So, I got to admit that I was excited

when the government had banned Fable out

of the gates. I think they were all

things that that like Opus probably

could have done with a lot of steering.

But, Fable pretty much one shot at them.

>> Converting a 120-page

operating manual on

>> [music]

>> building a hedge fund style financial

model is getting like quite a bit

easier.

Welcome to the next episode of Invest

with AI, where we explore the

intersection of fundamental investing

and artificial intelligence.

>> [music]

>> We have a short duo episode today where

we'll do just a little bit around the

horn vibe check

on AI in investing. And one of the key

topics people are discussing these days

uh is the new launch of Fable. Okay, so

what uh what are the vibes on on Fable

by your assessment?

>> Yeah, great to be here. Uh great to see

you. So, I got to admit that I was

excited when the government had banned

Fable out of the gates. Uh not because I

believe in any paternalistic

intervention of any sorts, uh but I was

just like, "Shoot, I'm not ready."

Uh

and I think that you know with all the

hype on these like big expensive models,

I I got to be honest with you and I

wonder how you approach this is I don't

have a great set of evals.

And by evals, it is like, do I have a a

a complex task

or series of tasks

with a lot of good context that is hard

and that prior models have not fully

succeeded at. And so,

I don't have like a predetermined set of

evals, but I do have one thing that I

run pretty regularly with our whenever

new model comes out. And that's

basically I take I have a giant folder

of like everything my business has ever

done. So, you're talking like

uh 200 granola calls,

50 decks, MCP into my email,

um

different like surveys that people have

filled out, the recordings of all my

trainings, and I dumped that in and I

say,

"Build me a website.

Create a strategy uh document for me.

Create a presentation from scratch just

based on like all of these materials."

And

I got to say, the result was pretty

good. Like, the website was beautiful.

The deck was pretty good.

Um but most importantly, the strategy

doc, cuz that's where, you know, I'm

intimately connected with my business.

Like, I know it better than anyone else.

So, like, if someone's going to give me

a unique insight, I I can like, that's a

unique insight.

And I'll point here, and again, this is

not the most robust eval, but what I'll

say is

I had a call, a prospecting call with

someone that was not even really a

prospect, just kind of like half

networking, half biz dev, but they said

something to me that I was like, "Ooh,

that's very

That's like, I feel seen. Like, you you

saw something that I didn't see in my

own business, and it's one of those

things that's like a little

uncomfortable where I like where I know

I should pursue it, but I'm not pursuing

it. And there's like reasons why I'm not

pursuing it."

And it was just one like one minute

inside probably

10,000 minutes of phone calls.

And it jumped right on that one minute

and it's like, "You should be pushing so

much harder on this."

And that was again, I don't want to over

analyze this, but I had forgotten

that that was a very salient point when

they made it. And it was once. It didn't

happen five times, it happened once. And

the fact that it read through 10,000

minutes of calls

and was like, "This is the thing you

need to focus on."

was pretty shocking to me.

So, that's what I'll say on the like

formal eval side of things or the

closest thing I have a formal. Uh I had

it build a lot of things from scratch.

So, I had it build like a tool that any

client of mine can describe themselves

in a multi-step process. Like, I'm a

fundamental investor. This is what I

look at. This is my investment horizon.

And then it generates like 20 skills

that they can then copy and paste into

Claude.

And I had it use all the skills that I

had ever created for any of my clients,

but also comb like YouTube and GitHub

and like any public repository.

So, that was pretty cool. I think they

were all things that that like Opus

probably could have done with steering.

But, Fable pretty much one shotted them.

Now, I can't tell you about the price

cuz I'm on the subsidy still. So, like

it's hard to know like was that like

$500 worth of credits or 30? I I don't

know.

Probably closer to 500 than 30 if I had

to guess. Was it worth $500? Probably

not. Was it worth 30? Absol- so freaking

lutely.

>> Very very interesting.

Um

I agree so far that like the eval is

hard. Like, eval is hard. I have a few

that I still in the pipeline to run.

Like, simple things like the level two

you know, financial modeling tasks that

are just

you know, complex and heretofore the the

models haven't been able to

accomplish. Uh so, I'll sort of get to

those.

Been traveling a little bit.

Um

but I was I was running one where I was

trying to turn

um uh the process to identify signals

around good times to short an individual

stock.

