Tableau, Claude dashboards and the future of the BI developer role

0:00In my last video I talked about the new capability with Claude

0:03dashboard

0:04being supported inside of the Tableau

0:06ecosystem.

0:06Again, this hasn't shipped yet, it's coming in 206.

0:093, but it came

0:10As a surprise, and to be honest, I had to join the dots up across multiple, let's say, announcements to sort of figure out what was actually going on.

0:17It basically went Dreamforce, Claude, then Tableau

0:2026.

0:213

0:21Now

0:22there's so many questions came out of that video.

0:24There's some great questions in the comments.

0:26I highly encourage you to engage and and take part in that.

0:28But additionally, I've been having some of these conversations, some of them for the last year and a half, some of them in the last few weeks, and some of them in the last few days.

0:36the majority of those questions and conversations have been with Kirk Monroe.

0:40So I asked him back on the channel so we could essentially have the conversation back in the open.

0:44And if you look at what we covered, we pretty much covered all the talking points.

0:48We also talked a little bit about

0:49where the role of a BI developer is going, but additionally

0:52what you can do to

0:54sort of stay relevant and

0:55understand how to upskill yourself in this new context.

1:00There's also one small detail here.

1:01I had audio issues at the very beginning of this episode, so I really apologize.

1:05I tried to fix them the best I can with a bit of AI magic, and there is some ongoing issue throughout the rest of the episode.

1:11But nonetheless, I've sort of gone through the whole thing and I've ironed out as many of those issues as I can.

1:16This is just a small apology to say I'm sorry the audio is not as good as it usually is.

1:20Um but anyway, all that said and done, enjoy the rest of the episode.

1:24Cut

1:24it's been

1:25uh it's been a crazy week.

1:26It has been a crazy week

1:28In the middle of a crazy few months, but a particularly crazy week, yeah.

1:32In the in the middle of a crazy year.

1:34It's a lot of change happening all at once.

1:36Before we get into it, obviously

1:38we decided to do this

1:39emergency podcast or whatever we want to call it, emergency video

1:43about the Tableau

1:45day trap capability

1:46and day traps in general, I guess, in in in in the whole world with Claude, but it might help for you to introduce yourself.

1:52Before we get into discussion so people can understand sort of where we're both coming from.

1:56Um

1:56yeah, and thanks for having me back on the channel.

1:59Yeah, I

2:00I've been in the VI space now for well, twenty-five plus years.

2:04Like I worked as a product manager at Cognos.

2:07I worked in sales enablement SRIBM

2:09BOTAS.

2:10I worked at Tableau

2:11for a bit.

2:11And for the last five years I've been

2:14an independent, you know, part of a small tableau consultancy.

2:17We're mostly Tableau

2:18at least.

2:19And so I've seen all these trends firsthand for the last twenty-five years through BI.

2:24So it's an another interesting time

2:27for sure.

2:27And you're too humble to also highlight that you've made probably some of the most important resources in the Tableau community around data models.

2:35But as well as that, I think you and I always chat behind the scenes about

2:39Pretty much everything tableau, right?

2:41So

2:41we we

2:42we have we have an iMessage

2:44thread that's just constantly going back and forth on a weekly basis.

2:47Yeah.

2:47I

2:48well I guess the other big thing is I wrote d

2:50data modeling and tableau, which is somehow

2:52still the only book on that.

2:53And you know, you and I have a couple of good

2:55videos that have gone

2:56really

2:56well.

2:57And I think the interesting thing is

2:59Data modeling

2:59something that Tableau

3:00almost did reluctantly at the start.

3:03Because the way they started

3:04was this like

3:05collapse everything into one tier and the way they helped

3:08the way they helped D

3:09DO and Cognos

3:11was

3:11Those things need this big data model before you can start.

3:13You don't need the

3:14nurse, which is kind of ironic now.

3:16And then

3:17so it's been hard

3:18a bit of people traditionally in the Tableau

3:20community to get them to care about

3:22the data model honestly, but

3:23but um people new to Tableau love it.

3:25'Cause it

3:26'cause it is good.

3:26It's just people have been around a long time are used to not having to rely on it.

3:30And it's probably more important now than ever, obviously.

3:33If

3:34you know everything's going

3:35headless.

3:36Exactly, exactly.

3:37And so uh let's get into it a little bit.

3:39Let's set the scene.

3:40Tableau

3:41R.

3:41I'm gonna say hinting because we've not seen it, I've not used it, so I can't say this is happening.

3:46But they're hinting at this idea of being able to publish

3:49something that you have vibe coded and

3:51called

3:52using the dashboard capability

3:53directly to the Tableau

3:55platform.

3:56And

3:56the missing piece, the thing that we have not seen yet

4:00is

4:00does it both support flat

4:02files uploaded to Claude

4:04that then get published to Tableau?

4:05And does it also support the ability to build on top of a tableau

4:09published data source?

4:10I'd assume it does because you can already do that today.

4:13So those are sort of the two pieces that are not called out explicitly.

4:16Right.

4:17But needless to say, yeah, Tableau's handing over the authoring

4:20capability to a feature in core.

4:23Yeah, that's partly true.

4:24But at the same time

4:26We have Studio coming out, which was talked about a dream

4:29for.

4:30So

4:30Yeah.

4:30Um

4:31I guess I should also say a metabolic visionary, so should be careful about NDA that anything that was

4:36talked about a Dreamforce, which certainly is positioned as the next version of

4:40desktop that's more agent

4:43enabled.

4:43So I think in

4:44code

4:45So you can ask agents to build dizzys in it, right?

4:48Using normal like what you would do is drag and graph inside Tableau, or you can do agent assistant inside it.

4:54It's interesting though.

4:55I don't know if it's a hedge bedding thing, but it's

4:58but I can't imagine you could take what you built in Claude, which is going to be all in

5:03HTML and D3

5:04and associated components

5:06and drag and drop a difference in it.

5:08Like I don't they're not the same path to a dashboard.

5:11Right.

5:11So I think that's interesting.

5:12At least that's my interpretation of it.

5:14So

5:15yeah, Kirk, it's interesting, isn't it?

