San Francisco Tableau User Group 30 Sep 2026
"From Idea to Extension: Creating Custom Viz Extensions with Claude" with Michael McCusker
54 min 9,031 words transcribed
Summary
The San Francisco Tableau User Group's September meeting pairs a career story about resilience with a live build of a custom Tableau Viz Extension using Claude.
Host Celia Fryar opens with community resources, including the DataFam getting-started material and the Tableau Community Slack, then recaps the Tableau news from Dreamforce: Tableau Studio, Tableau Data Apps, proactive intelligence, and Tableau Knowledge feeding resources such as PDFs and reference guides into the semantic model. Iron Viz is around the corner, and last year's winner Ann Pregler is speaking at the Austin TUG in October.
In the community spotlight, Eric Lacy, a term professor leading undergraduate analytics at the University of San Francisco (the group's sponsor), maps a 35-year career against the evolution of technology and the economy. The path runs from COBOL programming and an early ETL process at Hewlett-Packard, through a dot-com startup that burned through six million dollars in a year, Sequent, 2bridge and the 2001 tech bubble burst, to Deloitte, an independent consultancy, Business Objects and finally teaching at USF. The lesson: resilience comes from adapting your skills as conditions change, and AI is the 1997 of today.
Michael McCusker, a new Tableau Ambassador and Sr. Tableau Tech Lead at DaVita, then shows how to take an idea to a working Viz Extension with Claude. Viz Extensions are little websites, HTML and JavaScript hosted somewhere, that receive data through the Extensions API and render what Tableau can't do out of the box. Starting from a single project folder with a mock-up image and a skills file, three plain-English prompts build a KPI card: a first version that works with two drags onto the Marks card, a second that adds tooltips and a settings panel with period comparisons and dynamic number abbreviation, and a third that fixes a UTC time zone bug, adds goal tracking, a floating settings panel and dark mode, and writes the lessons back into the skills file, which grew from 439 to 628 lines. Michael also covers hosting options, an earlier REST API project that migrated a team from Tableau Server to Tableau Cloud, and where to download the skills file and prompts. In Q&A, Joshua Vijeh notes that a 21-minute prompt beats 21 hours of workarounds, and Michael explains why a reusable extension makes KPI cards consistent across a corporate team.
Listen to this TUG
Prefer to listen? This meeting is also an episode of the Tableau TUG recordings podcast.
Key points
- Resilience in a data career comes from adapting your skills as technology and economic conditions change.
- Learn AI the way you learned Tableau: pick something you love and use AI as a tool that extends your work, not a replacement for it.
- Custom Viz Extensions are little websites made of HTML and JavaScript; Tableau sends them data through the Extensions API and they render a picture back.
- Give your AI agent a single project folder as its working directory, with a mock-up image of what you want to build.
- A skills file holds the rules for a build plus your own preferences, and grows with every lesson so the next build starts further ahead.
- Ask the agent to plan before it writes any code, and to write a README with setup steps when it finishes.
- A KPI card built as a Viz Extension needs no extra calculations or multiple sheets, just a date and a measure.
- Stay in the loop and validate the numbers: Tableau passes dates to extensions in UTC, which shifted sales into the wrong month.
- A reusable extension creates a consistent KPI standard that any developer can recreate in seconds.
Jump to a chapter (37)
- 0:00 Welcome and meeting format
- 1:27 DataFam resources and the Tableau Community Slack
- 2:13 Dreamforce recap: Tableau Studio, Data Apps and proactive intelligence
- 3:18 Tableau Knowledge and the semantic model
- 4:25 Iron Viz is around the corner
- 5:26 Community spotlight: Eric Lacy
- 6:53 Hewlett-Packard, COBOL and early data warehousing
- 8:19 Dot-com startups and Xaos Tools
- 10:01 Sequent, IBM and 2bridge
- 11:59 The tech bubble bursts
- 12:23 Deloitte and starting ECL Solutions
- 13:29 Business Objects and independent consulting
- 14:07 A call from USF
- 15:40 Resilience, and AI as the next opportunity
- 16:34 Introducing Michael McCusker
- 18:13 Michael McCusker: from idea to Viz Extension with Claude
- 19:41 Learning AI as a tool
- 21:48 What Viz Extensions are
- 22:36 The plan: a KPI card in three prompts
- 23:18 Project folder, mock-up and skills file
- 24:50 Prompt 1 in plain English
- 26:31 The skills file
- 28:13 Watching Claude build
- 29:09 Stage 1: hosting the extension locally
- 30:52 Loading a local Viz Extension in Tableau
- 32:33 The iteration loop and Prompt 2
- 34:59 Stage 2: tooltips and a settings panel
- 37:57 Numbers that don't match
- 38:38 Prompt 3: fix it and explain
- 40:24 The UTC time zone bug
- 41:05 Stage 3: floating settings, goals and dark mode
- 44:04 A skills file that grew
- 44:31 Hosting options and Tableau Data Apps
- 45:33 What's next: the REST API and a cloud migration
- 47:51 Download the skills file and prompts
- 48:36 Q&A: build time, practicality and consistency
- 52:11 Wrap-up
Transcript
0:00 Celia Fryar: All right. Well, I think that we will get started here. Um, I, I'm so glad to see so many names I recognize. I'm just gonna encourage you to put in chat where you're dialing in from. Thank you so much for being here.
0:13 Uh, we have a great lineup and a full program today, so I'm gonna dive right in.
0:18 And, um, Palm Springs. I forget that you're in Palm Springs. How, how wonderful is that? So speaking of chat, um, we are gonna leave that open and do our best to respond to your chats. Please jump in, ask questions, make comments. We're super… It would be… make it way more fun for all of us, including our speakers. Great to see you, Ken Wall.
0:38 And the recording will be posted in a few days. Um, uh, our friend Chris McClellan out of Australia does such a great job with all that, so we're really grateful to him for that, what he does for our community. Little quick outline for our time together. Have a couple of quick announcements, a little bit of a few slides they've shared with us regarding a little bit of a recap from the keynote, uh, that happened at Dreamca- Dreamforce.
