Claude's New Code Interpreter Demo
Key Points
- Claude’s newest “code interpreter” lets users create and edit Excel sheets, PowerPoint decks, Word docs, and PDFs directly within its web and desktop interfaces, aiming to streamline core office workflows.
- The video demo features a live, screen‑shared session with Rod, who walks through real‑world prompts and workflows across Claude, OpenAI, and Perplexity to illustrate the feature in action.
- Rod, an entrepreneur who has been leveraging AI for risk intelligence and business building for over two years, emphasizes that virtually all of his weekly tasks involve those four file formats, highlighting the tool’s relevance.
- The discussion underscores Anthropic’s strategic focus on tackling the most common productivity pain points, positioning Claude’s new capabilities as a potentially transformative experiment for everyday office work.
Sections
- Live Walkthrough of Claude’s Multi‑File AI - The hosts demo Claude’s new capabilities to analyze and generate content from Excel, PDFs, and PowerPoint, while entrepreneur Rod discusses his AI‑driven workflows and prompting strategies.
- Claude's Spreadsheet Agent Impresses - The speaker tests Claude's new agent mode on generated data, discovering it handles formulas and dynamic spreadsheets far better than expected, surpassing OpenAI's earlier agent features.
- AI‑Generated PowerPoint Design Review - The speakers evaluate a slide produced by an LLM, highlighting its superior spacing, balance, and typographic hierarchy, and noting it surpasses typical PowerPoint designs.
- Using LLMs for Real‑Time Stock Analysis - The speakers discuss leveraging LLMs to create unified prompts and data for rapid valuation, exemplified by a quick analysis of Oracle’s earnings surprise and AI guidance.
- LLM Contrast: Excel Model Generation - The speaker compares two language models’ responses to a request for an Excel financial model, highlighting differences in the generated worksheets, documentation, and overall presentation.
- Comparing Claude vs ChatGPT Agent - The speaker contrasts Claude and ChatGPT's agent modes in speed, response style, and financial analysis accuracy, noting a request to format the output as an Excel worksheet.
- Design Review of AI-Generated Slides - The speakers outline a live exercise, then scrutinize a poorly designed AI‑created PowerPoint slide, contrasting its past impressiveness with current flaws, and note Perplexity’s suggestion to assign the agent only to preliminary analysis.
- Collaborative Fun Pivot Table Brainstorm - A team debates adopting playful, consumer‑oriented pivot‑table themes like “movie night” versus more corporate options, leveraging LLMs and Perplexity to dynamically generate and share ideas.
- Separating Research from Prompting - The speaker explains that using Perplexity for data gathering before constructing a detailed prompt for Claude—particularly when incorporating internal company data—produces more effective results, as illustrated by a movie‑dataset example.
- Awe at AI's Rapid Automation - The speakers applaud Perplexity’s ability to instantly generate Python code, citations, and Excel pivot tables, marveling at how swiftly modern LLMs transform labor‑intensive tasks into effortless, on‑the‑fly solutions.
- Excited Review of Auto‑Analytics Model - The participants enthusiastically examine a new analytics tool that auto‑generates pivot tables, heat maps, and pattern analyses in Excel, speculating on its broader impact on dynamic data products.
- Navigating Excel Pivot Features - The speakers troubleshoot finding and using pivot tables, heat‑mapping, and related analysis tools in Excel, while reflecting on how AI assistance shapes their workflow decisions.
- Informal Closing and Future Plans - A brief, relaxed wrap‑up where the participants thank each other for a last‑minute chat, make a light‑hearted joke, and express enthusiasm about meeting again.
Full Transcript
# Claude's New Code Interpreter Demo **Source:** [https://www.youtube.com/watch?v=JOUqXy_Ifmc](https://www.youtube.com/watch?v=JOUqXy_Ifmc) **Duration:** 00:40:49 ## Summary - Claude’s newest “code interpreter” lets users create and edit Excel sheets, PowerPoint decks, Word docs, and PDFs directly within its web and desktop interfaces, aiming to streamline core office workflows. - The video demo features a live, screen‑shared session with Rod, who walks through real‑world prompts and workflows across Claude, OpenAI, and Perplexity to illustrate the feature in action. - Rod, an entrepreneur who has been leveraging AI for risk intelligence and business building for over two years, emphasizes that virtually all of his weekly tasks involve those four file formats, highlighting the tool’s relevance. - The discussion underscores Anthropic’s strategic focus on tackling the most common productivity pain points, positioning Claude’s new capabilities as a potentially transformative experiment for everyday office work. ## Sections - [00:00:00](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=0s) **Live Walkthrough of Claude’s Multi‑File AI** - The hosts demo Claude’s new capabilities to analyze and generate content from Excel, PDFs, and PowerPoint, while entrepreneur Rod discusses his AI‑driven workflows and prompting strategies. - [00:03:15](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=195s) **Claude's Spreadsheet Agent Impresses** - The speaker tests Claude's new agent mode on generated data, discovering it handles formulas and dynamic spreadsheets far better than expected, surpassing OpenAI's earlier agent features. - [00:06:20](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=380s) **AI‑Generated PowerPoint Design Review** - The speakers evaluate a slide produced by an LLM, highlighting its superior spacing, balance, and typographic hierarchy, and noting it surpasses typical PowerPoint designs. - [00:09:27](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=567s) **Using LLMs for Real‑Time Stock Analysis** - The speakers discuss leveraging LLMs to create unified prompts and data for rapid valuation, exemplified by a quick analysis of Oracle’s earnings