5:16Because in some ways, over the last year, I think there's been sort of multiple points in the community where we've all kind of seen

5:23a move like this coming

5:25where

5:26the AI vendors, whether it's Chat GPT or Claude, allow you to build what we're going to call dashboarding capabilities in their solution.

5:34But I think the thing I didn't realize would happen so quickly is

5:38this sort of retreat back to the next

5:41sort of best backstop for the BI vendors, which is

5:45to just say, actually, fine, we'll let you

5:48Have your altering experience outside of our platform, but we'll become the governed place where you host those things.

5:55You have your data connections.

5:56And to be fair, that is a moat, right?

5:58That is probably the best mode, most defensible mode.

6:01it's easier to get Tableau

6:03on a, you know, PO

6:05than it is to get anthropic on a PO

6:07without drawing attention to it, right?

6:10So

6:10yeah, I don't know what you think about that side of things.

6:13And also maybe if you could expand on

6:15How long before that becomes another sort of pl

6:19point that they have to retreat from, if that makes sense?

6:22So

6:23yeah, so a couple of points on this one I think are important.

6:25One is

6:26I don't know if they've done that strategically or tactically because they've had to, but the

6:31probably a bit of both.

6:35I do think Tableau

6:36in particular

6:37has a great

6:38potential advantage here.

6:39It's hard to predict the future because

6:41because

6:41the moat's a good one because they already have these connectors to

6:4570

6:45plus databases, whatever it is, they have hyper, which you know

6:48might again start becoming important.

6:50Do

6:51Yeah, round

6:51compute

6:52class, et cetera.

6:53The reason why I think in this case

6:54the moat and

6:55retrenchment might be good that I don't think people are thinking about is anthropic

6:59have what now?

7:00I don't I think they have like twenty five hundred employees or something.

7:03And

7:04They are in every space.

7:05And what they're trying to do, I think, is get like

7:08the skinny front end of every space.

7:10Like

7:11if they try to build this back

7:14end

7:14to support all this BI

7:16stuff,

7:17I think they're smart enough to know that would probably get them off their mission.

7:20Do you know what I mean?

7:21Because what are they gonna be a BI vendor?

7:22They're gonna be an authoring vendor, they're gonna also individually take out Adobe, are they gonna

7:27Or are they gonna just say, you know what, we're gonna get us

7:29we're gonna get the front end kind of scheme of all that stuff.

7:32Right.

7:33And

7:33and they're probably smart enough to know

7:35I'll give you a question I asked Mark Record, which was probably unfair because you

7:39still knew it at Tableau

7:40at the time, but we had a little partner

7:41session at Tableau Conference when this idea of

7:44headless first

7:45started to come up.

7:48in May and I'm like

7:49I'm also old enough to live through the like middleware vendors of the knots, you know what I mean, of the two

7:54thousands

7:54and and I'm like

7:56It's hard to make money in that space

7:58is interesting.

7:59And and Thropic

8:00probably

8:00know that.

8:01Do you know what I mean?

8:02Like

8:02to go, you know.

8:03So

8:04uh

8:04all that to say, I think Tableau

8:06's

8:07are in a great a better spot than anyone else

8:09if they

8:09if they figure this out properly, right?

8:11Plus a huge installed base that they could keep.

8:14What I do worry

8:15and I had this question is like how can you continue to charge what you're charging now

8:19when

8:19all

8:19the authoring, right?

8:21So that that's what's gonna be I think if they figure this out, that's the challenge

8:25is gonna be the hardest one.

8:27Do you know what I mean?

8:27Because

8:28in my

8:28my

8:29they

8:29said

8:29well

8:30there's all this value, whatever.

8:31Of course they give the standard answer, but I go if I'm a typicals

8:34like

8:34C

8:34level person.

8:36Like

8:36the value I see is in the UI I see.

8:38Like no matter what where the value really is, don't you?

8:41Because I'm typing in

8:42pod, right?

8:43Yeah, yeah.

8:44So that's what I think.

8:45But I do think

8:47I I I think they have a huge opportunity to do things.

8:49Like we already know that like whatever the stats are, it'll change by the day anyway.

8:53The GitHub's full of swap now, right?

8:55But imagine

8:56all these

8:57like

8:58all the

8:58Uh you know, all the and

9:00if people buy

9:01code properly, right?

9:02And so

9:02they do it through MD files in the cloud, right?

9:04Instead of like a bazillion prompts, which is insanity and we could talk about, right?

9:08But someone they could also be the host of these MD files and all the changes and they could become

9:13Like the confluence of dashboard development

9:15because they could pull these MD files apart and show people what's in the dashboard, right?

9:20It would

9:21it would slide easily into Tableau

9:22Knowledge and what they have in the semantic layer.

9:25Like I

9:25Uh it's just hard to sell middleware because people don't see it.

9:28But I but I don't think cloud's gonna do that stuff, right?

9:31That is

9:31not

9:32and that's why we're in a bit of a mess and people go, well there's a

9:35Mess of a bazillion dashboards and they don't say can

9:37it's true, but it's only because

9:39people

9:40haven't yet figured this out.

9:42I mean

9:43the users of this stuff haven't yet figured it out.

9:45So I do think

9:46I do think they someone's gonna win this, I think.

9:48Like someone's gonna

9:49dashboards clearly aren't going away, right?

9:52Probably authoring dashboards through a GUI

9:54book's dead already, which a lot of people don't want to hear, but I think

9:57It's too late.

9:58Right.

9:58And

10:00and someone's gotta like make sense of this mess.

10:02So all this stuff around governance, security, whatever's true.

10:05They've just been flippant about it.

10:06Like no one's come up with a really good

10:09demonstration to people

10:10of this is why, right?

10:12So

10:12I do think this

10:14what I expect of

10:15these data apps

10:16like

10:16HTML contained within a

10:18Twibit, which

10:20I would have used the word harness

10:21before, but now

10:22it means something else.

10:23But it's effectively like a like a harness for your dashboard, security,

10:28all these different query layers, whatever, right?

10:30Because

10:30now

10:31Like Claude

10:32came out with what?

10:32Like Databricks, Snowf

10:34a bunch, DigQuery to be fair.