1:01 And then we have, uh, Michael McCusker h- here to talk about one of our top topics right now, Claude and, uh, and Tableau. And then, uh, we're honored to have Eric Lacy gonna… He's gonna be our community spotlight, and we'll, we'll get… I'm gonna let them introduce themselves in just a few minutes. But he is, um, part- largely responsible for us being able to have had this TUG and then have USF as our sponsor, so.
1:27 If you've not jumped into the, the DataFam resources, I just want to encourage you to do that. Um, there's a lot of getting started resources, a lot of new free training, asynchronous programs that are available if you're early in your start with Tableau. Uh, additionally, there is a Slack channel that is incredible. Last time I checked a week or so ago, there was about 24,000 of your closest friends waiting to c- talk with you, share ideas, troubleshoot, you name it, around the globe.
1:54 So, uh, it makes for an am- amazing place. Feel free to grab this, uh, QR code, and I'll make sure that I slide back around at the end and put this up again, so if anybody didn't get it now. But, uh, jumping into the Slack channel is, is powerful. You'd miss a… You'll miss a lot of, of really good connecting otherwise.
2:13 Dreamforce happened a week before last. It seems like it was just yesterday for me, but it was my first time to be there in person.
2:20 Uh, there were some pretty big announcements, and that I wouldn't pretend to fully understand quite yet, but we're gonna be finding this out over the next few weeks as they start to roll them out. But Tableau Studio is gonna bring us some… a new interface where we can switch back and forth, as I understand it, as they demoed it, between, uh, a cloud environment or a Tableau Desktop environment.
2:39 Well, it's supposed to be coming out at the end of, uh, October, so that should be a lot of fun, and it'll be amazing to see that happen. Tableau Data Apps, um, we're gonna have, uh, more that are deployed and governed and, um, again, coming soon, November, so that will be exciting. And then the, one of the ones that I really connected with is proactive intelligence.
3:02 Not just if we're getting alerts, but that we get to set things in motion where it comes to us and lets us know and/or takes action on workflows that we set up for ourselves and for clients. So they're, they're looking at that i- in, uh, October as well. So October ought to be a really huge month for Tableau.
3:18 So, uh, Tableau Knowledge, I saw that in the Slack channel I was referencing, I saw an announcement about that being GA'd today. So, uh, I'm sure that'll be rolling out across the environments as over the next few weeks. Um, but being able to see the resources that you have as a graph and having those things talk to each other, including things like reference guides, PDFs was one of the things they demoed at Tableau Conference, uh, at the keynote for, for Dreamnu- Dreamforce is T- Tableau keynote.
3:46 There we go. The words. Um, but any source to be able to bring those things together and feed them into your semantic model.
3:55 So, uh, it should be an exciting fall and wintertime for Tableau with some fresh takes on things. There's a lot of emphasis these days on sharing, uh, skills that are being built on the, uh, AI side of the house, on pre-built apps, on using MCP and more, um, APIs, um, more than just the read-only MCP, but a lot of where we also can interact and do creates.
4:20 And so we do encourage you to jump in the Slack channel. You'll, you'll hear more about that as we go. One thing we were talking about just at the top of the hour was Iron Viz is around the corner. If you are a, a dashborder, you really need to consider being involved in this process. The… They open up the topic and the dataset selection kind of thought, thinking usually s- later in October.
4:42 Ann Pregler was our winner last year. She did a phenomenal job. A lot of, um, just a really, um, a thoughtful, uh, done by her, like, creative… One of the, one of the criteria, by the way, as, as they have people judge it is, was it done in Tableau or was this, like, ported in from other environments? And what she did was truly masterful. And so anyway, uh, she's also gonna be speaking at the Austin TUG, uh, during October, so if you want to come hear her story in person, feel free to look us up.
5:15 So these were the kind of the final viz from those three, uh, final contestants, and it was quite the spectacle in, in person there. So it was really amazing. So, uh, with that, I'm gonna pass the baton to our, uh, first, um, speaker today, um, Professor Eric Lacy, somebody I have a lot of respect for and am really thrilled that he's willing to step up as our community spotlight today.
5:39 Eric Lacy: Um, so I thought I'd take you through my professional journey, um, basically post-graduation of, uh, uh, from, from San Fr- San Francisco State.
5:50 Uh, and I'm gonna spend a lot of time on this slide. You guys should, uh, basically see some pretty interesting things. So what I'm doing is, uh, is aligning the sort of technology evolution with the, uh, the economic, uh, you know, huge events, and then how I adapted my career, uh, across all of these events.
6:11 Um, as I stated, I started, uh, out at San Fr- San Francisco State University, uh, basically getting my, uh, undergrad in information systems.
6:20 And this was a time when, um, there wasn't a lot of, uh, technology companies out there. Primarily, you had, uh, a lot of, uh, uh, uh, professors that were teaching you a lot of theory. So I didn't have the opportunity to kinda lean on, like today, where there's a, a plethora of, uh, uh, of, uh, technology, uh, companies where folks, you know, can– we can find mentors and sorta help out with, uh, you know, guiding you through as you start your career.
6:53 Well, I didn't have that opportunity, but I did have the opportunity to start working at Hewlett-Packard in 1990.
7:00 Uh, there, I started off as a COBOL programmer for if, you know, anybody knows COBOL.
7:06 And, uh, I was working on their accounts receivable and credit application, uh, they called it Armada. And this is when, uh, you know, basically applications were developed in-house. They didn't have, you know, at– SAP or Oracle, large ERP systems. You know, basically all the applications were developed in-house. So, uh, along my COBOL journey, um, that went to about 1994, uh, I had the opportunity to build a ETL in COBOL, um, actually a extract, transform, and load process in, uh, in COBOL, uh, that, uh, you know, basically starting off, uh, doing data warehousing solutions.
7:50 Client-server was around, so, uh, we basically had, I believe it was an image database that was talking to Crystal Reports. Now, I don't know if you can remember Crystal Reports, but, uh, Crystal Reports was, uh, you know, was, was the tool de, uh, du jour. Uh, we had, uh, an application that allowed for us to, um, essentially interface with this image database and also build metadata in Crystal Reports and allow for us to create reports, and I thought that was interesting.