surprise and AI guidance. - [00:12:36](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=756s) **LLM Contrast: Excel Model Generation** - The speaker compares two language models’ responses to a request for an Excel financial model, highlighting differences in the generated worksheets, documentation, and overall presentation. - [00:16:03](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=963s) **Comparing Claude vs ChatGPT Agent** - The speaker contrasts Claude and ChatGPT's agent modes in speed, response style, and financial analysis accuracy, noting a request to format the output as an Excel worksheet. - [00:19:36](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=1176s) **Design Review of AI-Generated Slides** - The speakers outline a live exercise, then scrutinize a poorly designed AI‑created PowerPoint slide, contrasting its past impressiveness with current flaws, and note Perplexity’s suggestion to assign the agent only to preliminary analysis. - [00:22:48](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=1368s) **Collaborative Fun Pivot Table Brainstorm** - A team debates adopting playful, consumer‑oriented pivot‑table themes like “movie night” versus more corporate options, leveraging LLMs and Perplexity to dynamically generate and share ideas. - [00:27:15](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=1635s) **Separating Research from Prompting** - The speaker explains that using Perplexity for data gathering before constructing a detailed prompt for Claude—particularly when incorporating internal company data—produces more effective results, as illustrated by a movie‑dataset example. - [00:30:32](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=1832s) **Awe at AI's Rapid Automation** - The speakers applaud Perplexity’s ability to instantly generate Python code, citations, and Excel pivot tables, marveling at how swiftly modern LLMs transform labor‑intensive tasks into effortless, on‑the‑fly solutions. - [00:33:39](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=2019s) **Excited Review of Auto‑Analytics Model** - The participants enthusiastically examine a new analytics tool that auto‑generates pivot tables, heat maps, and pattern analyses in Excel, speculating on its broader impact on dynamic data products. - [00:37:24](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=2244s) **Navigating Excel Pivot Features** - The speakers troubleshoot finding and using pivot tables, heat‑mapping, and related analysis tools in Excel, while reflecting on how AI assistance shapes their workflow decisions. - [00:40:33](https://www.youtube.com/watch?v=JOUqXy_Ifmc&t=2433s) **Informal Closing and Future Plans** - A brief, relaxed wrap‑up where the participants thank each other for a last‑minute chat, make a light‑hearted joke, and express enthusiasm about meeting again. ## Full Transcript
Claude launched totally new capabilities
to revolutionize Excel, PDFs,
PowerPoints, all of that stuff. I
thought it would be boring for me to
just get on a video and talk about it.
And I think it's way more useful if you
actually see how it works in practice.
So, I got on the line with my friend
Rod, and we went through a bunch of
workflows across Claude, across OpenAI,
across Perplexity, and we broke down how
this new feature from Claude works for
multiple file formats, how you actually
use it, how you prompt it, lots of
screen sharing, lots of back and forth.
I hope you enjoy this one. All right,
I'm on here with Rod. Rod, do you want
to just introduce yourself and sort of
tell the audience a little bit about you
and what makes you interested in AI?
>> Yeah, sure. Um, thanks for having me on
first of all. Great to be here.
Basically, right now I'm an
entrepreneur. Uh, I started my own
company about two and a half years ago
and I b I've used AI extensively through
that process and it's not my first um
it's not my first experience with AI and
when I did use it before we used it
around things like third party risk
intelligence, right? where you're trying
to understand these very nuanced
scenarios that that ladder up to these
more complex situations that you need to
derisk. And then two and a half years
ago when JAGPT came out um I I had that
reckoning. I saw it. I said, "I've seen
this before. This looks awesome." And I
saw there was an opportunity for me to
actually take the skill sets that I've
been working on and bet on, you know,
the foundation that AI and was building.
And I've been doing it ever since. And
it's been really fun and exciting just
both sharing this stuff with other
people, learning more about it, and then
actually building my business around
this.
>> Yeah. No, that totally makes sense.
Well, we have a focus for today. I want
to get into some of that entrepreneurial
piece, but I think we'll get into it via
what launched from anthropic. So what
what Claude launched and uh you can add
on to this but like as far as I know the
way I would put it is yesterday Claude
launched a code interpreter which sounds
boring but isn't because it means it can
create and edit Excel spreadsheets
PowerPoint slide decks word documents
and PDFs directly in the web interface
in the desktop app. And the devil's in
the details there because if it does it
well, those four document formats are
this huge part of work. Like Rod, how
how much of the work that you're doing
touches Excel, PowerPoint, Word, or PDF?
One of those four.
>> Everything. Everything I do touches
what? All of those things at various
points in a week. Yeah. This is awesome.
from the fact that Anthropic had the
foresight to see that you know what is
it that people are having difficult time
with and saying like let's position our
product in a way to simplify that right
and I think that that's I mean I think
that just it makes for a very exciting
experiment that we're about
>> yeah no I think it is and I what I want
to do as part of this is I want to sort
of share my screen in a minute and show
you some of the tests that I've been
running today um and get your live
reaction to kind of like dig in because
what I did was I was like okay these
claims. We've heard these claims before.