10:36But through Tapo

10:37they have access to way more than that

10:39Like as an example, right?

10:41So

10:41or

10:42and and even more importantly, which I don't see Tableau

10:45promoting, which they could, they have

10:46a huge advantage on is

10:48what if your

10:48same data model that answers with a single dashboard, those tables are in different databases?

10:53Right.

10:54Someone

10:54can handle that seamlessly.

10:55To the end user, they don't even

10:56you know to the author, they don't even

10:58need to know that they're in different databases that was

11:01modeled properly.

11:02So

11:02there is a there is a possibility for a big

11:04moat there for sure, but they

11:06I'm

11:07more worried about desktop and new studio what's coming.

11:10Like I'm like, how

11:11long are people going to continue to author that way?

11:13I get

11:13you know, I guess people who are already doing it maybe, but I don't know, man.

11:17It's just so impressed with how quickly

11:19It's not

11:19it's not been um

11:21I have to say, like

11:23Tableau's rollout of this stuff is, I'm gonna say

11:26a little bit slow.

11:27And so

11:28for the majority of the people, you're still having to use desktop, right?

11:32If if I put it that way.

11:33Like you can't really

11:35mainline

11:36agentic

11:37like

11:38building of dashboards just yet

11:40because

11:40the capability has been demoed, it's been showed, but it's not really there, there

11:44in a productionized way.

11:46And then the other side to your point around um

11:49Like, you know, where does the value sit?

11:52For a long time

11:53we've had competitions like Iron Viz

11:56basically showcasing that the the value is here in the build, in the in the

12:00developer experience.

12:01The value is

12:02here

12:02in this

12:03what we get you to certify against, which is desktop.

12:06The value is

12:06look

12:07how artisan and how beautiful these public dashboards can be.

12:11And I think overnight, like,

12:13Claude is

12:14about to do things that are like

12:16just

12:16gonna look far better than everything on Tableau

12:19Public in like in in in in a in a snap.

12:22And so

12:23i

12:23yeah, customers will have a strong leg to stand on and say, No, no, no, no, you pitch the value over here and we bought into that.

12:29Now that's gone.

12:30That's the majority of the value.

12:32So you just become a

12:34A connector, right, to to

12:36lots of different data sources

12:37and you become a host.

12:38Last time I checked, that's what Databricks

12:40do, you know

12:41And and that then becomes a problem.

12:43And

12:43and and that value prop between the two starts to become a very different thing.

12:47We

12:47we'll come back to the Databricks point, but

12:49I think

12:50I think it's a

12:51Because the value proposition is changing, suddenly it's it's almost like Tableau's wearing the wrong costume to the Halloween

12:58party, right?

12:59Yeah.

13:00It's interesting how they could lock into it though

13:02because I think

13:03so

13:04the

13:04I think the Databricks

13:05challenge

13:06is going

13:07there's two challenges.

13:08There's one to Tad

13:09Well, there's one to Databricks, right?

13:10So

13:11I think

13:12not to dump on data engineers, but

13:14But the

13:15because

13:15they've had to be, right?

13:17They're pretty process focused and this moves quicker.

13:19And what Tableau's

13:20big advantage has always been

13:22is how quickly you can move.

13:23So if they say

13:24this

13:25if they can make the case

13:27that this isn't a job for data engineers is the bigger problem.

13:30It's a

13:31it is for business analysts, whatever

13:33term.

13:33They'll come up with some fancier terms than that.

13:35But you know, say business analyst.

13:37Um, who

13:37has to do it in in they're never gonna

13:40get the keys to Databricks, right?

13:42And this

13:42and this logic has to be one level higher than that

13:45because

13:46The data engineers are a bunch of data peerists.

13:48They're not

13:48the sloppiness of business process.

13:50So this has to reside in a different persona.

13:53Like, I think they have an opportunity there, Jimmy, because that's what Tableau

13:56's kind of always been

13:57anyway.

13:58Do you know what I mean?

13:59Like that's

13:59true.

14:00It's kinda how they

14:01got started and beat the Cognosas and the BOs.

14:04They also got lucky that both those companies got acquired, right?

14:07But

14:07but they beat them in in a good part

14:09because they're like, we can be more nimble, you can do it quicker, you don't need to rely on these other teams, right?

14:15They still have that opportunity.

14:17And and the reason I think Cabo

14:18was so slow and I wasn't getting data apps.

14:20You know, we got as a visionary, we got an inside peak for a lot of stuff.

14:23I'll tell you we got

14:24no inside on data apps, like

14:26zero, right?

14:27I think a problem that that ironically

14:30maybe

14:31D we didn't see until last week when the

14:33when t

14:33when cloud dashboards came out

14:35is

14:36is it's one thing to vibe code it, but then how do you take the vibe coded

14:39result, wrap it inside a twin

14:42a

14:42Twed

14:43or Twibics I guess.

14:44Twib properly and like

14:45to do that.

14:45And it looks

14:46that Claude might have done that for

14:48those.

14:48Like

14:49do you know

14:49like so

14:50Yeah.

14:55And it's probably just a skill, right?

14:56It's probably just a skill that developed and they said, here's our

14:59Here's our like workbook architecture

15:02to send us stuff for us to be able to have this.

15:04We have something called a Viz

15:05extension.

15:05It supports these attributes.

15:07So when you build what you build,

15:09Just use this Viz extension capability in order to be able to articulate the whole dashboard.

15:15And boom, it just comes across as a Viz

15:17Extension package in a workbook, which is

15:19Which has always been possible, right?

15:21So

15:21it's it's sort of individual

15:23it's VizQL data service

15:24API, it's

15:25Viz extensions, it's I'm sure

15:28a bunch of other things in the RAST

15:29API to be able to publish the workbook and

15:31And get it sent up, right?

15:33And then you have this information architecture, which is the way projects

15:37should have been called folders, but projects are set up and then there's

15:40Secure it to certain people, so you pick all that stuff up as well.

15:43Right.

15:43Yeah, exactly.

15:44Exactly.

15:45The the MCP, I'm sure, as hell, it grabs the data in the first place, which would respect row level security.