And Fable did an amazing job of sort of

going back into the history of good

times to short that stock and created a

really coherent, you know, pattern

recognition engine that historically I

would have had would have had to steer

quite a bit. It would have taken me two

or three iterations to say, "Hey, no, it

doesn't you know, we're not worried

about a goodwill impairment or gap

earnings." These are sort of the things

that ultimately are triggers to

to stocks and I may have to feed that

in, you know, transcripts, etc. or my

own research and

it one shotted it, which was which was a

little bit scary. It sort of catch

catching that same fundamental dynamic

that you um, you just mentioned like

materiality or taste. Like what's the

what's the white right right right right

word for it?

Um,

>> Yeah, it's like incisive judgment.

>> Yeah. Yeah.

>> Like

>> Yeah, which is it's it's wild. Like I

guess we would need to bring some

experts in to explain exactly how

how they they cook that in. Um, but I am

start I am launching a new eval, which

is a real real-life eval. I'm taking 100

companies

and I'm running 100 earnings previews as

earnings season starts next week and I'm

going to basically give Fable

two tasks uh,

forecast revenue EBITDA and EPS

accurately and accurately forecast where

the stock's going to trade up or down on

the earnings print. So, that's a very

tight sandbox of

tight sandbox of judgment

and there's no hot there's no hiding

from that. I'm going to keep maybe I'll

maybe we'll talk about that, you know,

the results once earnings season's over.

I'll put some of that on Twitter.

Obviously,

but I thought that'd would fun I was

inspired I think the the talk in my

network

for the last week was this uh Thinking

Machines Bridgewater report where

uh the the researchers took a large

language model so it which had native

judgment in financial tasks in the high

40s, overlaid financial

sort of

a judgment layer financial expertise and

judgment tasks raise up to 75% like the

mid 70s.

Um that wasn't the most academically

rigorous report but I think

directionally was inspiring to a lot of

people thinking about this combination

of

human judgment with large language model

judgment that the combination could

drive better better results.

Um

and so I've taken a lot of time to

codify into my earnings preview process

the frameworks of judgment that lead to

my estimation of whether a stock will

trade up or down.

Things like setups and expectations and

buy-side whisper and business momentum

and guidance trajectory etc.

Um and so I'm excited to unleash that

unleash that with the Fable

uh model in this exercise. Um not I'd be

shocked if it does really well

um but I want to just at least kick this

off cuz then the iteration into version

2.0 and feeding that back I think is a

fun data set to start to start uh

building.

>> Can I ask so I hadn't seen that paper.

Um is it are they fine when you say

they're adding the judgment are they

fine-tuning the model with that or is it

in a skill or in a prompt? How is that

judgment being

uh incorporated into the

>> I don't know I don't I don't know the

specifics. I read it a you know read it

sort of relatively quickly and maybe we

need to try and get one of the authors

in that would be a that would be a fun

conversation.

But I think the the conclusion was they

they took this sort of codified

analytical judgment of the Bridgewater

team and overlay that on the large

language model approach.

And the com- combination the combination

was valuable. And sort of in a in a um

in a in a very few way, I've started to

see that, right? When I started to see,

okay, take this take this approach

almost train it out of think codify some

of the setup patterns that, you know,

I've traded in the past. It it's a quick

study. Like the models have been a quick

study on those patterns and and

following on the rails of those

those that that pattern recognition

power. There's been a few instances

where I'm like, "Huh, that was like

nailed it pretty pretty pretty

accurately." Not in a scientific way,

hence I want to start sort of doing this

in a in a broader subset to see if did I

just get lucky and heads flip you know,

head flipped on the coin twice in a row.

Um so I'm inferring skill, which is why

um my plan to do this in a more

systematic way in in the earnings prints

coming up starting next week.

>> That's cool. Wow. One thing I'll flag, I

know we advertise this pod- podcast as

not having the answers, but us just

talking through what we're what's on our

radars and um

someone

who knows a lot more about it LLMs than

I do had mentioned something called I

think it's called braintrust.dev

and it's

it's a platform for running evals.

And he had actually suggested

that that I do an exercise kind of like

the one you're describing. I'm not I'm

not close enough to fundamental equity

and investing to to do that. But he had

suggested that I do that using this

platform. So just putting it on your

radar to maybe check it out that that

there's actually like a structured

environment to run these evals and we

can run

>> Oh, that's interesting.