I think if you remember, agent mode came
out, right? And Chad GPT was like, "Wow,
you know, you can do Excel and
spreadsheet." And I did a review on
that. And it was like kind of
forgettable. I didn't do a whole lot.
And so my assumption when Claude
launched this was to say they're trying
to be play me too. They're trying to
play catchup. That is not what I found.
I found that it was actually
surprisingly useful. So this is a uh
artificial company model. So I generated
a bunch of artificial data and I
basically fed that data to Claude and I
fed it to the agent model that OpenAI
has and I wanted to see specifically
>> how does it do at formulas because
formulas are one of those things that
has really beeled spreadsheets from LLMs
and I was astonished because you go
through here you have eight different
tabs. It starts with an executive
summary tab. I don't know how well you
can see that we're zooming in a little
bit. starts with an executive summary
tab where you can see the performance of
the division, the annual revenue, all of
this good stuff. If you click on a cell,
what you'll notice is you have these are
not hard-coded numbers. These are
formulas that it's actually using. And
then you say, well, where do these
formulas come from? You can start to
click through. This is the revenue data
tab that it has created, and it actually
has hardcoded these so it can reference
them. Uh, these are the parameters it's
using. So you it's listing its
assumptions. These are the scenario
planning uh features it's using. So what
are its multipliers for base,
optimistic, etc. This is the financial
analysis tab where you can actually see
how it breaks down the different
divisions in the company. So software,
hardware, consulting division. It's
calculating bonus calculations based on
how the company did. And again, these
are like dynamic, right? You see this is
equals another tab F24. So it's actually
like using that for calculations. I
won't pretend these are complex
calculations. I don't want to oversell
this, right? But the fact that we are
talking about an eight tab spreadsheet
that is correct that has formulas in it
that has nicely headlined like very
readable headers. That is a breakthrough
moment for me, Rod. Like I had a moment
where I was like I think that work is
going to change.
>> Yeah, thousand%. I mean I'm still, you
know, copying and pasting, you know,
tables from some of the LLMs or asking
them to create a CSV. Um I'm also
noticing that there's a documentation.
>> There is. Let's go look at it. Um so
this has a different section. So it
actually talks about like how it defines
these things like what is revenue
adjustment, what is IBITA, uh what are
the advanced features that it
implemented. And I really appreciate
this because like you can actually see
it documented its logic like it's using
a VLOOKUP here. It's using an if
statement here.
>> Honestly, when spreadsheets I didn't
documentation tasks.
>> Absolutely not. Who has time for that?
>> But that's the thing. It did this thing
in like 3 minutes and it came back and
it was like, "Holy crap." Like when I
was doing spreadsheets as a marketing
analyst, this this would have taken me
all day.
>> Oh, I spent plenty of days in
spreadsheets just trying to make a
VLOOKUP work.
>> Yeah.
>> Yeah. Exactly.
So,
>> that's cool.
>> I think the the wow factor for me was
something else. And then I want to show
you also um because we were talking
about powerpoints, right? And like how
does it do with powerpoints etc.?
>> Yeah.
>> I asked it to produce a PowerPoint
because do you do you recall and we may
look at some agent stuff because I did
ran agent through the same paces. Uh we
may do that a little bit later in the
call. This is what it did for
PowerPoint. And again it's not I don't
want to pretend that Figma designed this
and this is the most impressive thing.
Rod, like folks don't know how good you
are at design. and you kind of hid that
under a bushel in your introduction. Um,
you're you're good at design. Like you
could design a better slide than this,
but from an LLM oneshot perspective,
this is vastly better than I've seen
before.
>> This is actually really good.
>> It's good, right?
>> The spacing, it's doing the the small
things that you take for granted. It I
mean, that's a big part of, you know,
the principles of design is just
balance. And you know, at first glance,
this is very presentable stuff.
>> It's really like look at how they've
chosen to center this in the boxes. Like
that was the kind of thing that drove me
nuts when I was doing PowerPoint design.
>> Absolutely. I mean, the spacing between
the boxes themselves is about balanced,
right?
>> The type faces like executive summary
like proximally is the largest then it
gets smaller, right? like the hier the
type hierarchy.
>> This is really really cool. The question
I have for you is
>> what did you feed
>> what did you feed Claude for you to get
this output?
>> Let's look at the prompts. Right. What
what Nate video is complete without some
prompts.
>> That's right.
>> So I'm going to go back. Give me a
second to pull up my Claude here. Um,
and we're going to check out these
prompts cuz I I think that a lot of the
key to it is, and I found this when I
was digging in and comparing Claude and
agents. So, let me like state the
principle and then we'll get into the
example. So, what I found is that agent
mode, which is the most directly
comparable example, has eyes that are
bigger than its stomach. It tends to do
more tabs. It tends to do a more
advanced spreadsheet layout plan. it
tends to get deeper into the weeds on
analysis, but the execution is much more
poor and so the spreadsheet doesn't have
complete formulas and the layout on the
PowerPoint is absolutely terrible and
you can't ultimately use it even though
it was really aggressive. Whereas Claude
is more conservative. It focuses a lot
on tool use. It may not be as
sophisticated at raw analysis, but what
it delivers is very strong.