15:50The only

15:51ones that then

15:52wraps it back into

15:54It's a bunch of bets

15:56paying off, right?

15:57Paying off, yeah, that's a good way we think about it.

15:59Do you know what I mean?

16:02Each of them haven't landed to

16:04to the everyday developer.

16:05Does that make sense, right?

16:07A hundred percent.

16:08This is

16:08this is sort of like

16:10IK, no.

16:11This is what we were waiting for, if that makes sense.

16:13But sorry to cut you off.

16:14I know you were about to go.

16:14Yeah, no, that's what I meant by

16:16it was it

16:16tactical or strategic.

16:17If we use those words too much, did they have all these things then locked into

16:20it?

16:20It was the right time for them.

16:22Or was it strategic and they were building out ahead and now finally their time's come?

16:26I don't know that we'll ever

16:27be able to answer that question.

16:30Yeah, yeah, yeah.

16:31Or it doesn't matter.

16:32We planned it all along.

16:34Right.

16:34But you're right.

16:35I think what it is is you're probably developing the mother of all

16:38this

16:38extensions.

16:39That's what it is, effectively.

16:40Right?

16:41Which we see people all the time.

16:43Like

16:43on, you know, LinkedIn's full of people.

16:45FID coding

16:46Viz extensions

16:47really

16:47quick.

16:48So this is just uh

16:49uh

16:49a much easier, more managed way to do that, which is pretty slick.

16:53Right.

16:53Oh, and I guess the other thing they've must have added, obvious

16:56Lee

16:57is

16:58effectively a web server to render them.

17:00Because the other pain to this point was even if you vodcoded an extension

17:04Unless you got it approved by Tableau

17:07as a partner and the partner

17:08like you had to host it somewhere else.

17:10Clearly

17:11they're doing

17:12down that wall fast, haven't they?

17:14Yeah.

17:14It took 'em I don't know how many years to

17:16do that, but then now it's just like

17:17what

17:18Knock it

17:18out.

17:19Right.

17:19Well I guess they'd ha I guess they've always had

17:22I don't know, there might be some native

17:23I haven't thought this out till

17:24right now, but maybe the fact that it comes in from Claude

17:27gives it some level of

17:29it being unlikely to call

17:32Like something

17:33I don't know.

17:34I don't know.

17:34You can build some pretty creative things with web browsers.

17:37You know what I mean?

17:38It's it's

17:39would I rather trust something that someone has vibe coded?

17:43to be injected into my platform if that's

17:46or would I rather trust a developer

17:49who

17:50stands to lose a lot of reputational

17:53like

17:53cachet if they get something like that wrong

17:56And

17:57I'm just gonna sort of simplify this.

17:59For a long time, Salesforce and Tableau

18:01have thought the latter

18:03was

18:03uh

18:04less likely if that makes sense.

18:06They could

18:06because they kind of

18:07they didn't they didn't

18:09they didn't sort of bring the walls down and enable developers to really go hand.

18:13But they think that, yeah, Claude's not going to have those same issues.

18:16And I think that's actually probably going to be like a huge conversation

18:19once we get into it.

18:20I I assume also

18:21Tableau

18:22have probably done something to

18:24to kind of firewall them.

18:25So the personal space to me is a really good solution to that

18:28because

18:29it kind of puts it

18:30into a context where

18:31only that user can really

18:33have

18:34so it will respect that person's individual access to the existing data sources, any

18:38row level security it

18:39kind of gets carried over because it's in that personal space.

18:42But then the question I have and I don't know anything about

18:45is

18:45well, how do you then promote from that personal space to s to to like projects or shared assets?

18:49Or

18:50collection even, right?

18:51And I think collections are probably gonna get the biggest

18:53upgrade ever

18:54because I don't know if you saw the um the Dreamforce keynote, but it's like someone remembered that we already have something that allows you to group assets across different projects.

19:03Ah, collections.

19:05Right.

19:05Let's really upgrade these so that, you know, this whole project folder thing that no one uses properly and we can't consistently use.

19:13can actually just be dumped and collections become this more of a social sharing thing where people are vibe coding stuff they're adding it to collections and people are promoting it in and out and that way you get this more organic

19:24sort of ecosystem, which is why Tableau

19:26knowledge is so important.

19:28It's almost like they needed to build the back

19:30end algorithm for their social platform.

19:32And Tableau

19:33knowledge is essentially that it's sort of stitching things together

19:36so that the AI isn't useless when you start to then ask things in studio

19:41or inside of the platform.

19:42It has that context that people are sort of building in.

19:45Right.

19:46Right.

19:46So already

19:47knowledge should enable

19:49so

19:49new.

19:49Doing

19:50I've started playing with it and thinking about

19:51a video and stuff, but I haven't

19:53and working through

19:54Like if you have a sales process that you want to define, now you don't have to convert that to an MD file and remember to call it every time, or you can upload the doc

20:02as is, right?

20:03But

20:04Again, something I didn't think about to now, so I don't know if they've thought about or thinking on the fly, but my whole like git GitHub for dashboards thing, maybe that slides into knowledge as well, might be the logical place for it, because it's knowledge around

20:16And then you get lineage and history.

20:18I don't mean like data lineage, but like dashboard lineage, like how if it was vodka, what previous versions look like.

20:25Maybe that's a logical place.

20:27Yeah, yeah.

20:27I do think though, to go back to the Databricks point and then Snowflake as well,

20:31throw them in there.

20:32This is a real battle now.

20:33It's almost like the race between BI

20:35vendors is entirely different.

20:37Before I think they were very much going up against each other

20:39I think they can just safely drop those the

20:42drop those sort of

20:43focuses

20:44and really just squarely focus at this sort of

20:47build

20:47out of what is the new

20:49I'm gonna call it landscape or canvas upon which you build stuff on, right?

20:53So Databricks have this view that hey you already have your

20:56your

20:56data, you already have your AI workloads over here.

20:58Why not also just add your BI workloads on top of where your data already lives?

21:02They love that.

21:03Good consumption.

21:04On the other side, you've also got, you know, fabric.

21:07We've talked about Fabric

21:08apps has already

21:08just

21:09been announced.

