Prak, Inc

Professional Everyman, AI Silent Partner for RIAs

Sign up for beta here:

[Prak, Inc @PraktikantAI]

#Phillies Baserunning

[RIABiz @RIABiz]

Bain Capital funds Envestnet buy of Vestmark to surpass $10 trillion and define 'adaptive modularity' strategy, leaving Orion a clear RIA lane https://t.co/Gx5aA4QW4O

[homanp @pelaseyed]

Thanks but no thanks

[Harrison Chase @hwchase17]

an increasingly hard part of agents is handling auth

does the agent act as itself, or on behalf of a user

we're trying to make this as easy as possible in managed deepagents!

[Prak, Inc @PraktikantAI]

how many cliecks does it take before your employees just set #Claude to "Dangerously Always Allow"?

[Rohan Paul @rohanpaul_ai]

New Microsoft paper. Long agent runs expose failures that short benchmarks miss. Agents can look reliable at 2 or 4 steps and fall apart by 16.

every agent step has some chance of going wrong, and those small errors compound as the workflow gets longer.

Across 9 models, success usually dropped as the number of dependent steps increased.

On ToolQA, models that were near-perfect on short runs fell to just 0-33% success by 16 steps.

Long context was not the main driver: shortening the context made the decline worse, so blindly trimming history is not a reliability fix.

For builders, the recommendation is straightforward: stop treating a benchmark pass rate as proof that an agent is production-ready.

Test agents at the workflow lengths you actually expect, measure per-step reliability, and add checks or checkpoints before a bad step poisons everything that follows.

[Tom Goodwin @tomfgoodwin]

I swear bots are patches on broken design.

We shouldn't need AI to do things that we could do with 3 button presses.

Perhaps bot traffic will force companies to make it easy to self serve, even with edge cases

I also get confused as to why you'd want AI to make a DMV booking, I'd rather look at options and ponder on what works best. "I don't mind waking early on Thurs, and need to be close after, whereas the week after I'll be stressed thinking about X"

[Prak, Inc @PraktikantAI]

@UseCorgi using millenials for cold outreach because they cost less than tokens.

[Josh Pigford @Shpigford]

partially filled out a form on @UseCorgi and in the 10 minutes since i've been texted, emailed, called twice and freaking video facetimed twice.

i assure you i won't complete anything and certainly will never use corgi.

[Prak, Inc @PraktikantAI]

i'm not not watching #DollyParton biographies on @eentertainment

[Rudy Havenstein, Senior Markets Commentator. @RudyHavenstein]

Pop quiz, hotshot.

What is the average and median 10-year U.S. Treasury yield since 1962?

5.81% and 5.41%.

Today 4.64%.

#WGS10YR

[Prak, Inc @PraktikantAI]

@MattLoszak Shapiro RN.

[Addepar @Addepar]

Most RIAs are experimenting with AI, but few have a real strategy behind implementation and adoption.

With Addepar, AI is built into the platform — powering data management, reporting and analysis. Because AI is native, rather than bolted on, firms get actionable intelligence without the strategic burden.

Learn more: https://t.co/SAnk3U7tge

[Olivia Moore @omooretweets]

Consumer agents have gotten so good in the past month

For the first time, I can imagine allowing AI to fully intermediate my email or calendar

Some massive incumbent interfaces are about to become disruptable

[Prak, Inc @PraktikantAI]

today's discovery revelation

[Tom Goodwin @tomfgoodwin]

A lot of people expecting AI to transform work fast don't get how messy most of it is.

It's 40 x 7 min tasks that involve working across 6 programs and 2 devices.

The change happens when we completely rethink how we work

[Bedrock Robotics @BedrockRobotics]

Tune in to @BloombergTV to watch @EdLudlow and Bedrock CEO @bsofman discuss a first-ever milestone for construction: three fully autonomous excavator deployments, happening concurrently on two active construction sites in TX and one in NV.

The momentum and demand for our autonomous systems is strong, and we’re proud to share this achievement with our team and partners just two years into this journey. The future is bright for this industry, and for an America that relies on construction to build all it needs.

[Prak, Inc @PraktikantAI]

@MartinGTobias i respectfully disagree. the base docs, yes, have core tenets that get repeated as step 1 of 15. But the application and execution of a good estate plan (steps 2-15+n) is unique to each family. This is where contextuall customization must occur-not ai standardization

[cinesthetic. @TheCinesthetic]

What TV show intro is so good you NEVER press 'Skip Intro'?

[Gregory Blotnick @gregoryblotnick]

"Operating Leverage" (Mauboussin)

I would tell my 23 year old self, "spend the next 6-12 months on this topic and nothing else, until u have become fluent in this math."