8:19 So over the next four years, uh, I, you know, basically was, you know, getting kinda hungry. More or less, all my friends were going out to, uh, startup land, 'cause if you can recall, uh, ba-basically, uh, mid-1995, here comes all the commercial internet players, right? You had things like, uh, eToys and, uh, uh, what else? Uh, well, Amazon, uh, you know, uh, Pets.com, right?
8:47 So you had all these different commercial internet, uh, pop-ups that were coming. So this was s-essentially the start of the tech bubble.
8:54 And so I basically left, uh, Hewlett-Packard and went to work for this company called Xaos Tools. Xaos Tools basically created these plugins, uh, for Photoshop.
9:06 And what they wanted to do was, is essentially leverage the internet, uh, to essentially provide, you know, customized, uh, components for, I think it was a picture app uh, that allowed for you to take pictures and kinda doll them up with, uh, you know, different, uh, um, graphic splashes and so forth.
9:25 Uh, well, basically that went on and, uh, during that time in Xaos Tools, we basically– they blew through six million in essentially one year, but they weren't making any money.
9:37 So I basically had a wake-up call at that point, going, "Hmm, I think I need to go back into, uh, you know, uh, large company land." But I basically was taking my experience that I, that I got from Hewlett-Packard in terms of, uh, consulting.
9:53 So I was basically consulting, uh, data warehousing and business intelligence solutions, uh, for internal HP customers. So I thought, "Hmm, I think I'm going to go work for a-another large, uh, company." And there was a company called Sequent Computer Systems, and Sequent Computer Systems basically had this hardware that was, uh, basically able to handle large data warehousing solutions, uh, uh, had in-memory, uh, uh, uh, uh, basically compute, which at that time was, was really big with, with data warehousing solutions.
10:31 Uh, and at Sequent, I had the opportunity to, to work at various, uh, cust– uh, you know, basically, um, clients that, that were not internal, so basically real clients.
10:42 Uh, and a lot of them were, were, were a lot of the dotcoms, some of the big players that, uh, were happening at the time. I had the opportunity to, uh, uh, to work for those companies.
10:51 Uh, Sequent Computer got acquired by IBM. Uh, this was in 1988.
10:58 And I was thinking to myself, uh, "I really don't necessarily wanna be at IBM." Uh, y- you know, it was great that Sequent actually, you know, got acquired and, you know, again, this, um, you know, essentially accelerated the stock options that I got. It was really good. However, you know, IBM. So I decided to jump out of IBM and went to this company called 2bridge.
11:20 And 2bridge, it was basically a software company. I want you to think about, like, Wix and, uh, uh, you know, some of the, the, the, um, internet, uh, uh, HTML, uh, you know, creators. So it allow-allows for you to create, uh, HTML pages, and it was all software driven.
11:41 So there I was actually a consultant, uh, essentially, uh, delivering on this 2bridge software platform.
11:48 Uh, very exciting. Uh, you know, we were hopefully looking at the opportunity for them to go public. I think we were actually in the, in the midst of that.
11:57 And burst. Basically, there, there goes the tech bubble burst in 2001.
12:04 So there I am, basically laid off from 2bridge, thinking, "Okay, where am I gonna go?" So I basically was picking up, uh, uh, you know, onesie-twosie kind of, uh, uh, e-engagements, uh, where, uh, you know, I was basically, you know, um You know, hired in as a contract, uh, uh, developer.
12:23 And then I had the opportunity to work for Deloitte Computing. So, or Deloitte that is, uh, and worked there in 2002, and I had a little, a little bit of a wake-up call there because, uh, it was a lot of work. I don't know if any of you know, uh, what it's like to work for a large system integrator such as Deloitte, but it's a lot of work. And I was a manager there, and I thought to myself, "This is something I really am not liking." And so I decided to jump out of there and take the experience that I got from Xaos Tools, uh, from Sequent, and from 2bridge, and basically start my own company.
13:05 So in 2003, I launched ECL Solutions, which essentially I was doing exactly what I was doing at Deloitte. Primarily, you know, basically doing data warehousing, business intelligence solutions, um, but as an independent consultant.
13:19 And that, that was great. Had a good run there. Um, you know, you know, had a lot of clients and, and a lot of work there working on different, different, uh, opportunities.
13:29 And I got a call from Business Objects, and Business Objects, uh, really liked, you know, my background, and so they, they hired me in to essentially, uh, work with them for, I think it was until 2008, where I decided to… At that time, you know, I, I… You know, I, I, I really liked, uh, being independent, you know, having my independence and, uh, you know, working for clients a- as, as an independent consultant versus actually through, uh, you know, a software company.
14:02 So from about 2008 to 213, 2013 that is, uh, I get a phone call in 2013, uh, basically USF asking, uh, for, you know, uh, we need somebody to come and teach a, uh, a basically a basic information systems course to our executive MBAs.
14:25 And I'm like, well, you know, I'm not necessarily really wanting to teach at this point, you know. So that was something I was gonna do in my, my retirement.
14:34 And the recruiter on the, on the other line basically said that, you know, there really is no time like the present, right? You can just give it an opportunity, right? You know, you d- you don't know how it's gonna be. So with that, I said, okay. Uh, and I basically went and taught this course, and I never looked back on educating. So I basically jumped out of my business and decided to focus on basically, uh, being an educator in higher ed.
15:03 Uh, I was an, an adjunct through 2019. Uh, I had to get my master's degree so I can continue, uh, teaching.
15:14 Uh, and so I got my master's degree at USF, and from 2020 until today, I am basically, uh, employed with, you know, with, with USF as, as a, uh, uh, as a term professor that, uh, uh, basically I'm leading a, uh, the undergrads for, uh, the analytics department. So undergrad an- analytics.
15:40 So I think the net-net of this story primarily is, is that resilience that I had came from adapting my skills through this journey, uh, and really looking at, you know, what's going on with technology as it, as it, as it's evolving and the economics, economic conditions at that time. So that kind of… That was my, my drive, uh, to, to, to bring me to where I am now.
