sympathy for a complicated step instead of an interruption.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