1/6

fluency may be overstating it, there are just odd exponential effects at work here that are hard for our brains to grasp

sort of like compound interest, we just have a very hard time with non-linear growth rates, and here, u have many of them colliding with one another

the core math:

rev dollars split into px/vol, how those flow thru various cost structures (fixed vs. variable), incremental margins + quickly triangulating $1 of revs --> how many cents of ebit --> how many cents of eps

and then layer financial leverage onto operating leverage -- the way a chg in EBITDA affects equity value on a net cash B/S vs a 4x levered one

study all this until the dots start to align, its way more important than twtr celebrity gossip

"ooo leopolds wedding...ooo down to $10B AUM" like dog what sorority kitchen is this

but the combo of op lvg + fin lvg, the lollapalooza effect, is prob the most important shit to understand tbh in all of L/S equity

there's always spare $$ laying around, its always mismodeled, there's a sustainable source of edge in unit economics + fixed cost absorption + the fact that most equity guys don't really understand cap structure

2.

because op lvg + financial leverage...sales growth of 4% vs 5% can knock 1/3rd off the mkt cap in a highly-levered + high DOL cost structure

1% revs --> 10% EBITDA --> 15% EPS...capitalize the lost EBITDA @ 8x or wherever it trades, to get the delta in EV, and remember that the debt burden doesn't change...the mkt cap does.

say $250m debt + $250m equity = total capitalization

the lost EBITDA @ 8x, say $125m of EV

that $250m of debt is still $250m...those r contractual claims, they dont move with earnings, equity is residual and equity is guaranteed nada

so that $125m in EV comes right off the $250m in mkt cap.

bang, 50% equity decline

3.

u learn this shit by building a billion models and just tinkering with px/vol + cogs/opex items, and watching what happens to ebit and eps

lots of consumer, industrials, tmt, this is the math ur doing

a cruise ship, an airline, a semiconductor co, Mattress Firm...almost all these costs are fixed, the "factory" must run regardless of units moved

and u add in that these are cyclical as fuck, so u have a 3x lollapalooza at econonomic inflection points

sales up 20%, costs stay the same, ebitda up 50%, EPS 60%, levered out the ass so the levered equity stub 2-3x's in a month

so if I was coaching my 23 year old self, I'd say pull up 4 LLMs, chatgpt, claude, grok, gemini, and make ur way thru this mauboussin piece line by line, hitting LLMs until u understand everything

4.

and then after that, build 10 models of the nastiest high-fixed cost businesses u can think of, historicals going back 20 years, and study the way the quarterly P&L shifts around cyclical peaks and troughs

do a few shitco retailers, a casino, a hotel, an iron ore smelter, an offshore driller, a theme park, and hell, u may as well throw in micron

"why does this trade at 5x eps"

bro what do u think these companies do lol

"AI technology"

no. they make widgets

these are all widget companies

go stare at the quarterly P&L from 3-4 years ago, and ask grok if its good when u print a negative 30% gross margin

and then study what happens when negative gross profit dollars hit a fixed cost base

good intro to operating deleverage

5.

u know u are in high DOL shitco paradise when ur just staring at a P&L, like, "how did sales go down 5% and EPS went from $2 dollars to negative $12?"

but I'm saying, this is where all the $ is, always...a shitco that printed x margin 20 years ago, can print that same margin today, and its never in ests despite that its the exact same business.

its all factories, boxes, stores, a pile of heavy assets...a dollar of revenue is gonna be damn near a dollar of net profit, in both directions, because the costs dont move

every co has new fancy multiple-accretive recurring revenue stability streams to reduce the lumpiness of earnings

and then it doesn't snow in Colorado, Vail's revenue goes down 3% and their EBITDA goes down 40%

and u realize that u overpaid for a levered shitco multiple arbitrage roll-up, and that ski resorts have always been shitty low-multiple businesses

this is what buffett talks about with "when good mgmt meets shitty economics, the economics always win"

"our revenue is high-quality, because of our new loyalty program which grew 30% y/y"

like...u sell furniture over the internet

or the economy rolls over, the loyalty program isnt so loyal, Vegas traffic and casino revenues get cut in half, and EPS goes negative

the operating leverage inherent in a business model is not something that changes, thats why this has to be mastered and is worth the time

6.

go stare at HTZ financials. play around with tiny changes in capacity utilization or pricing, and u will see why it chapters out every 5 years

or AMC, see what a dollar of revenue lollapalooza's itself into. u can prob trace 1 movie ticket --> $10m of market cap

thats what happens when 1 rev dollar = 1 net profit dollar, and when current net profit = 1

1 + 1 = 2

aka, your earnings doubled

if it weren't for that torque, where the equity is effectively just a call option, the mkt cap of this should be about $2B lower (current: $2.3B)

"wait, how can an enterprise have market cap and be worthless at the same time?"

u will learn, my friend

its called

"schrodinger's shitco"

END

if I find other reading on this I'll post it but I don't think u can read ur way thru this, this is 100% modeling intensity

just staring at cogs/opex items trying to figure out wtf all this shit is and how it moves around

labor, R&D, are these fixed, are these variable, what about stock based comp, what about depreciation, how are each of these connected to revenue dollars, assuming they are at all, and why do people care about shit that isnt on the income statement like "inventories?"

after 10 or so years, u will arrive at the truth

"wow, 99% of GAAP accounting is complete fucking fiction."

and thats when u say, u know what, I'm just gonna find stocks that are already going up and buy the ones with sweet sounding names

50% long "SharkNinja" and 50% long SpaceX into the lock-up expiry

buffett spoke of this

good luck modeling

https://t.co/M1kbbBqXHd

[Prak, Inc @PraktikantAI]

Bessent dropping facts from the Top Rope.

[Treasury Secretary Scott Bessent @SecScottBessent]

One of the highlights of the Warsh Fed has been watching stenographers posing as journalists, like the WSJ’s Nick Timiraos, reduced to reporting Fed backroom gossip because they’re incapable of performing real economic or monetary policy analysis without being spoon-fed.