And so with that in mind, I'll show you
the um the prompt I actually used. So
this is interesting. I used Perplexity
to help build the prompt for this
because I found that Perplexity is
really good at collecting real- time
data and then combining the data into
like make this into a prompt, right?
where you have the self-contained uh
financial data or whatever it is and
also the prompt all in one thing so that
every model gets the same start.
>> That's really smart.
>> And so that's an example of using an LLM
to actually have some fun with it. And
so
>> great great trick too for anyone that's
listening. I mean
>> let let the LLMs help you to craft
better props. The the best thing is when
you get into that circular rhythm, you
start to really leverage the value pre
like upfront and that's really powerful
stuff. All right, so now I'm going to go
over and you can see Claude here, right?
This is for I did multiple problems. So
I'm hopping into the prompt I used for
valuing Oracle. So I don't know if you
noticed, but Oracle had I think a 40%
pop at the open because of their uh
reporting. Uh, and ironically they
missed on topline revenue and I think
they missed on something else too. And
the reason their stock still popped was
because they issued forward guidance on
AI that was so strong the market look.
So I felt like that was a wonderful
timely question, right? I was like, what
do we make of that? Um, and so I
actually
>> why didn't you why didn't we record this
yesterday before the pop so I could have
invested a little bit of money into
something. Thanks, Nate.
>> Listen,
right? Yeah, we need to do these like
right when the earnings drop so we can
have earnings uh reflections before
things happen.
>> Let's learn for next month. So, this is
what Perplexity came up with. I was
like, Perplexity like set this up. So,
basically what Perplexi did was it said
here's the challenge, right? You have to
value Oracle Corporation correctly. It's
going to feed you data current as of
September 10th. And it just like it just
makes it so easy to pull all of this.
It's like this would take me forever to
pull out, you know, EBA to margin and
cloud revenue growth and this and that.
And so it just pulls all of that out. Um
it pulls out a
challenge. Please provide a discounted
cash flow valuation with sensitivity
analysis. Um and then it goes in and
says here's the current market data.
Here's the fiscal. Here's the
profitability. Here's the cash
generation. All of this other stuff,
right? Like I can go on and on and on.
Is is producing this? I don't want to
pretend I produced it. Like perplexity.
I just asked it to do this. I had the
clarity of intent. Um, and it's a
complete data package. And what's
interesting is I learned about prompting
this model a little bit from this
exercise. This prompt does not specify
the output format for the model. And
what Claude defaults to is not Excel.
And so what you find over here on the
left is Claude says, "Great, let me do
the discounted cash flow valuation." It
starts to kind of go through. It does
all of this stuff here. It's sort of
spitting out free cash flow projections,
looking at terminal value, where the
where the company should be valued at,
uh what is the sum of present value, all
of that. It does all the work. It goes
to the end. It has key observations. It
has an investment recommendations. It
says it's fundamentally overvalued by
80%.
>> That's insight right there. The
intrinsic value cannot support the
current market price. Quad is dead,
ladies and gentlemen.
>> Oracle.
>> Time to short short Oracle.
>> Short Oracle. I apparently that's the
lesson the AI is giving us. This is not
investment advice, etc.
>> And
I I I say, "Okay, uh, you did not do
what I asked, but I didn't really
clearly ask for it, so I don't fault it
because if you look in this prompt,
there's nothing here that says do an
Excel." And so I say, "Okay, please
formulate this as an Excel." I don't
ding it. Right.
>> Right.
>> And it goes through and it then creates
the Excel model. Um, and what's
interesting is it gives me both a model
and it gives me a user guide with
documentation and it tells me what's in
the model and everything it puts in
there.
>> Um, and so I can just download it and
immediately start to use it. Um, or I
think I can actually click on it here
and you can see a sneak preview. So this
is not really fun like it's more fun to
look at it in Excel. Fun in Excel and
but you can see like it's telling the
truth. It has like four tabs. It has a
sensitivity analysis etc.
Um,
>> this is so cool.
>> So, that's how we're doing.
>> Do you want to see how agent handled the
same challenge?
>> I'd love to.
>> I'm gonna have to stop sharing my
screen. I'm going have to go back and uh
I'm going to go and dig up agent here.
And what's interesting is agent had
exactly the same challenge. It had
exactly the same prompt. And it just the
the response is just so different. Um,
so let me pull
>> Isn't that interesting? like across
these different LLMs, your miles may
vary every time and some of them are
just so much better at certain things
and the other ones are not and it's it's
kind of a shift.
>> I mean, that's part of why I want this
conversation, Rod, because like I feel
like one of the things I am taking away,
one of the learnings I have as of
September 10th is that if you have
PowerPoint or Excel work that you need
done, you need to go to Claude for that.
>> Yeah. You should not be messing around
with anything else at this point. Like
it just doesn't justify.