21:10They're trying to protect the Azure

21:12and Fabric moat, right?

21:13So, you know, Power

21:14BI has really only ever been this moat for these

21:17fabric and Azure

21:18sort of value props that Microsoft likes to put into an organization.

21:22And then Snowflake is kind of out there on its own, being I guess the most agnostic of the databases

21:27to

21:28everything else.

21:29So AWS, Snowflake, wherever you have your data, Snowflake seems to be like the

21:33the place to to put stuff in.

21:35So it it's really interesting that you have database vendors versus BI vendors versus

21:40I guess you call it the the the claws.

21:42I don't know what we call them.

21:43I'm not going to call them a superscaler but

21:45Claude is skimming very much.

21:46It wants to be that aggregator, maybe let's call it that, the aggregator

21:50and the interface to all of these things, right?

21:52And that that is super interesting.

21:54That is super interesting.

21:56Yeah, like

21:57they

21:58probably are thinking about themselves as like

22:00the

22:01ubiquitous UI

22:02layer

22:03of everything.

22:04Which is ironic because they almost don't have a

22:06Uh.

22:06But

22:07you know, like again, the creation

22:09almost like uh

22:10almost like an operating system for creation.

22:13Yes.

22:13So

22:14exactly.

22:14Why did they take on the messiness of governments and

22:18I mean

22:18other than things like

22:19stopping injection in the code and stuff dude, there's some things they'll have to do, but like most of it

22:24in your point, I do think

22:26Yeah, the BI vendors have to most worry about now

22:30for sure is these cloud data warehouses.

22:34I don't know.

22:34Well

22:35will

22:35products like Looker maybe

22:36become

22:37well

22:37may

22:37not looker, we shouldn't think about that, but now our Google also locked into a bunch of stuff.

22:41I haven't had enough time to think about that

22:43because they

22:44always felt like

22:45At least in the BI space

22:46third class citizen kind of thing, but maybe, you know, some companies would be like, we use them, how

22:50can you say that?

22:51But for the most part.

22:52But

22:52but

22:53but I do wonder about all these startup

22:55BI vendors or

22:57At least in theory, I think we'd have a harder time than Tableau

23:00and like the likes of Tableau

23:01if they get this right because

23:03the hardest time of all.

23:04Yeah.

23:04Yeah, I was talking to an exec

23:06from a company just yesterday and I think

23:08that

23:09the biggest challenge is very

23:12much

23:12how do you

23:13how do you

23:14how do you sell your products?

23:15What what conversation do you have with customers?

23:18When they are largely waiting to see what happens before they make any investment, right?

23:23And you're out of the way.

23:24That's the biggest thing.

23:25Right.

23:25Yeah.

23:25Yeah.

23:26But people are just stalling

23:27because

23:28stalling, dare I say it, is paying up.

23:31Right.

23:31People are just building right into the space.

23:33So

23:34literally if you

23:34If you're thinking of a renewal like eight months ago and you think of a renewal today, the landscape is entirely different.

23:40And that's

23:41one one cycle and

23:42These contracts go for two to three years at a time.

23:45So it's not like they renew yearly.

23:47So

23:47for a lot of companies who are coming up

23:49renewal right now, it's like, oh, we'll extend for a year

23:51just to see where things go, right?

23:53That's sort of the same.

23:55And then long term

23:56though, there's just

23:57gonna be this massive influx, massive influx of

24:00change.

24:01We should have another session on like

24:04this whole build.

24:06uh

24:06and curation culture, right?

24:07Because I think that now becomes the prominent skill.

24:10We've we've been talking about, okay, what is the most important skill set?

24:13Um

24:14let's get into it now actually.

24:15What is the most important skill set for the future?

24:17I'm gonna say

24:18BI developer

24:19and inverted communism.

24:21I think we can safely have like

24:22hold

24:23cold of morning.

24:25Right.

24:27It's no longer a really a thing.

24:29Yeah.

24:31I think most are scared, which I can understand why, especially because in the short term

24:36and we even feel this is a consulting company, things have gotten

24:39slow

24:40lately, but I can tell it's not we're not losing things.

24:42People are delaying everything.

24:44Right.

24:45So

24:45you feel it, right?

24:46So

24:46but I think

24:47the ones who are like

24:49The ones who have always felt like, hey, listen, like I'm a trusted advisor for them

24:53that can help them get the information they need the right way.

24:57should be happy because it's been hard to do that because you try to be that and what you're actually doing is you're fiddling with settings inside a dashboard.

25:05You know, like in the report, life cycles

25:07never really went away.

25:08It's done a little shorter.

25:09It's not gone away like Kablo

25:11would like to think it has, right?

25:12So you spend all your time building.

25:14So

25:14now

25:15you should be able to be

25:16I

25:17talked about

25:17any

25:17cockroof

25:18about this about before we used to be like twenty

25:20percent thinking, eighty percent building.

25:22Now we should be able to be like eighty, ninety

25:24percent thinking

25:25and ten, twenty

25:26percent building, right?

25:27So now we've got

25:28this cadre

25:29of agents who can actually build the stuff for us once we describe it to them.

25:33Right.

25:34So

25:35the

25:35the role

25:36should be a lot should be a lot more exciting for the people who always sell themselves as business consultants or business analyst

25:42types anyway, I think.

25:44It should be a really good time.

25:46No

25:46I do feel bad for all the people who've

25:49and there's a lot of them who went, no, I know Tableau

25:52or Power BI or whatever, I build dashboards, other people can worry about what goes in them.

25:55I'm like, yikes.

25:56Do you know what I mean?

25:57Like that's gonna be rough.

25:58Do you know what I mean?

25:59Because

26:00Correct.

26:00Yeah, absolutely.

26:01Absolutely.

26:02Um

26:03it's interesting.

26:04A

26:04a year and a half ago I was asked to

26:07For work at

26:08endpoint, I was asked to go to

26:10um

26:10like a

26:11an off

26:12site for a company

26:14and they asked me to talk about the future of Tableau

26:17for the next two years.

26:18It's one of these things where I was like, oh

26:19God, like

26:20jeez

26:20Like, uh

26:21how

26:22how on earth do I

26:23do I do that?