That's interesting.

Um

yeah, that's one of the one of the areas

in our work where we're really like I've

done a I've done evals a lot on vibes,

right? It's like, you know, you try you

you shoot you you shoot off a few

workflows. You're like, "Ah, this looks

good."

But the problem is even when you shot

off five things in Opus 4.8, like you

kind of like sometimes it works,

sometimes it doesn't. So, it's very hard

to scientifically measure vibes.

>> Totally.

>> One of the one of the things we're

trying to do with clients is create a

more scientific eval set of the various

MCP inputs, right? So, who are the

modeling MCP vendors? Let's create an

eval set so we can more scientifically

score some of those vendors on on

accuracy uh accuracy and capability. Uh

so, it's a I'd say that's like an work

in progress

work in work in progress on our on our

side, but yeah, good call out on the

brain trust I'll definitely take a take

a look at that.

>> Cool.

On my end, I'm a little late to the

game, but it's something that I'm having

a lot of fun with. So,

we're both Karpathy Andre Karpathy fan

boys

and probably, you know, like he drops a

banger of a tweet every 6 weeks and the

AI world like it realigns around this

new idea. So, maybe two or three tweet

cycles ago, so that might be, you know,

18 weeks ago,

he had something he called it kind of

this like self-updating

wiki, like a knowledge wiki.

And and I kind of didn't have time to

focus on it when it came out and I'll

circle back to the wiki with a very

common problem that many of my clients

are having right now. And that problem

is they've gotten really good at

co-work. Some are using code, you know,

and so now they've got good prompts,

they've got good skills, all that stuff.

They've connected their MCPs.

But let's say they've got an investment

and in this this like private equity

fund

they have an investment and there's 500

documents in that folder. Let's say

that's you know, the investment is I

don't know, Sweetgreen, right? So they

have this Sweetgreen

assuming it was private. They point it

to the Sweetgreen folder, they've got

500 documents in that folder.

And it's not even organized and so on.

And so if they want to ask questions

about it, they could be very tactical

questions like

you know, how has the store same sales

growth changed over quarter over

quarter? Or they could be more

qualitative questions like how should

this thesis how is this thesis being

informed by inflation? All right? Our

our tariffs, right?

So you point at Cork. Cork's Cork can't

go through 500 files, right? And so what

it does is it kind of

cherry-picks it and what it really does

is it does like basically like a lot of

like command F.

So like let's say you're doing tariffs,

it's going to do like command F on your

500 files and be like tariff, command F

uh inflation, command F and then it's

going to grab 17 files, pull them into

context and then try to answer your

question.

And it starts to break down more and

more as you have more and more

documents, more and more MCP. MCP breaks

down too because the context windows are

too small when you're dealing with like

huge volumes of data.

So that's the problem that people are

starting to have and it's even more

complicated by the fact that there's

PDFs, there's Excel files, there's you

know, 200-page, you know,

uh credit agreements and so on.

So here's where the Karpathy wiki comes

in is that the way this this works is it

says we're not going to we're we're

going to create a map

of your information, an an ontology

of your information. You might have hear

that word, people throwing that word

around.

And so what it does is it basically,

let's take it takes the 500 files

and one at a time

well, first it will convert it into

markdown.

And then it will basically read it.

And as it reads it, it kind of creates a

summary at the top with tags and

metadata.

And then it creates like links of ideas

inside

the text that it just read. So, if you

have something like, I don't know,

tariffs, right? It will just like find

all the references to tariffs.

Then you ingest the next file and it

does the same thing ex- with one

exception. As it finds tariffs, it then

looks in all the other files and says,

are there any references to tariffs in

the other files?

And then they start to link and you kind

of have this kind of like you could

think of it as kind of

like a private worldwide web, right?

Like a little little wiki a wiki, right?

And so what's cool about this is that

the LLM is doing the ingestion.

So, like taking the file, creating the

links. And then as you get new

information

it just keeps updating it. So, there's

this like self-improving process.

Now, I'm going to be honest with you

like

again, it's another one of these things

like hard to know if it's working versus

the naive approach, which is, you know,

just pointed at 500 at Fable.