>> Exactly. And you know what? You know
what? This also opens up a conversation
around
>> all of those rappers that have built
their tooling around,
>> you know, really changing like, you
know, the themes in, you know, Google
Sheets or
>> or PowerPoint. You know, this is this is
one of those rapper killers that, you
know, you you look at it. I can take
that presentation and I could walk I
could walk into a meeting with that
>> walk right in right like it works
>> and that's powerful you know like one of
the most powerful things that I found in
my career my job is you always want to
make sure that you're able to
effectively communicate right
>> the things that you're trying to share
with people and you you might have a
presentation or maybe it's a power up an
Excel doc right and the funny thing is
you need to rehearse that you need to
spend the time tweaking it, going
through the cycles. And you're kind of
doing that in the LLM as you're going
through. Like you can actually rehearse
as you're building up those
presentations. And all of a sudden, you
don't need to stop 3 hours, 4 hours, 2
days before you need to go out and have
that conversation because you could
rehearse up to about 30 minutes and then
say, "Hey, give me this presentation
because I need to go out there and I'm
well rehearsed. I know exactly what I'm
talking about. There's no gotchas." And
that's
>> that's powerful.
>> That's right. I love that call out
because like there's that sense of
iteriveness, right? where you are
learning how to say the line, how to
find the talk track as you chat with the
other.
>> Yeah,
>> that's fantastic. That's really cool.
>> All right, we're going to see what
happened over on agent side. Uh, so you
got this
>> same prompt go through da da da da da.
Uh, so it then comes back, it thinks for
longer. I want to call that out because
I felt the difference. Claude would come
back consistently faster than chat GPT
agent mode.
Um, and then it comes back and says,
"Okay, I'm going to give you the
answer." Just like Claude, just like
Claude, it comes back with an answer in
text. And so it spits out a bunch of
text. And by the way, I had the text
here analyzed by Perplexity for both
Claude and OpenAI agent mode. And the
assessment is that you get a slight
extra point or two on the raw analysis
for agent mode because it was a little
bit more conservative as a financial
analyst
>> specifically $57 a share I think and
Claude had 62 and Perplexity was like
you know looking at the current
performance blah blah blah I think 57
probably makes more sense and I was like
okay and then it was like but I admit
that like both of them like think that
Oracle is severely overvalued. So, it's
very much a directional thing, right?
So, anyway, it comes back, it gives me
all this and I say, "This is not an
Excel, right? Formulate this as an
Excel." Here's the problem. Here's where
things go south. You can see already
there is no structure to this
spreadsheet.
>> Yeah, this spreadsheet is dead in the
water. It's unreadable. It's unuseful.
And it there there's nothing I can do to
pass this around. You can't send this to
someone and say, "Look at my analysis of
Oracle. What do you think?" Or really
honestly, "Look at Ched GPT's analysis
of Oracle. What do you think?" It's use
it's useless.
>> Right.
>> And and this is where I got really
frustrated with agent. Like I I I wrote
up a a summary of agent that like I'm
sure did not make new friends in San
Francisco, Rod, but like they were
trying to say like it's the best thing
since sliced bread. And I kept running
it through stuff like this and I was
like it doesn't deliver real workable
value. I see the hype. I see the promise
when it does stuff like this,
>> right?
>> But I can't get any work outputs.
>> And that is the that I think is the
chasm that all of these tools are trying
to solve. And I think that seeing Claude
do this so effectively is demonstrative
of how powerful these tools are getting.
You know, I mean,
>> really is.
>> It's also changed my workflow. I mean,
I'm used to the stopping point or the
edge of value of these LLMs to be, you
know, the text and the structure. and I
will go in and I'll spend the time to
say, "Okay, I need to get it to be able
to play in the playground that I am
external to it. I'm not going to use any
of these background tools. I'm going to
I'm going to have to copy and paste
things." And that itself, you know,
takes time. And the fact that we're
seeing Claude able to actually move from
within the workspace into these tools, I
mean, we're we're talking about, you
know, some gamechanging stuff. And we're
not really even talking about like
things when I'm really curious. I want
to push this to its limit. I almost want
to see it do a pivot table, you know?
>> We can do a live exercise here. I think
we should like in a minute here. I want
to like have you give me a live exercise
and we'll like set it up and we'll run
it while we chat and then we'll go back
and look at it. I think that would be
really fun.
>> That would be a lot of fun.
>> But first, I want to make you wse.
You're a designer. I want to make you
suffer just a little bit.
>> I'm here for it. You are going to for
better or worse, you are going to see
what agent put together for the
PowerPoint side of things. So this is
that same report you looked at earlier.
It's the same slide.
>> Look how painful this is. It's so
>> You know, you know what's funny is a
year and a half ago, this was really
impressive.
>> I know. Time flies.
>> This was really This was really
impressive. And then you showed me
Claude
>> and now this just doesn't
>> just doesn't work the bar.
>> It doesn't work.
>> Look at the these little footnotes are
completely unreadable and like
look this isn't
title like it's just it's just kind of
bad.
>> Well, listen. I think what we're
understanding here is that, you know,
the certain the agent is not the
designer, you know, between the two.
It's clearly the agent is just, you
know, he's just about the raw numbers
>> or she well and what's interesting was
that the conclusion that Perplexity came
to. So, Perplexity looked at the like
text outputs and basically was like if I
had to pick, I would say give your
assignment to agent to do some initial
analysis and then feed that to Claude to
actually do the report preparation and
the and the PowerPoint and all of that
because then you get the best of both
worlds.
>> 100%. It's funny that you bring that up
too because like daisy chaining is such
a powerful tool when you're using
working across LLMs. It's like getting
the best of all world. So
>> yeah,
>> that's a great call.