16:06 Now, I'm at the point now where I want to pass the baton off to, uh, the, you know, next generation, uh, of, uh, you know, technology folks. I think AI is, is a really big opportunity. I think that is the 1997, uh, of, of today. Uh, and being able to be resilient and adapt, I think will make, uh, that journey successful.
16:31 Thank you.
16:34 Celia Fryar: That's, that's awesome. I love how you've made the tie to the transitions that happened in industry as people have been curious about the what will the workforce look like after this wave settles. I feel like that resilience in, in what you've discussed here is just invaluable. So thank you so much. Um, quick, any quick questions for-
16:54 Eric Lacy: Yeah, any questions?
16:55 Celia Fryar: For E- for Eric?
16:59 Eric Lacy: No questions.
17:00 Celia Fryar: Awesome.
17:01 Well done.
17:01 Eric Lacy: I guess I addressed everything.
17:03 Celia Fryar: Okay, can run on.
17:04 Eric Lacy: Well, thanks. Thanks, everybody.
17:05 Celia Fryar: Thank you. Appreciate that.
17:07 All right. And with that, let's, um…
17:11 I'm gonna pass the baton on to, uh, Eric… Uh, sorry, Michael McCusker. And, um, Michael, I've had the pleasure to meet… I guess we met at Tableau Conference this year probably, right?
17:24 And, um, he has become renowned in our, uh, user group circles and leader circles because he's such an excellent communicator, and, uh, leans in early with some of the progressive or new cutting edge kind of techniques that we use for Tableau, and has made, um, them very approachable to very many people, so myself included, so I really appreciate that.
17:48 He was awarded Tableau Ambassador this year, which is for sure, um, probably overdue. Um, and I didn't realize that you had, uh, the Viz of the Days, but I'm not surprised. You have amazing work and, um, so, uh, with that, let me pass the baton to Michael. We were very fortunate to be able to get on his calendar. He's been in Japan the last couple weeks, and so I really appreciate you being willing to jump in today.
18:13 Michael McCusker: Yeah. Thanks, Celia. Uh, can you see my screen okay? And-
18:16 Celia Fryar: I can.
18:17 Michael McCusker: Great.
18:17 Celia Fryar: Looks great.
18:19 Michael McCusker: All right.
18:20 So, um- Let me just get orientated a little bit here. So this presentation, I'll be jumping around a little bit, so please bear with me. But, um, this presentation was kind of born out of, uh, Celia had me on to talk about a different topic. Uh, I was talking about map layers, something that I'm really passionate about, and she just liked the last five minutes where I talked about some viz Extension work that I had been doing, and so asked me to spin up a fuller, um, presentation on viz Extensions.
18:50 So, uh, really excited to do this. And I'm, I'm, I'm glad that you did, Celia. I- I'm really excited about this presentation, where I'm gonna talk about taking something from an idea to a working extension and creating custom viz Extensions with Claude or whatever AI agent you happen to be working in. Now, Celia already covered this. My name is Michael McCusker, uh, Tableau Ambassador this year for the first time.
19:14 Uh, really, really excited about that. I help run a couple of different Tableau user groups, and I have, um, been working at a company called DaVita for about the past six years, and we're a healthcare company that helps provide… helps patients with kidney disease by providing them with dialysis. And I- I've worked with them for six years. I've worked with Tableau for, uh, and then data for about the same amount of time.
19:41 And it's the last couple of years, especially 2026, it's kind of been hard to ignore the fact that AI is everywhere, and it's very prevalent in our industry, in data, and in tech. And I found what Eric just said really, really interesting about resiliency, because I think like many people, uh, I've spent most of 2026 kind of reeling and a little concerned that I no longer have the skills necessary to sort of keep up with how quickly that the industry is changing.
20:12 And so I recognize that I need to go out and learn AI, but what does that mean to learn AI? It's such a general, all-encomp- encompassing term. And so I just had no idea where to start. So I thought about, how did I learn Tableau? And when you sort of… Beg your pardon. I wanted to learn AI in such a way where I was using it as a tool. I didn't want to replace things that I was already doing.
20:37 I wanted to harness it and expand what I was already working on.
20:42 And I thought about how I learned Tableau. And when you learn Tableau, they tell you to viz what you love, right? So take something that you're passionate about and go and build a dashboard. And it doesn't matter if nobody looks at this dashboard. I have a dashboard up on Tableau Public, and it's all about a comedy theater arts festival that happens every year in Edinburgh called the Edinburgh Fringe Festival, and I love it.
21:05 And I use this as a sort of demo dashboard to just practice basic skills and then add more advanced techniques in as and when I learn them. It's had very view- few views on Tableau Public, but that doesn't matter. This was just for me.
21:21 And so I wanted to do something similar when it came to learning AI.
21:25 And so what's something that I love? Um, I love Tableau, and like many of you. Um, and there's these areas of Tableau that I've always been interested in exploring, but for whatever reason, I don't have the technical skills or know-how. Um, I wondered if Tableau, uh, AI would be able to help me explore these more. And so today I'm gonna talk about custom viz Extensions.
21:50 Now, if you're not familiar, um, custom viz Extensions are basically like little websites. Um, they're mostly made up of HTML and JavaScript, and you have to put them somewhere. You have to host them. And they might be up on the internet, they might be on a server at your company, or on your local machine. And what kind of happens is data is passed via the extension API to that location, and then it renders a picture and passes it back.
22:19 And that gives us a lot of really cool capabilities and flexibility to kind of build something that is not possible out of the box with regular Tableau.
22:29 So, um, we are going to try and build one of these viz Extensions today. Um, and the example that I've chosen is a KPI card. Now, if you've built a KPI card in Tableau before, um, they're great. I love them. I think they're really… Uh, they deliver a lot of information really fast. Um, but generally speaking, they take quite a few calculations. Sometimes you're doing it over two or three sheets.
22:56 And once you've built it, it just lives there in your workbook. You, you need to rebuild it if you want to do this again.
23:03 So we're gonna attempt to build this as a viz Extension, and we're going to build one KPI card. Uh, we are going to just use two shelves on the marks card, and we're hopefully gonna do that in just three steps.
23:18 So couple of things that we need to get started.