But that being said

I I know that some clients have done

this like, you know, maybe like a

concentrated like a PE fund that's like

15 investments. They're like, we're

going to go through this process for our

15 investments. We're not going to look

at the 200 we passed on.

But we'll do the 15, we understand it

well, every analyst can kind of own

their name in the wiki. It's not like

your style of investing where the turn

over, you know, you're talking six-year

investment period, so you're not

flipping names and there's not data

you're responding to every second. It's

a little bit different. But what you

start to have is this knowledge graph

that is much easier for an LLM to

traverse

without blowing a ton of tokens and by

increasing the depth of accuracy.

So, early days early days in my own

exploration on it. I do have a few

clients that like I I don't want to say

like I brought it to them like they kind

of came up with it we like they came up

with themselves and we can compared

notes. You know, a lot of these a lot of

this is around this idea of like if you

can get all your files neatly organized

in markdown with good naming, good

metadata inside like a nice clean folder

structure that the LLM can go in and

read it much better. And so, you're

starting to see in again

I encourage everyone or I discourage

anyone from doing this like oh, we're

going to like do 20 years of data across

this like no, no. Pick like five names

and start. Pick one name and start. And

so, that's starting to happen and and

I'll be able to report back on the

progress. But I have seen some green

shoots that again, this is more in

private equity but it works.

>> It's It's interesting and I've heard

more about the same concept of

um

just making your file system more

legible to to the agents. How are How

are clients doing that? Is that a skill

that co-work or code x can go in and

organize like that even gives me hives

when I start to think about organize

like letting code x rip on my internal

files to make them

>> Oh, yeah.

>> legible or people just pressing the

button and hoping for the best?

>> Yeah, I mean I think like any of these

things you start with a little pilot.

And so, maybe you've got you know, the

sweet green folder with 500 files and

you'll say

uh find me all the financials and put

them in the financials folder and then

give it a descriptive file name.

And then you go in and you're like "Did

it do it?" You know, and as the analyst

you know, you know, sometimes an Excel

file is a client list. You're like, "I

hope it didn't take the client list and

put it in the models folder, right?" So,

you kind of have to do that testing, but

it's it's very good.

And so, uh so I So, then you can kind of

make this a more

robust process like

every time a PDF comes in, immediately

turn it convert it to markdown.

Right? That That I know a lot of groups

are doing that. Um and then add the

metadata while you're at it. So,

it's actually quite powerful The models

are very good at doing that. All right,

you could even do that with a Sonnet

model. You don't even need Opus um for

like the the the just like the the the

tagging and the cleaning and the moving

and so on.

Now, when you want to do it in, you

know, bigger batches or with more

complexity to it, more one-shotting, you

would use a a bigger model, but I would

say again like kind of start small, but

you I think you would be very surprised

that, you know, this is the file And

again, it all comes back to having good

skills. Like, this is the

naming convention that we use for our

files. Year year year month month day

day, you know, quarter, you know, if

it's a financial this.

Um

and then another advant- another thing

you do is like you literally have a

document that's like, "These are how the

folders are organized." So, you think

about the LM comes in, it reads this

kind of index document, and it's you're

asking a question that involves models,

then it's like, "Oh, let me read this."

And And this file says, "All of the

models live in the financials folder,

right?" Uh and so, it just saves the

model and saves the agent a lot of work

in like, "Where Where do the models

live? You know, do I need to open this

file?"

>> Yeah. Yeah. That makes uh that makes

sense. It's interesting we

I think we both have the same

fundamental rule. I call it the Karpathy

rule of wait for Andre to tweet

something and then trying trying to

through how it applies to our space. I

I've used that same Wiki knowledge Wiki

concept to build skills architectures.

And a lot of what I've been doing is is

really like the brain dump of everything

I know about a certain concept, putting

that into as deep of a

sort of knowledge pool as possible. And

then taking that same knowledge graph

concept to sort of to make it agent

legible for skills creation. Uh and

that's been quite slick uh from a

process perspective. It makes a lot of

the hard part actually is like compiling

compiling the knowledge.

Um

but converting uh you know, for example,

converting a 120-page

operating manual on building a hedge

fund style financial model,

taking that raw data into actual

operational skills architecture is

getting like quite a bit easier if you

take that interim interim step. Um

so that's been um

that's that's that's been fun.