>> Yeah. All right. Well, let's have some
fun. I want to go to Perplexity together
and I want you to help me build a prompt
in Perplexity that we can feed to Claude
Live and sort of have some fun with.
>> Awesome. Because you said you had like a
pivot table idea. You wanted to have
fun. So what's what's your idea? What do
you want to do to like challenge Claude
a little bit here? And
>> we'll see what we get.
>> Interesting. Let's see what would make
for a good
You know what?
Given in the spirit of using AI, let me
come let me let me uh brainstorm some
ideas
>> as I would normally do.
>> Yes, GPT.
>> Do you want to share your screen while
you're doing that so we can kind of see
you in the brainstorming mode or or uh
are you too shy?
No, I'm not shy enough, I think. Here we
go.
>> All right. Well, then great. What I will
do is I will stop showing vertex for a
second. This is why this is live. Um,
and you can share your screen, I think.
>> Share button down there.
>> And here we go. So, I'm just going to
straight up ask, help me brainstorm
a few ideas around making a fun
pivot table.
>> I love that we're using the phrase fun
pivot table. Like, who says that? This
is why I want to
>> It's 2025, you know?
LLMs have changed our entire
perspective.
>> Pivot tables are now fun.
>> Make it dynamic. Make the suggestion
a
>> All right.
>> GBD5. Let's see what it says.
>> Movie night.
>> Pivot hacker.
Fitness fun.
>> Travel vibes.
>> Recipe. These are very consumer focused.
>> They are. Should we make them a little
more Should we change them and make them
a little more corporate?
>> I mean, we could or we could stick with
this. You get the test either way, I
think. What do you think? What do you
want to do?
>> Let's go with movie night. Let's make a
movie night pivot table.
>> Movie night pivot table. Okay. Let's
throw that into Perplexity. So, I'm
going to go over to Perplexity. Uh, or
do you want to go to Perplexity?
>> What I'll do is I'll copy and paste this
over to you so you can
>> paste it, right?
>> And then I can pop it in. Okay. Go
ahead.
>> A little team effort. We're gonna have
to find a way to bridge our
uh our hive mind AI.
>> That's right. I love those.
>> While I'm while I'm sending this over to
you via the interwebs.
>> Yes.
>> What are you What are you thinking that
is the most unique part of this launch
that made you just say, "Oh, this is
this is different. This is worthwhile."
What made you really excited about this
one?
>> I Well, I think I shared this at the
top. I was primed to be bored. I was
primed to not be excited. My expectation
was this is not going to be that much
because I've seen this hype before. And
what I found was just the ability to
deliver on promises around work done
became really really compelling for me
because it meant that I could be in a
spot where
I felt comfortable giving a bigger piece
of responsibility
to AI which in turn like if if I think
about it like a lot of the question we
face that we wrestle with is what do we
trust AI to do? How do we delegate
effectively? And how much time does that
give us back? How do we multiply
ourselves as AI professionals? And this
feels like one of those things where
like I can I can see all of these places
through my day where I can start to
delegate differently and I can push more
to Claude because I can trust him.
>> That makes a lot of sense.
I mean, that's one of the things I'm I'm
looking for whenever I see these um a
launch or any of these guys, whether
it's Anthropic or OpenAI
put out a new announcement, always
curious on like how can it change my
day-to-day, you know, and then
>> quickly try to get in there and and and
tease out that value.
>> Exactly. Okay, I picked up the text. Um
so, I'm gonna go back. I'm gonna share
my screen. And I have the text I think
in my clipboard and we're just going to
start to build a prompt live that we're
going to hand to Claude and see what we
get. So the pasted text, it's movie
night. Um, and what I basically want to
ask Perplexes is I want to say uh I want
you to uh take this seed of an idea and
expand on it. Uh specifically, you need
to build a prompt for another LLM to
construct a pivot table spreadsheet
uh that includes
uh a list of 20 uh to 25 recent movies
uh categorized
as described below. Um and the pivot
table
needs to be very usable um and
userfriendly. So this is like me
actually like prompting raw here. Uh in
addition,
the prompt you construct must contain
all of the data the LLM needs to
construct the pivot table. do not assume
the LM can go and get movie data,
you get the movie data.
And that I found was really key. Uh
because you can't try and give it like
the research job and the Excel creation
job in one go at this point. um it will
go and do research, but I find it's much
more effective to have Perplexity do the
research and then come back with
something like a self-contained prompt
>> and then work. Um, and what that
suggests, by the way, is that if you're
working with internal company data and
you're working with Claude, you're going
to end up in a position where you need
to be constructing a prompt very
intentionally pulling in internal
company data as part of the prompt
construction process and then giving it
to Claude.
Yeah, 100%. And I think that that's
going to be a nice exercise for us as we
see this first version that it spits out
>> because I'm already thinking about like,
you know, maybe we want to just focus on
say the Marvel Cinematic Universe
>> and then we want to overlay characters,
you know, and then we want to
>> add actresses and all this fun stuff.
So, I'm really excited to see what what
uh what Perplexity puts out there.
generating a complete data set of recent
movies. So like I cannot believe that
here we are in 2025 and this is just a
casual prompt that you and I are just
throwing up.
>> Look out IMDb, you know,
>> right?