23:22 Uh, we need a folder on our computer. And inside that folder, you can see this folder, is SFTUG nine thirty twenty-six. And I've just placed a couple of things in there to help me get started. I've placed an image of what I want to build, a kind of mock-up. And then I have this second document called a skills file, and I'll come back to that a little bit later on.
23:45 And then of course, you need an AI agent. Today I'm gonna use Claude, but you can achieve this with other agents. And so if I jump over, this is my folder that I'm gonna be working in today, SFTUG. And kind of think of this as like a project folder or a working directory. And it's seems silly to say that it's very obvious that you'd need a folder, but this becomes very important later on.
24:10 And it's just something I never thought about. I thought these things just lived all over the place. But no, a single folder, a single working directory, and things are gonna, uh, work out really smoothly for you.
24:22 And then, um, we have the image and the skills folder.
24:27 And then of course, we have Claude. And I'm gonna just kick Claude off with, uh, a prompt, and don't worry about that just now. I'm gonna cover what that prompt is doing. But I'm kicking this job off in the hope that maybe by the time I've covered the next section, this bit will be done. So just jumping back to my presentation, let's take a look at that prompt.
24:50 Now, quite a lot of text here. I don't expect anybody to read through it, but you will notice that it's written in plain English. Um, and I'm gonna highlight just a couple of the points.
25:02 First of all, I'm telling Claude to read that skills folder that is sitting in my folder. Uh, that skills document that's sitting in my folder.
25:11 And this skills document, I'm gonna go into more detail in a second, but it's kind of like, uh, the scaffolding, the wireframe of how to build a viz Extension.
25:20 I'm telling it I want a KPI card that looks like the image that's already in the folder, and I'm importantly, I want my first step to plan out the, before it starts to write any code.
25:32 And then, um, when it's finished, I want it to build a sort of readme file that gives me very clear instructions of how to, um, use this viz Extension, how to access this viz Extension, that kind of thing.
25:45 So let's jump over to Claude and see how it is getting on.
25:53 Oh, I missed a very important step.
25:57 Let me just go back real quick.
26:00 Right, back in Claude. Um, very important, um, otherwise this happens. We need to tell Claude which folder it has access to. So I'm going to go and add that folder, uh, which was this San Francisco TUG folder, and then allow access. Without that, it's not gonna know what it's looking at. And then I'm gonna re, uh, submit that prompt.
26:25 And whilst that's working, I'll jump over back to the slide deck, and let's talk about that skills folder. Now, it's just a text document, and it has all of the instructions of how to build a viz Extension. Now, you do not need to start with this.
26:44 However, you will thank me if you do, because I developed my first skills folder by, um, trial and error, and it was a lot of steps, a lot of iteration, a lot of back and forth with Claude, and ultimately, you do end up with a good skills file, and you could argue it's a really good way to learn AI. However, um, the idea here is that you want to accelerate your workflow, and using a tool like a skills folder is an excellent way to go.
27:12 And it's just a text file. This one's pretty long. I think it's about six hundred lines or so, and it's just all of the rules about how to do this build. But not just how to do this build, but, like, some of my preferences on colors and fonts and that kind of thing. And what's really great is you can add to this skills folder, uh, this skills file as you are going along.
27:36 So I'm gonna jump back to Claude real quick. And it has done some stuff, and it has a few questions for me. This is part of the planning process. And so I'm just gonna answer a few of these questions just now just to get it going. It's wanting to clarify on how to, um, display that big number in my KPI, and let's just do that and that.
28:02 And so it's gonna start to work on all of the steps it needs to build out this KPI. It's asking, "Should I go ahead and build it?" So I'm gonna say yes.
28:12 And there's something about live demos which, um, I had to think long and hard about how to do this with AI because I've built a bunch of, um, viz Extensions now, and they all take different lengths of time. So it does sometimes… I've been on demos where we just watch this little icon turning round, but there is actually quite a lot to see here.
28:34 And up in the top right-hand corner, uh, if you click up there, you can see Claude is starting to plan out the build. Um, and it's gonna show you the progress at each step, so you're not just watching the icon spinning.
28:47 And if you click on there, you can actually see steps that it's taken. So this is gonna continue to run in the background. But if any of you are old enough to remember cooking shows being on television, uh, rather than maybe, like, on TikTok, I am gonna jump over to one that I prepared earlier.
29:09 So this is prompt one, um, picked up at the same place. It's ready to build, and it has given me those instructions on how to get started in using my, um, viz Extension. It's given me some instructions that are one-time setups.
29:24 And then importantly, it's given me this step three here. This is something that I need to do every single time. I talked about how a viz Extension has to be hosted somewhere, and today we're just gonna host it on our computer. And this step is gonna allow me to sort of boot up that server and place my viz Extension on it.
29:43 So, uh, jumping over to… This is the only bit that's appears slightly technical, but it's not. So I'm gonna use my saved stages.
29:54 And in prompt one, after my Claude Code had finished, I've got this new folder that's in there. Claude has written this into my computer. I've not put any of it there.
30:04 And we can go in and take a look and see what it's put. It's got some certificates. Uh, it has that readme file, which I asked it to provide. It has a couple of these files called TRex, which we'll look at in a minute, and a bunch of JavaScript and HTML scripts as well. So
30:22 the instruction that I got from Claude in order to boot up my server is I need to open a terminal window on the card. So I'm just gonna come down here, a new terminal window app folder.
30:34 And this to me feels like hacking. It's not. It's just, I've got two words. Um, I'm gonna put that on there, and this is great. This is now telling me that my, uh, viz Extension is hosted, so I can go and look at it in Tableau.
30:48 So let's switch over to Tableau and check it out.
30:52 So I'm in Tableau. This is just Superstore. I've cleaned it up just a little bit for the purposes of this demo.
30:59 And on the Marks card, I'm gonna come down to add an extension.
31:04 And when you do that, you are presented with the Tableau Exchange, and this is cool. There's a lot of, um, Tableau extensions and extensions built by third parties that have been approved, and they're hosted by Tableau or these third parties, and you can download them from here. But we are not doing that. We are using a local viz Extension. So down here, Access local viz Extension, and then you just navigate to the folder in your computer.