One of the things I've been spending a

lot of time on is thinking about just

the the the um

uh the stack of agents. Like what's the

right Like we sort of had this concept

back last year like four to six pages is

the right

uh depth for a prompt, right? Because,

you know, a one-page prompt was probably

too thin, not descriptive enough, but an

eight-page prompt was just too much, and

anything in the middle

the large language model would lose

attention.

You know, that I'm curious your

perspective on this in skills building,

too. It's sort of trying to dial in

myself of like how thin is too thin or

how thick is too thick for a skill.

And um one of the concept I've been

using is a sort of like raw primitives,

almost like single-purpose skills,

single-purpose agents

that are composable into these

orchestrated pipelines.

And the sub-agents, like the ability of

a ramp skill to go pull in a dozen

different sub-agents,

individual skills into this workflow

pipeline.

I'm getting some really interesting

stuff out of that. Um that's sort of

composer composability

uh process which has been pretty

exciting to me so far. Exciting and my

wife would tell me I'm a nerd and then I

get excited about skill skill

composition, but it's a little bit of

like a breakthrough on my side of uh you

know, consistency and um

each of each individual piece is sort of

doing a much better job adhering to that

individual skill. Whereas, if I load

everything into one skill, it's like

things get lost in tran- things get lost

in translation.

>> And you know, I haven't spent as much

time on that. My skills do tend to be

pretty narrow just by default. But I

haven't like it's not been a design

choice, just kind of the way the puck

landed.

Um

but one thing you'll notice, I don't

know if you notice this on Fable,

but it does uh it's kind of

doing a lot of the sub agent work on

your behalf even without skills.

And so, for example, if you ask it to,

you know, create a

uh I don't know, create a

uh final memo, you could see it it's

like

uh I'm going to create a bunch of

different sub agents to read these

documents and then report back.

And one of the things that I'm just

starting to realize on sub agents is

that sub agents use their own context.

Right? So, they ring-fence their own

context window so they're not polluting

the context window of the larger thread.

And so, if you have these like very

discrete agents

driven by skills as you've just

described, you have this like really

efficient kind of context gathering

exercise

that then extracts the key information

up to the month, you know, the

orchestrator or the main thread.

Uh that then can kind of reassemble it

and uses like the heavy duty Intel, you

know, the fable, the open open style

intelligence. Now,

again, this is more through observation

than from things I've like consciously

architected myself, but I know that

people, you know, you know, I was a big

open claw person back in the day. Like

people who are using open claw were

thinking through this

6 months ago of like

>> Yeah.

>> when do you need [clears throat] to

close ring fence contacts, when you need

to feed contacts into the main chat.

Like this is why I always love talking

about open claws cuz they were 6 months

ahead of the conversation, whether it

worked or not, right?

>> Yeah. Yeah, no, it's interesting. It um

you know, like complicated Excel

modeling has been one of the areas

that's been quite disappointing um

with with AI and I've had to really

chunk things into individual steps.

And for whatever reason, AI in the Excel

wrapper just doesn't listen to my skill.

It doesn't follow directions well.

I asked a

I asked a um a contact about that and

they're working, we will bring him on.

Um

his firm is working on a sub agent

approach. And that's sort of like that's

very interesting to me. Like if I could

take a highly complex LBO model or hedge

fund style model,

it's just too much like it's too context

inefficient, token inefficient today

to do that in AI Excel system, but if

you can spawn 30 different sub agents,

one to go clean up the cash flow

statement, one to go build out the

interest schedule,

and then you bring in a fable on top of

that to assemble those analyses in,

um you know, deploying deterministic

code where it makes sense.

Um obviously the proof is in the

pudding, but conceptually that feels

like a really interesting engineering

approach

um

approach to this this this problem.

>> And again, I I don't know enough about

Fable, but it seems like it's trying to

do with all of that on the fly.

>> Yeah. Yeah.

>> Which would be crazy.

>> Yeah. Yeah, exactly. Exactly.

>> we're like talking about customizing

every sub-agent, but

again, we don't want to get, you know,

um, get in front of, you know, get in

front of the story, right? Or mislead.

But it's there are little breadcrumbs

that it has

the capability to do at least some of

that.