26:24Now I'm quite happy with myself

26:26because I ha

26:27I

26:27at the time I was very adamant that

26:30A skill that I think already existed, but wasn't sort of respected enough was this sort of role of a curator, you know, who is

26:37not necessarily always building, but making sure things are done right.

26:41I I kind of use the analogy of um

26:43uh

26:43neighborhood police officer, you know, patrolling the streets, getting involved when they need to be, but generally just making observations and and then just sort of making sure the community is a harmonious place that people want to be in.

26:55Right.

26:55And then there's another side of that, which is uh getting your hands dirty, right?

26:59And and actually building the things that people are going to use.

27:01So

27:01identifying problems where

27:03where things should go.

27:04And I I I very much sort of leaned on this curator thing.

27:06Now, in that in that session, I said

27:09that it would take roughly two to three years to get to this place.

27:12And then

27:13it's taken like one year.

27:15Well,

27:16on an infinite timescale.

27:17I got it woefully wrong because anything is possible in an infinite

27:20timescale.

27:21But

27:22I

27:23I am

27:24I am

27:25more certain

27:26than ever that if you're in this space

27:29Whilst a lot of your time was taken up with what I think is quite mundane, which is just making things beautiful, focusing on UX in a BI tool where it was never really that good.

27:38So

27:38like it was it was it was kind of a clunky experience.

27:40I'm famous for hating hacks, right?

27:42Like this is

27:43like

27:44I'm so happy to be done with that

27:46if that makes sense.

27:47And I think it should be rewarding to now really say, actually

27:51I can

27:51move my skill set

27:53a little bit further back, not too much further back, but actually get close to the data, maybe pick up some AE

27:58sort of skills, analytics

28:00engineering skills, right?

28:01How do I really

28:02understand

28:03what we need as a business, where it should live and how it should work.

28:07And then how do I build that ecosystem so that my analysts, and I say my analysts, because I'm now the curator,

28:13My analysts who are the business people

28:16can just go and build the things they need.

28:18And then all I'm doing is making sure they're doing it and using the best practice.

28:22I'm supplying them with the assets.

28:23I'm s I'm teaching them how to prompt this stuff.

28:26And then lastly, I'm also

28:28Joining the dots

28:29where

28:30big solutions need

28:32a concerted effort

28:33because those solutions still need

28:36people like us.

28:37So

28:37you might have

28:38Three or four dashboards in your company that everyone uses, right?

28:42Everyone

28:42it is the universal

28:43tried and tested thing.

28:44Those are still going to be built in a BI solution, generally speaking.

28:48But the offshoots from that, let's say someone wants to take a, you know,

28:52finance view of that, one

28:53someone wants to take a HR view off that, those it

28:55is

28:56if

28:56if the data models are good,

28:58There's no reason why those can't be replicated off those trusted data sources.

29:02And you can even tie them out and check them out and get Claude to do the validation for you between them.

29:06So

29:07I'd be really excited about that.

29:08I think that is the sort of future of where we go.

29:11Um

29:11and

29:12Another thing I'm gonna say here is

29:14keeping things simple

29:17is actually a huge skill.

29:20It's gonna be so easy to vibe code these absolute monstrosities, but it's gonna be so hard to bring your design skills, your UX skills

29:29And your experience of how the business works to be able to make these things so simple that they get used time and time and time again.

29:36Yeah.

29:37Right.

29:37And

29:38in

29:38a token efficient way, as you say.

29:39Yeah.

29:40Yeah, I think

29:41it felt like

29:42Say

29:44five to ten

29:45years ago, like

29:46analytic

29:47COEs

29:48with

29:48centers of excellence or centers of enablement.

29:51We're big things in companies.

29:53Yep.

29:53And they and they seem to die out.

29:55I don't know why.

29:55But they don't seem to be a big thing anymore.

29:57People don't go to like internal

29:59user group like there's exceptions to all these things of course but for the most part

30:03they might make a huge comeback

30:05do you mean

30:05and now they're a lot more tangible

30:07do you know

30:08because

30:09You know, it used to be such a fight to get people

30:11to do

30:12use a corporate template, say for a deskboard.

30:14I'm not sure that's that important anyway.

30:16Like I

30:16I've never been that's as important as other people, but still

30:18take that example, right?

30:19But certainly

30:20certain visual best practices

30:21Right.

30:22Like

30:22now you could be like, this is how when you vibe

30:25code something, you make sure that those in the quad world

30:28skill files are used.

30:30Like this is where they are.

30:31This is where you gotta make sure.

30:32So you don't even have to think about it.

30:33Like it's gonna advance.

30:34It's going to apply the template for you.

30:35So you don't have to learn how to build the template

30:37or fit into it or replace things in the template, which is great.

30:40But you still need to know where to go to get it.

30:42Do

30:43like

30:43there could be classes on

30:44because

30:45the problem is

30:46if you look

30:47back t

30:47the first ten years of Tableau

30:49before that ten

30:50years

30:50I think the problem was

30:52everyone thought, ooh,

30:53Tableau

30:53's so easy.

30:54I don't need any training.

30:55I'm just gonna do

30:56and that's where you got these crazy LODs,

30:58dumb calculations and things.

31:00And then at some point people like, I better learn like it's still easier than before, but I should still learn it.

31:06We're at the dumb stage right now.

31:07Like people keep prompting their way to dashboards instead of going, look, before I even start.

31:12I'm gonna describe

31:13exactly

31:14what my dashboard is, right?

31:16And I'm gonna have all these associated skill files around visual best practices and I think

31:20corporate templates or whatever, right?

31:22Color guides, you know, to make sure that

31:25Contrast

31:25is right and colorblind people will be fine, but

31:27like

31:28and then

31:29I think the answer to all this is gonna be like

31:32single prompting these things.

31:34Because otherwise it's a mess.

31:36Do you know what I mean?

31:37That's where you get into bad data, I think.

31:39Do you know what I mean?

31:39Like you single

31:40prompt it and then you could quickly have

31:43a single

31:44prompt UA behind it to check numbers.

31:46So

31:46all this the numbers aren't right.