31:31 So mine's is just here.
31:36 And the file that I want is this T-Rex file.
31:41 I'm gonna click on that.
31:43 It's told me that it's being hosted locally, and that was created by me. And if I click on Okay, fingers crossed, I get this information about how to use this KPI card. It's pretty bare bones right now, but it is telling me that I need to drag a measure and a date on. And it does say what it's still waiting on. So just now it says it needs those two fields.
32:02 If I drag date onto the Date card. Let's just make that continuous.
32:10 It's now just waiting on a measure. That's pretty cool. And I'm just gonna take sales and put it on to Measure. And just like that, we've built a KPI card. So click, drag, two steps, um, KPI card. No additional calculations, um, no multiple sheets. Uh, a really, really solid first effort at building a KPI card, at building a viz Extension.
32:33 And the great thing about working with, uh, an AI assistant like Claude is, if I had written this myself, um, if I'd written the JavaScript, if I'd written the HTML, this is probably hours of development.
32:46 I have very little energy now to go in and iterate on this. But this is the real power, um, of working with a, an assistant like this, is the ability to iterate, um, or the iteration loop. So this looks good for a start. But notice, uh, I've got sum of sales written there rather than just sales. And if I hover over this, there's, um, you know, no tool tip.
33:10 There's not very many options for me to interact with. So I think we'll go back and kick off stage two.
33:22 So…
33:27 I have prompt two, and actually let's look at it on the…
33:39 Prompt two is a lot wordier. This is gonna be the really fleshing out our viz Extension, so there's a lot of stuff going on in here. And again, we're not gonna read through it all, but it is still in plain English. And you'll notice that the first instruction is still to look at the skills, um, file, and this time it's also adding in the KPI card project that was built in step one.
34:02 I'm gonna– I've got a couple of quick fixes that I want from V one, but next up I'm gonna work on a settings panel. And this is my absolute favorite thing about viz Extensions, because w- when you build a viz Extension, we can't use the regular Tableau color palette or fonts or things like that. We get to design and develop the UI and the UX of the settings menu, and I love it.
34:23 Um, let me show you a couple of things that we're gonna add here. So, um, as well as the settings panel, we're gonna add this period-over-period comparison. Just now it's just month over month. We're going to throw in some tooltips and maybe a couple color palettes and some font preferences.
34:45 And let's jump back over to Tableau. Where has my mouse gone?
34:56 So after prompt two had run in Claude, this is what my file now looks like. It looks– It's still got that screenshot. It's still got the skills file, the KPI card. But it has added this extra sort of item where it's starting to give me a mock-up of some of the things that I'm building.
35:18 And I'm going to add an extension. For you, you would just have to refresh, uh, your extension that you built in step one. But because I've staged this out, I've got to add it again.
35:44 Okay.
35:47 Same intro message. This is looking good. I'm gonna drag order date on to Date and click OK, and then Sales on to Measure.
35:57 All right. So it's fixed that, um, sum of sales piece. And now if I hover over, I actually have a pretty cool-looking tool tip, and it's got a legend built inside the tool tip. But as I mentioned, my favorite thing about viz Extensions is the settings. So I've got this tiny little cog up there. I think I probably wanna make that bigger later.
36:18 And this is gonna give me some options. So just now, the big number is showing the latest period.
36:25 Um, but I could change that out to be year to date. And so now this is representing the entire year and then broken out by month.
36:32 You can change the grain. So perhaps I want to look at the quarter.
36:37 And different view here.
36:40 I'm gonna switch that back to month for now and back to now. And we have a bunch of cool, um, number format and menus. I like this abbreviated large numbers. Um, it does this dynamic abbreviation which Tableau doesn't do, where it will automatically decide what's best. Does it do K, M, B, T, depending on how successful your company is.
37:01 Uh, some information about the trend chart. I've even got the ability to add some goals in here as well. And then I've got some layout preferences. So I also have the option to s- kind of pivot the KPI card so that the big number's on the left and the trend is on the right. So I'm giving my users a lot of flexibility here. I'm gonna change that back for now.
37:24 I've got my nice color palette. Maybe I'll change it out to forest just now.
37:29 And I needed that to be back at a month.
37:34 So good-looking KPI card, and if I drag it onto a dashboard, you know, it looks, I mean, it's excellent.
37:44 And the good thing about this is now another developer can just, uh, recreate this in seconds. I've done some work here, but that work is, um, front-loaded. It doesn't need to be repeated later on.
37:57 However, um, I'm looking at this number, seventy-nine point eight K, and if I look at my tool tip, December 2026, seventy-nine point eight K, that's not ringing quite true with me. And if I jump over to just something that I've built in Tableau itself, I'm not using the viz Extension, my sales for December 2026 is actually eighty-five K.
38:20 And so something's not quite right there. So I could spend some time trying to troubleshoot what is going on there. But in fact, I think I would rather have Claude figure it out for me. So jumping over to my slide deck, let's look at prompt three. So again, I'm gonna tell it to look at the KPI card and look at the skills file before it does anything.
38:48 And then I've highlighted the problem, uh, just saying, "Hey, my numbers aren't matching. Can you look into that? And I want you to ex- fix it, but also explain to me what went wrong. And, you know, whilst we're here, let's add some polish onto the card as well. Let's add some more palettes, uh, some more color palettes.
39:06 Um, I want to get a better instruction, sort of that front menu, um, on my canvas. Uh, I want it to say 'Built by Michael McCusker' on there as well. And, um, I like my settings panel, but you noticed when I clicked on it, it covered up my whole view. So I'm gonna pop that out and have this floating settings panel. It's gonna look really, really good.
39:29 And then finally, um, it's going to ask me to… I'm gonna ask it to build that, these preferences back into that skills file." So I jump over to Claude.
39:46 And this was prompt three.
39:49 All just pasted in there, and I can see the progress that Claude made during the, this build. So it fixed, um, and identified the ti- the, the date or the time zone bug. Um, and it's built out those extra things for me. And, uh, you can see all of the steps that were taken. Sixty steps it took to build this stage three. So I'm expecting a lot of polish here.