>> It makes sense. One one question I had,

um,

how are you seeing your clients think

about the tradeoffs between Claude Code

and Claude Co-work, which is really just

a different harness? Like where do you

find Like what do you find are the pros

and cons of that that that decision to

be?

>> It's It's funny you say that. Do my

clients freaking love Co-work.

Uh, and Co-work I don't I like Code, but

I'm kind of nerdy. I like the weirdness

about it. Um,

they love Co-work and I'll give you uh,

so so

and I teach Code if people want to learn

it, but now, even after I teach it,

they're like, "What's the difference

between Co-work?"

And there are some differences, but the

differences are actually like collapsing

for knowledge work. Like it they'd be

hard to notice. Like for coders, you

would know the difference. Like a coder

would never use Co-work.

But it's the blur the lines are being

more blurred. So a few things that are

pretty cool on on

on Co-work. One is the There's this

concept of living artifacts, live

artifacts,

where you can run a dashboard off of an

an Excel spreadsheet.

Uh, and as you change that Excel

spreadsheet, the dashboard changes.

Which is like one of the many one of the

main reasons why people were using

Co-work Claude Code anyway was to just

create more interactive visualizations

of data.

Uh

so, that like knocks out a problem.

Uh

another one is that Claude co-work

is pretty aggressive in solving problems

with code.

So, even like if it can't figure

something out, it might say like, "Can I

Can I write a Python script to do that?"

And it it didn't used to do that, or I

felt more constrained in the past on

that. So, it's getting a little bit more

aggressive, co-work.

Um

the other reason why people like co-work

is

the schedule tasks, right? That's kind

of how you trigger agentic workflows.

Um

they're so clean to run in co-work.

There's a You can use your mouse, you

could see the which ones ran, which ones

didn't want run, which ones need

approval. When you run schedule tasks in

code, it just goes into this black hole.

Like, it's like a it's like a There's a

markdown file that like, "Did your thing

run?" Like, you know, and so, the ease

of the user interface.

And then, this is something I actually

just learned the other day cuz you and I

don't get too deep into the infosec side

of things. Um we kind of assume that

people come to us once they've cleared

the infosec bar.

Uh information security for those

unfamiliar with the acronym.

But, one thing I just recently learned

is that So, what Claude co-work does

is it actually creates a virtual

machine. So, you can think of this like

a disposable desktop. And so,

it's actually really hard to blow up

your computer cuz you're actually using

a copy of your desktop. And I don't know

exactly which cloud it lives on.

But, with co-work with Claude code,

you're actually giving it root access.

So, it's not a copy, like this

cloud-based copy, it's the real shebang.

It's your computer, right? And so what

I've heard from IT departments is that

especially if the gap is converging in

what they can do, they're like, "I don't

want to give, you know, Brett root

access when he's not even a coder.

I just want him to be able to use the

cool features of Co-work in this like

virtual machine environment that's much

safer." So I didn't actually know this

distinction until like a couple days

ago, but

even since then it's come up twice

already.

>> Interesting. Interesting.

Um

Yeah, and if you know, if you if you can

sort of get the same power of Cloud Code

through the the Co-work wrapper and has

a few of these bells and whistles that's

sort of causing people. Do you lose any

of the flexibility? Like do you have

more flexibility in Cloud Code to build

dashboards or is there anything that

Co-work is constraining constraining of?

>> Yeah, I for knowledge work per se, no.

But if you want to

like,

you know, some some of the things people

want to do is like build these like more

complex data scrapers

where you need like a heavier build of

code. I think that I I

I could be wrong on this, but basically

in coding there you you can like there's

these open-source libraries that's like,

"Oh, if I want to run a scraper, there's

like 5,000 people have created scraper

tools on GitHub open-source

and you could just go grab one. You

don't have to like rebuild the scraper,

right? Uh I don't think Co-work has the

ability to grab those packages

for the security reason. So I think the

more you want to reuse like more

traditional coding components,

then you'd want to shift to code, but

again, you'd have to really have more of

a coding use case than a research

knowledge work use case.

>> Yeah, so the develop like it's more more

developer-minded.

Person who's going to use Claude Code

and the more just end-user who's

working with files and building, you

know, PDFs and

uh is

Excel spreadsheets is more living in

Claude Co- Claude Co-work. But the

distinction distinction is is

compressing anyways from a community.

>> Definitely compressing.