>> Yeah. You know, one of the things that I
think that people don't realize, I was
talking with someone uh earlier today
who uh I mean he's a company leader and
he hadn't he he uses perplexity. He
loves Perplexity, but he didn't realize
that you could hit these three different
options on the bottom, and he just
defaults to search and loves it. And I
was like, there is more to Perplexity.
You can do even cooler.
>> You You unshackled him.
>> I did. I Well, I'm hoping I did. We'll
see. Um, so it gives me like a bunch of
uh recent movie titles. It gives me a
bunch of view records data uh showing
who watched what when. Uh, it gives me
pivot table specs. It gives me expected
output examples. user experience
requirements. This is very complete.
This is this is very very thorough. Do
we want to just like run it and see what
happens?
>> Absolutely.
>> All right. All right. We're going to go
run it and see what happens. Okay. I'm
going to stop sharing and I'm going to
paste it into Claude and we're going to
see what we get.
>> I'm excited. I'm rubbing my hands. I'm
not even doing anything.
>> I'm like, this is going to be really
fun. All right. So, we're going to copy
it. Um, and I'm glad I told it to be
self-contained because there's no way I
would be writing all of that stuff out.
All right. So, I'm pasting it in and I
actually want to do this live. So, once
I get it queued up, I'm going to share
my screen again and we're just gonna see
how it goes. Okay, here we are. Gonna
share my screen and go back to Claude.
Lot of screen sharing in this one
because there's so much cool stuff to
look at. Um, okay, here we are. And
literally like it pasted it in and you
can open it up and look at it and you
can see uh this is the instruction. It
does specifically say create a
spreadsheet. So, I'm hoping I don't have
to remind it to create the spreadsheet
this time. It has all the data we looked
at. Uh, in addition, it dumped in a
bunch of URLs because it just can't help
itself. That makes the prompt slightly
dirty, but I don't care. I'm just
leaving it in for now, and we're just
going to see what it does.
>> I love how Perplexity loves citation.
>> It does. It really does. Um, and so it's
starting to go through. Uh,
it's going to create a So, so it's first
analyzing it. So, it's got the Python
script up here. You can actually see it
like starting to code it in, which I
This is just addictive to me. Like, I
love this.
>> Look at the data.
>> Yeah, the world is a different place
now. It's just wild.
>> It's It's so funny that there was a time
where doing something like this was
extremely time inensive. And
>> I say a time as if that wasn't
>> as if that was a couple years ago.
>> It just feels so far away. Like we are
such creatures of convenience that the
moment you put things like this in front
of us, we were like, "Give me. This is
the new the new floor." You know, it's
really cool.
>> And so it's just going through and as
far as I know, like it's not even
touched the spreadsheet. So, it all it
did was it used Python to figure out
what it was going to do and then it's
decided, okay, now we're going to build
a pivot table. Um, and we're going to
like actually start to bring an Excel
file with quote unquote advanced
features. So, we'll see how it does on
that.
>> I really love the way it thinks. I love
the way it's it does the thinking out
loud and
>> yeah, the
>> and the way that it's leveraging Python
to give it a P, right? so that it can
then use that. And I think that's that's
something that I'm noticing across the
different LLMs. And I think that the
engineers are getting smarter about this
like they know that there are certain
languages that just lend themselves well
to going from human language to you know
computer
>> that computers speak and Python is just
one of those languages. So if you could
get into Python, all of a sudden
shifting into other languages becomes a
lot easier because you're getting that
structure built in. So really really
cool stuff when you when you're actually
paying attention to what's um what's
going on there. Uh you're not just pure
vibing, which I love to do too. I mean,
you know, a good a good vibe is great. I
think that when I'm able to pay
attention, I notice that Claude, like if
I'm not running back and forth between
20 different things, I notice that
Claude is much better about this than
OpenAI. Like the the tool record here is
really clean. I can see exactly what it
did all the way through. And now like it
loves documentation, so it's creating me
a read me. I didn't even ask it to
create me a read me.
>> This is so great.
>> How to use the spreadsheet. Thank you,
Claude. Sometimes Claude is a little bit
a little bit stuck up, but I love it so
much.
>> I don't mind it. Can Can you laminate
it? Can you collate it? Laminated and
and mail it to me too, please, Claude.
That'd be great.
>> So, you remember Calvin and Hobbs?
Claude has the personality of Susie
Derkens to me.
Like, it's it's that kind of a model.
>> That's a great call back.
>> Yeah,
>> I love Calvin Hobbs.
>> It's so good. My kids are addicted to
it.
Um, okay. Okay. Oh, wow. It
overachieved. It has a a complete pivot
table and apparently an enhanced
analytics version. What?
>> Amazing.
>> Amazing.
>> Okay, we have to see what's in here. Uh,
so I'm going to download both of these.
Um, and I'm going to pull down the
readme. I actually want to open these in
Excel because I think this is just a
crappy way to look at it. So, give me a
second. We're look at this in Excel and
see what's in the box. It's gonna be
very hot.
>> Excited.
>> I think we
>> While you're pulling that up, I'm I'm
also excited to see how
>> how Microsoft and these other companies
respond to this, right? Like
>> the amount of dynamic behavior that
they're going to have to start
introducing to their products. It's just
opening so many doors.
>> Yeah.
>> So, it's so cool.