40:16 And if I scroll down, let's see what explanation it's given me for the mismatch in numbers.
40:24 Ah, okay. So Tableau has taken the date and passed it off to the viz Extension in UTC. Um, and so some of my sales are counting in a different day when I think about being on the Pacific Coast. And so glad that we fixed that, glad that we caught that. And something that's kinda highlights the importance of staying in the loop as a person and not just letting AI run off with this, this kind of validation, iteration loop, how it kind of goes back and forward.
40:54 So it's fixed that. I'll check it when I reload this new dashboard. Um, and it has gone on and made some of those changes that I asked it for.
41:05 So let's check it out in production.
41:11 So let's add extension.
41:14 Access local viz Extension, and it's KPI card stage three.
41:21 All right. So I've got a more detailed information sheet here, um, and telling me exactly what it needs. And, um, it also has that "Built by Michael McCusker" down here in the bottom corner.
41:33 And let's take order date, drag it onto date, and make it continuous.
41:39 And sales onto measure.
41:42 Great. This is now returning eighty-five one seven five, which if I go back and check, eighty-five one seven five. So I have fixed that time zone issue. And it's also filled out the space. Um, in our earlier build, there was a little bit of white space going on, so that's something else that I identified and had it pick out.
42:01 Uh, but the… Let's take a look. My favorite thing, the settings menu. When I click on the card settings, I have this floating settings card that I can resize and move about. And what I particularly like about this is I can change stuff and apply it, and it keeps my settings open, and I can see what's going on in the background. Um, and I don't have to constantly open and close a settings window.
42:26 So let's do that to there and look at the number format. So I'm gonna have that abbreviate that number, and it's gonna do it smartly, um, and correctly picks the sort of right, um, abbreviation to do.
42:41 Then, I don't know, let's track it against a goal. So let's say 90,000. Sales are better if they're higher. And then I'll apply that. And I get this indicator against the goal, uh, along the side, this, like, long red line, and it, it'll be green if it was on goal.
43:02 And I've even given them some options where I don't really like the long bar, uh, maybe I just want a slightly shorter bar. I think that looks pretty cool. Notice everything switches around. And sometimes people like the bar to go across the top. I don't love this, but it does look good when you bring it out onto a dashboard.
43:22 I've got some layout and appearance options, including those color palettes, and if you're like me, you will love the fact that I've added a dark mode in here.
43:32 Um, and I can keep that there. And what's cool is as you click around the color palettes, it's not just affecting the viz, it's affecting my, um, my settings menu. Uh, that's changed from gray to kind of like a orange, to green. So really, really like that there.
43:54 All right.
43:56 So that was three prompts to get to a pretty good-looking KPI card.
44:04 Along the way, this is harder to visualize. I talked about the importance of that skills file. It grew. Um, it started off at 439 lines. It finished at 629, uh, 28 lines. And those additions were things like, "Hey, I like this color format," and, "Make sure you get the time zone bug correct every single time." And this skills file is now available for you to use on your next extension without having to rework the things that you've just gone through.
44:31 Something we've not covered today, uh, hosting your extension. And, um, there are some popular options. You can take this and put it up on Vercel or GitHub Pages. Heroku's pretty popular for this. Um, Tableau Ops is a website set up by Eric Summers. He's a Tableau ambassador and just a fantastic developer in this space, um, of viz extensions, and he's built something where you can…
44:55 It just takes all of the thought about hosting, um, away, and he will host that for you. Um, really, really cool there. But as Celia mentioned, uh, maybe the future of hosting builds is gonna change in Tableau because Tableau Data Apps was announced at Dreamforce earlier this year. It looked really cool. We don't have a lot of details about what it's gonna look like for us as developers, as customers, or you having to be a Salesforce customer.
45:21 Who knows?
45:22 Um, but it does kind of show that Tableau is really embracing this idea of users building their own stuff and empowering them to do so.
45:33 Now, what's next? So I just built a KPI card. It's pretty straightforward, but I think it's an excellent demo. Uh, but take that skills file and go and build something else, like this really cool, uh, viz extension. Again, this is one of Eric's, this spike lollipop chart. I love it. Um, he does a lot of dark mode stuff, but I'm gonna switch this out to light mode.
45:54 He has this floating settings menu as well. So a really good-looking viz extension there.
46:01 But that was just one use, um, of, um, AI to assist in your Tableau journey. Something that I was, uh, really passionate about was learning the Tableau REST API. And I was doing this back when, uh, Claude or the agents were more like chatbots, where you only got, uh, messages returned in the field. They didn't write anything to your computer. And so I learned Tableau REST API.
46:29 I had learned a little bit of it before I'd started exploring AI. But with the help of AI assistance, I was able to build a cloud migration tool, and I moved our team from server to cloud entirely by myself, um, and Claude.
46:44 Um, and so, and all of that came just by learning as, um, learning by doing. So I took AI, I turned it into a tool to help me with the gaps of the knowledge that I had or to really lean into the things that AI is good at and that I'm not good at. Coding is one of those ones. Um, we, there's a lot of opinions about, um, or AI is good at some stuff and not good at others.
47:11 One of the things it's really good at is coding.
47:14 So leaning into that, I was able to build a, um, cloud migration tool, which was, uh, really, uh, useful for our team.
47:23 And I also built a website, and then I built an app. I, I took the learnings from here and just applied it to other stuff. And it was not really about having a website or having an app or having a viz extension. It's just about, like, how can I pick up some of these skills and kind of keep up with how the industry's changing? How can I be resilient like Eric, um, through this kind of massive change that is happening, whether we like it or not?
47:51 So if you liked that, you'd like to try it out, I have made those skills file and those prompts available for you up on my website. That is the QR code there. And you can just click on that, and it'll download it straight onto your computer.
48:06 Uh, skills file, screenshot, three prompts, and you're done.
48:12 So thank you very much for having me on, for letting me present on this. It was a little bit of a jump-about presentation, so I appreciate your patience there. Um, but I'm happy to take any questions that you have. And any other kind of, like, all of my other AI explorations have been up on my website as a blog or under the presentations section.