31:47Like it's easily solved too.

31:49These things are easily solved

31:50if people

31:51step back and try to learn how to do it, as opposed to thinking

31:56this is

31:56like

31:57It's

31:58it

31:58it's a chcause I guess they started as chatbots, right?

32:01But don't build a dashboard like it's a chatbot.

32:03Like build it like it's a developer

32:05and you're the architect, right?

32:07So you come up or the product manager, you come up with the spec.

32:10And go, okay, now you go build it.

32:12Because like you're good at writing SQL and all that kind of stuff, but I don't want you

32:15trying to figure out

32:17interpret what revenue is, right?

32:18Like

32:19um, like you know, for your finance dashboard, we want to use

32:22Invoice date, right?

32:23And for sales maybe we want to use sales data.

32:25That could all be in skill files, right?

32:26Or it could be

32:28right

32:28and then

32:28so

32:29I don't want

32:30like

32:31a very simple example I'll give you of how of how these things try to do it and how and and why semantic models

32:37and skill files

32:38and things are so important in this world is real

32:39life client, transportation clients.

32:42I've used the example before, but where they ha

32:44one

32:44it's no cost per mile.

32:46And in their model, in their tableau model, they had

32:48CPM.

32:49Right.

32:49And for anyone who knows

32:51like

32:51business like that's

32:53CPM's most famous for cost per 1000

32:55impressions, right?

32:56And merfitting.

32:57So I

32:57asked this thing

32:58cost per mile, cloud goes for a while

33:00through the MCP, the Tableau

33:02MCP, works at it for a while, comes back, creates its own calculation

33:06Which was close but wrong.

33:07And everyone's like, see, it's wrong.

33:08But I'm like, why didn't you CPM?

33:11And it goes, I didn't think what that's what it was, right?

33:13You go back into the Tableau

33:15Semantic model, say

33:16CPM is cost per mile.

33:18Anytime anyone asks a little

33:19cost

33:19for mile.

33:20Yep.

33:20This is the

33:21and

33:21then you ask

33:22it uses it.

33:23I'm like, well, why did you use it now?

33:24It's like, well, because this

33:25because the model told me right here there's a description to use.

33:27Like

33:28you can test these things to know they're smart

33:30to do.

33:30Yeah.

33:31Or sorry, did they

33:32or sorry.

33:32But

33:33the problem with them is is

33:34they're

33:35Yeah, I don't know if it's a problem, but they're overly keen analysts

33:38who aren't

33:39true

33:39who aren't that great at coming back and asking you for clarification.

33:42They're getting better at this though.

33:43They are getting better with you mean one, two, three.

33:45They're getting better at it.

33:47But

33:47But you

33:48they

33:48tend

33:48on pleasing you, right?

33:50That's that's sort of what what they want.

33:51Yeah, they want to impress

33:53new employees are like too.

33:57humans

33:57or that's what their training site is at least, right?

34:00Like

34:00no fear, but they want to please, right?

34:02So

34:02yeah.

34:03So I just keep thinking, if you instead, when you do dashboard development, just hand it the complete

34:08spec

34:09right and then the other thing that becomes easy is if you want to make a chain

34:13like

34:13you just change that file

34:16Yeah.

34:16And then reprompt the single prompt and it's gonna build it right

34:20because

34:20you can just change it and

34:22there is a best practice evolving at the moment

34:24where

34:25instead of

34:26iterating through prompts, it's okay to go back to the initial prompt, change that, and resend that

34:34because that's actually more token efficient, especially if you

34:37go three or four

34:38five prompts down.

34:39What it's actually doing is sending the whole conversation again just to make a tiny little change.

34:43When it could just be

34:44sufficient just starting from scratch with the correct context.

34:48And you

34:48take a lot less sort of deviations.

34:50So

34:51great.

34:51Yeah.

34:52So we had this conversation offline about how I saw a a software developer build schemes

34:56doing this.

34:57And

34:58In it's so good he

34:59goes because he's been doing it for a long time, like this single prompt game development.

35:02There's

35:03in that world what he also

35:04does is he use frontier models to test his requirements.

35:08And then builds with cheaper models because who cares?

35:10Right.

35:11So

35:11maybe in the dashboard development world, it's like

35:14we use our

35:15frontier

35:15models

35:16to

35:20Well he was even using an even cheaper one

35:22sometimes, right?

35:23Because it's like

35:24the requirements are so good.

35:25It doesn't matter.

35:26Exactly

35:26true.

35:27And

35:27if it takes forever, who cares

35:29And

35:29it can test on the cheaper model because actually the goal is very clearly stated.

35:33So

35:36then

35:36but then he talked about

35:37he wanted to change his game and change one of the tenants of it

35:40like

35:41one of the core kind of tenants.

35:42Hey, like when you hit a

35:43I don't know, when you jump over a log, you can't fall directly on your face.

35:47Your hands

35:47always catch

35:48you or something, right?

35:49And then he changes it to go, nah, let the person fall on their face.

35:52And

35:52And his Frontier Model 18, which runs in cloud, right, whatever cloud model

35:56at the time, would say, before I do this, you gave me

36:00a requirement not to do it anymore.

36:01He goes, cut

36:02change my mind, put it

36:03Like if

36:04because the way he does it, it always knows the reasons he

36:07did things in the past, right?

36:08Yeah.

36:09Right?

36:10Like

36:10That's the context you need.

36:11And that's where Tableau knowledge should really actually

36:14I hope they get it right.

36:15Because the thing is if it ends up being

36:18this thing where you wish someone didn't add something to it and then you end up having to do the surgery of trying to figure out

36:25what in the context

36:26is it keep on picking up and having this behavior.

36:29that becomes like a right

36:30pain in the neck because

36:31you have so much stuff that can be uploaded there that the the

36:34the

36:34the etymology is a ro

36:36wrong word, but the original sort of cause of the outcome becomes to start

36:41st

36:41starts to become opaque.

36:43almost to the

36:43to the to the curator who's trying to create this experience like

36:47I have this thing in my design md file but for some reason it keeps going around and doing

36:52this instead.