>> Yeah, okay.

>> And I've got some pretty technical

clients that know how to use code and

they're just like,

"I can't be bothered. I just use

Co-work. It just it does what I need it

to do." And I to be honest, I'm falling

more and more in that category. Like,

let me use my mouse more, show me more

things on the screen.

Um and then just today they announced

that

they're going to improve the mobile

capability of Co-work. And so like the

schedule tasks

used to only run if your computer was

on.

>> Yeah.

>> And apparently now they're going to run

if your like laptops off. So they're

moving some of the those powers to the

cloud.

So the the announce it's not out yet,

but they announced it this morning.

>> Oh, interesting. Interesting. Okay,

great.

Uh any last any last thoughts on on on

Vibe Check? Like what are how are people

feeling about AI adoption in in finance

right now?

>> I think I I feel like I'm encountering a

little bit more

I wouldn't say frustration,

but I think the skepticism

level has gone up like

15% in the past quarter.

>> Interesting. What

What are What are the pushbacks? Yeah,

sorry. What are the pushbacks on that?

>> I think one is

the we're spending so much time

babysitting the AI when we could have

just done it ourselves. That's one.

And I feel like after 18 months or so,

they're like, "Oh, we thought we would

have like kind of left that period,

but we're still in it." I don't know if

you saw Glean had this report called

bots, they called bot sitting.

It's the invisible work in just like

updating your contacts, like checking

for hallucinations, reprompting, you

know, I think they It's like, I don't

know, it's like 30% of your AI time is

bot sitting, right? So, you're actually

[snorts] only 70% more effective. So, I

think there's a little bit of that.

And and then the cost the cost side of

things. Just like,

is this worth the money?

And I think,

you know, you're definitely seeing token

budgets, you know,

Co-work is great. It uses a lot of

tokens. I don't think it's I think to

make it so friendly to non-coders, it

probably,

you know, I saw the analogy is probably

taking a blowtorch to light a cigarette,

you know, a few times.

Uh

so, there's definitely

a little bit of that. And I think a few

people have said, "Look, like, they're

like, "Look, if I'm really honest, I

didn't notice a difference between Opus

4.5, 4.6, 4.8. I'm not sure if I'm going

to notice a difference with Fable.

I'm not sure if that's like a user

error, and I put myself in that category

as well, like I'm not pushing this hard

enough.

Or it's the reality that there is some

kind of plateau for the types of things

that we're doing, right? Synthesizing

documents, writing reports, analyzing

data. That's an open I I

No one's issuing a verdict on that, but

there's a little bit more like,

I'm really honest, I don't know if it's

really got if AI, like the models,

have gotten that much better in the past

3 months, 6 months.

>> Yeah. Yeah.

It's interesting. I think um

you know, the the the the public market

like

you know, most of the clients we work

with are you know, scaled public market

investors that have a tight coverage

area. And so chatbots were just like did

not hit

you know, effective product market fit

for that. They're great for generalist

firms that want to get up to speed and

do you know, research quickly.

I think many of my clients are probably

only in the last three to five months

like getting really serious about

implementation.

So they're probably later on that curve

than some of your clients who are more

in the private market space have been at

this a little bit little bit um

longer. This concept of a digital twin

sort of overlaying the entire process

like a digital analyst uh is something

that we've been we've been scoping

um and really was a little bit science

fiction until

>> Mhm.

>> very very very recently.

Um

So the vibes, you know, the vibes on

those conversations I think have gotten

quite exciting and maybe that hits the

same wall at some point. Um it's easy to

have hope and and build prototypes. It's

another to actually deploy these things

to to uh the effect of better

decision-making.

Um

>> Yeah.

>> that's the hill we will we will all work

to climb and hopefully bring in some

some more guests to uh help inform that

inform that climb.

>> Yeah.

I'm excited.

This is a good vibe check.

>> Good vibe check. Uh

great great great to connect K and we

will be back we'll be back next next

week with another guest and maybe next

month with another vibe check. We'll

we'll we'll sort of monitor the vibes

and bring those vibes back back to you.

So thanks everyone for us know in the

comments. Yeah, let us know in the

comments if what you think of these vibe

checks, what kind of things you want us

to cover in future ones.

>> That sounds great. Sounds great. All

right. See everyone soon.