48:36 Celia Fryar: Very nicely done. Not at all a jump-about. Th- very much like m- one step away from being hands-on, so thank you so much for that.
48:45 Michael McCusker: Yeah. I, I wonder if, um, that first query actually finished running. Um…
48:57 Celia Fryar: Interesting how sometimes they execute really quickly and sometimes there's a little bit more time involved. Yeah?
49:02 Michael McCusker: Yeah. It took about 21 minutes, so I'm glad that-
49:04 Celia Fryar: Okay
49:04 Michael McCusker: … we didn't sit and watch that. And I had to
49:09 to get to the stage one. But if I look at the, uh, our original folder-
49:16 Celia Fryar: Mm-hmm
49:17 Michael McCusker: … the SFTUG has now got that KPI card in there.
49:20 Celia Fryar: Yeah. Yeah, yeah.
49:20 Michael McCusker: So back at the end of stage one. So, and it's written that in there for me, so.
49:25 Celia Fryar: Yeah.
49:27 All right, team, any questions for Michael?
49:31 Oh, agenda for the weekend. Love that. Same.
49:35 So he did come and speak at, um, Austin TUG in July, and I did pop quiz him with asking him to add more onto it, and that was my next Saturday after that. So Saturday evening, to be fair. It wasn't… I had just been, uh, feeling like it was too far off my path and, but turns out, not at all. So very grateful for that.
49:59 And, and what a great-
50:00 Joshua Vijeh: It is
50:00 Celia Fryar: … methodology he's just shown as well, folks.
50:03 Joshua Vijeh: Yeah. This is, this is great, Michael. And I was taking a peek to see the comments and, and questions. I don't see too many, you know, questions, but I know, you know, it's hard to present and look at the chat. But, um, just the examples that you used were incredibly practical. I think everybody can identify with the pain of trying to create a KPI and then have a, have someone changes their mind on formatting, and they don't realize just, uh, what a daunting task that can be.
50:28 So, um, uh, the number of comments about, uh, the units, you know, whether it's dollars or even abbreviating, you know, using a decimal point, I've done that with incredible workarounds that just, you know, uh, your profi- you know, your proficiency suffers and your dashboard. And something like this, just incredibly practical. I mean, 21 minutes for that, uh, prompt, that sure beats, uh, you know, 21 hours worth of work, uh, trying to put that together.
50:59 But yeah, this, I mean, this is just a great example of, like, resiliency. Like, you know, like you tied it well with, with Eric's conversation, and, um, just the curiosity. Um, if, as long as you're curious, right, and, uh, willing to kind of roll up your sleeves and, and learn, I think this just, yeah, just incredible. Now, now I'm looking at it going, "Ah, man, maybe I should look at, uh, migrating to cloud on my own when the time comes." So this, yeah, great resources, Michael.
51:26 I, um, yeah. It's given me some ideas, too. So I'm, I'm ready to go.
51:31 Celia Fryar: Agree.
51:32 Michael McCusker: Yeah. I think the thing that we found, especially with KPI cards, like if I build a KPI card, Külli builds a KPI card-
51:38 Celia Fryar: Mm
51:38 Michael McCusker: … they're probably gonna look different. So it's kind of hard-
51:40 Celia Fryar: Mm-hmm
51:40 Michael McCusker: … to create consistency, um, like if you're a corporate organization. So for us-
51:44 Celia Fryar: Yeah
51:44 Michael McCusker: … this was like, let's build a standard and then recreate it very easily. So, um, that's why I love the KPI example. It's super relatable. Um, but, um, the best thing about this is now just go and be super creative with it. Like, um, Eric's spike one, and there's some really great examples coming out that it's a little easier for us to do it using Claude.
52:06 So, uh, yeah, I'd love to see what you all come up with.
52:11 Celia Fryar: Excellent.
52:12 All right. Well, you guys please catch, um, Michael's website there. And, uh, what a huge advantage to get to start with, um, a jump start from you.
52:22 And I would also encourage you to jump in on the, the Tableau Community Slack channel.
52:29 There's a lot happening in there that you'll miss otherwise, and everything from product announcements to int- knowledge about webinars and what's upcoming, what's working, what's not. None, it's not all just, uh, just the positives, but it's a, a mixture of, uh, good, hands-on, uh, feedback and experiences. So, um, so I just wanna thank you guys for joining us today.
52:50 And, um, s- my special thanks to Michael and to Eric for participating in our, our program. And thank you all for carving out time for, um, being here and participating.
53:02 And, well, Michael, thank you for making it so approachable.
53:06 Greatly appreciate it.
53:08 Michael McCusker: Great.
53:08 Celia Fryar: So.
53:08 Michael McCusker: Thank you, everybody. Thanks for having me.
53:10 Joshua Vijeh: Thanks, everyone.
53:11 Celia Fryar: All right.
53:12 Eric Lacy: Thank you. It was very inspiring.
53:15 Celia Fryar: Oh, thank you. Yeah, Michael and, and Eric. Such a good timing for both of you to be speaking. Thank you.
53:22 All right, team, we'll hope to see you soon. And, uh-
53:25 Joshua Vijeh: All right. Bye bye, everybody
53:26 Celia Fryar: … jump, jump in. Thanks, everybody.
53:28 Michael McCusker: That's good.
53:28 Celia Fryar: Bye.
53:29 Michael McCusker: Take care.
53:29 Attendee: Thank you.
Shorts for this TUG
- $6M gone in one year: a dot-com wake-up call 8:54
- The call that turned a consultant into a professor 14:07
- AI is the 1997 of today 15:40
- Learn AI as a tool, not a replacement 19:41
- Learn AI the way you learned Tableau 20:46
- What is a Tableau Viz Extension? 21:52
- Give your AI agent one working folder 23:59
- Start with a skills file. You'll thank me 26:41
- A KPI card in two drags, with a Viz Extension 32:02
- Why Michael's numbers didn't match: the UTC bug 40:18
- A skills file that learns from every build 44:06
- Moving Server to Cloud with just me and Claude 46:08
- 21 minutes of prompting beats 21 hours of work 50:13
- Build a standard KPI card, then reuse it 51:32