36:53Why is that even though I keep telling it not to

36:55And

36:55you get this sort of layer cake of

36:57a prompt here

36:58says

36:58do this

36:58and another one says

36:59don't do that and then another one says

37:01do that and another one says

37:02don't do that and it's just like

37:04It's getting crazy.

37:05Now you're just burning tokens

37:06steering

37:07the the ship.

37:08So

37:08right, right.

37:09A hundred percent.

37:10It's gonna be interesting.

37:11It's not gonna be smooth for sure.

37:13But

37:13again, these are all opportunities for this role to

37:16Help clarify run that down.

37:18Yeah.

37:18Yeah.

37:19Yeah.

37:19Yeah.

37:19Your curator

37:20role.

37:21For sure.

37:21I like that.

37:22Yeah, yeah.

37:22Because you gotta curate all these.

37:24Because we're gonna run into a lot of I can tell you

37:26Like

37:27overly

37:28pedantic people doing things like

37:30a dashboard should never have more than four charts on it.

37:33And they're gonna put it in a design

37:34MD file and someone's gonna have a good use case for six and they're gonna be fighting the living

37:39What is a dashboard?

37:40What is a dashboard?

37:41What is a sheet?

37:41What is a met

37:42like

37:42is a metric a chart?

37:43Like what if I can have twenty

37:45metrics on a page?

37:45It makes total sense.

37:47Like all these rules just go out of the window.

37:49And and and actually there is a product mindset that people haven't had and a curator is going to have

37:55to have, which is

37:56I'm looking after an analytics

37:58product.

37:59On it, people are using it to execute different things.

38:02The experience of that product becomes your sort of duty.

38:06And

38:06What that means is you have to appreciate technical debt will exist in your in your solution.

38:11You also have to appreciate that the frontier technologies will keep adapting and evolving.

38:16So just 'cause you you planned it and designed it today doesn't mean in six months we need to ship it.

38:20That's

38:20where the industry is going to be.

38:21So you have to almost be thinking ahead and going, how do I avoid vendor lock-in?

38:26How do I avoid having all of these things?

38:27Like Tableau

38:28knowledge to me is kind of like a mini red flag because I'm like

38:31If I go in there and upload all these things, how do I get them out, right?

38:34You know, I'd actually

38:35I'd actually want to be able to very quickly do that

38:38because they're actually quite useful in other contexts.

38:41But

38:41You know, I don't want to play into that game again.

38:43So there's all these

38:45small things that I think the last twenty years have taught us.

38:48A product owner is going to take those into account, knowing full well that every six months

38:53There'll be as much innovation in six months as there has been in five years of BI development.

39:03You think about it

39:04because

39:04It's going to have to expose everything through an MCP or otherwise, some kind of

39:10interface

39:11to

39:11the agent for it to be useful.

39:13And if it can do that, you're going to easily be able to ask the agent

39:16To download or take it, transform it and put it in somebody else's.

39:20Or reverse engineer.

39:21Have you heard of this

39:22GitHub repo, which which re-engineers

39:25products?

39:25I

39:25I keep meaning to do a v a video to see if we can re-engineer Tableau

39:29Desktop interface, but yeah.

39:30Uh it's it's a pretty interesting GitHub project.

39:33Didn't someone just

39:34I should talk at the I didn't do enough research on this, but someone effectively I think didn't they in some a short period of time

39:40like a month?

39:41completely redid

39:43like Adobe

39:44's whole suite.

39:44Yeah.

39:45Yeah.

39:45Yeah.

39:46Yeah.

39:46Yeah.

39:46Yeah.

39:47Yeah.

39:47REV yeah.

39:48Yep.

39:48Yeah.

39:49So

39:50this is I

39:51I was gonna

39:51I was gonna try and end this on a

39:53Very interesting night which is

39:55I've always had this observation with Tableau

39:58that it is

39:59Tableau

40:00as a company, like

40:01just I'm going to put Salesforce to one side.

40:03Tableau the product, Tableau the brand.

40:06almost exactly follows the Adobe

40:09model, but a decade behind.

40:12Like

40:13everything about it.

40:14Like you had the PDF, you had Tableau

40:16Reader, like literally a decade later, right?

40:19You had the core product desktop, Adobe

40:21It's Photoshop.

40:22And then you have all these other sort of little things that have popped up around it.

40:25It's exactly the same.

40:26Then you had Tableau

40:28Public, which is basically the Adobe sort of

40:30They have like a light version that creators use and stuff like that.

40:33Which is basic

40:34basically the same thing, right?

40:36And then now with this sort of, you know, move

40:39of hosting and becoming almost

40:41this sort of platform.

40:42It's doing what Adobe did a while back, which is essentially give creators like a place within the ecosystem.

40:47to share and create assets, almost become like the the portal.

40:51The difference is

40:51Adobe had ultimate control

40:53of the design and creativity space, right?

40:56Where

40:56Tableau

40:57haven't quite achieved that level of dominance.

40:59And so it doesn't quite work

41:01the same way.

41:01But anyway, it's just an observation I've always

41:03got to do.

41:03Yeah, yeah, yeah.

41:04No, it isn't

41:05spacey following Adobe, but a decade behind in terms of

41:08Including renaming desktop studio.

41:10Yeah.

41:10Now I've said it, you're not gonna unsay it.

41:12No.

41:13Anyway, Kirk, it's been

41:15it's been fantastic.

41:15Thanks again for for joining me.

41:17Yeah, it's been

41:18great

41:18for the chat.

41:19We'll probably have another one of these in a month or so.

41:21I think we're gonna have to just to keep up and see what's going on.

41:25We'll call it like a AI check-in.

41:27I don't know.

41:27We have to come up with a better name rather than emergency podcast.

41:30So

41:30I like it.

41:31And then we can look back and see how wrong we inevitably were.

41:33How well we were, yeah, we made some calls.

41:35Maybe I'll do that.

41:36I'll I'll do

41:36like a running list of here are all the things we said w w

41:39would happen and every video we'll just do like a a a checklist and say, okay, last time you said this, okay, it hasn't changed yet.

41:45Passeng thing.

41:46Yeah.

41:47Alas

41:47and first

41:48two.

41:48Okay.