>> My gosh, look at this. Okay. I didn't
even know it could do this. I know. I
just I I'm killing you. The suspense
alone h half the audience is is keeled
over.
>> Yeah, I think has become exciting thanks
to Nate and Rod. Um,
>> so this is the enhancement.
>> It did color coding.
>> Oh, heat coloring.
>> Yeah,
>> it heat maps. I'm like kind of
impressed. I didn't ask it to heat map.
It volunteered.
>> Cool.
>> It did pattern analysis. Look at this.
Wow. I'm kind of blown away. I didn't
know it could do this. Rod, this is also
me reacting.
>> Are those
>> Those are little like uh
>> Do you zoom in? They're like
>> spark lines or something.
>> Smart graph line thingies. Whatever you
call these things. I don't even know.
Like I I I'm not the the Excel expert
here. What do you call these things?
Spark lines.
>> No, I don't. I I use language that I
didn't didn't really know. Like little
bar graphs or something.
>> Yeah, they're little bar graphs right in
the cell. And then it has this cute
little like okay like you know take away
a little credit because the genre is on
top of the bar but whatever like I would
say good
>> uh you get friend statistics so you can
see what these different friends like
unique movies diversity etc. You get
your top 10 movies ranked.
So like this is the Suzie Derkens
overachievement version
>> and then it has all the raw data right
>> and the raw data that's what we love to
see. That's so good.
>> Yeah. So, like you can tell I'm not
lying is the implication, right? Um,
and so there it is. And so that's the
enhanced version. I'm going to go ahead
and share with you what it called the
basic version, which I thought was still
perfectly usable. And you can kind of
see like I think you get a sense of like
what it thinks is special, which is its
own little piece of insight to grab. So
yeah, this is it. This is regular movie
night without the heat mapping. Um, so
it's exactly the same table. So you see
the 12 is there and all you have is a
nice clean traditional finance table
with blue and white rows data perfectly
understand
>> but you know
>> they segment out each of the uh viewers
>> that's
>> that was not in the enhanced version so
they actually data on Alex Jordan Casey
Morgan and Riley and you can see like
how they go by day what their genres are
and so in a sense like one of the things
that I am learning from this is that you
have to the the more interesting you
make the prompt for Claude, the more
likely Claude is to give you some
permutations. And what I might do in a
followup here is basically say, "Claude,
I loved the heat mapping on the pivot
table.
>> Can you please preserve the heat
mapping, but also give me the viewing
pattern pivots and like do a
modification like as a
>> That'd be really cool." Actually, open
up the pivot table.
>> Uh, this one all friends. Yeah.
>> Yeah. And can we pull in can we go into
the If you click on data. Uh where's
data? Okay. Yeah, up here. Please go
away. Um
yeah, data.
>> Let me take a look. It's been a long
time since I've been inside Excel. I
will tell you that.
>> Are you looking for or something?
>> I'm looking for the pivot. There's a
there's a table that you could actually
see how it creates the cross
relationships. Oh, is it in analysis
tools?
Is it?
>> No, it's not. Um,
what if analysis, text to columns,
flashfill, queries, and connections.
That's it, isn't it?
Is that it? No, that's not it. Um, we'll
have to find that later. I don't think
we're finding.
>> We will. That looks really cool. This is
This is awesome.
>> Yeah. No, it's super fun and I'm
learning a lot, too, as we chat. So,
there you go. That is it. That is what
Claude has launched. Do you have any
takeaways from this?
>> I'm I'm excited. Say that again.
>> Do you have any takeaways? Like what
what are you taking away from this
conversation?
>> I mean, for me, what I'm realizing is
that the idea of using these tools is no
longer relegated to just me as a person.
The amount of decisions that I need to
make around getting something valuable
out of the outputs that I'm getting can
really really start diving into the more
interesting bits, right? One of the
awesome things that we've seen over the
past couple of years with these LLMs is
that we all get, you know, we're all
over the first 20, 30, 40%.
>> And most of us by the time we're really
proficient at what we do, the 80% is
pretty much, you know, you could phone
it in,
>> but it's that last 20%
>> that makes things go from good to great.
And the fact that we could spend more
time focusing on that 20% and this is a
perfect example of that makes for a
really interesting like future in like
both you know the professional space
just the communication aspect. I mean
this is just really really cool stuff.
So
>> so I want to give you a philosopher
quote that I promise is relevant that I
am taking away as I look at this. Um
this is from Alfred North Whitehead
um who famously said, "Civilization
advances by extending the number of
important operations which we can
perform without thinking of them."
>> That's actually really that's a really
good one right there.
>> And I feel like extended civilization
because there are important operations
where I didn't tell it to put the heat
map in. I didn't tell it to center the
the text on the PowerPoint slides. It
just did it and I didn't have to think
about it.
That's awesome. Yeah, I like that. I
also like that positioning.
Anthropic might come knocking, mate. At
this point, you're you're saying that
Anthropic by extension is extending
civilization. Just fantastic.
>> Yeah. No, I'm sure. Well, if Daario
calls, I'll pick up the phone.
>> Well, thank you for having Thank you for
for chatting a little bit. Thank you for
making time. I know this was a bit of a
last minute thing. We're having fun.
First of May,
>> loved it. Looking forward to doing
another. Have a good one, Nate. Cheers.