AI Agents Driving Business Savings
Key Points
- Amazon’s internal AI assistant “Q” automated Java‑17 upgrades, saving the company an estimated $260 million and about 4,500 developer‑years, illustrating how agentic workflows can create huge efficiency gains at scale.
- These developer‑focused savings highlight a broader trend: AI‑driven automation can free up engineering time for higher‑value work, though quantifying the impact on the bottom line remains a challenge.
- In the legal sector, Spellbook’s new “Spellbook Associate” AI agent is designed to handle complex, multi‑document matters, demonstrating the need for agentic, not just query‑based, workflows to make large‑scale AI usage practical.
- The rapid launch of competing products, such as Harvey’s AI platform touting 70 % year‑long user retention, shows how fierce competition is driving rapid adoption and refinement of AI agents across industries.
Full Transcript
# AI Agents Driving Business Savings **Source:** [https://www.youtube.com/watch?v=7qMWTXNNdOI](https://www.youtube.com/watch?v=7qMWTXNNdOI) **Duration:** 00:09:08 ## Summary - Amazon’s internal AI assistant “Q” automated Java‑17 upgrades, saving the company an estimated $260 million and about 4,500 developer‑years, illustrating how agentic workflows can create huge efficiency gains at scale. - These developer‑focused savings highlight a broader trend: AI‑driven automation can free up engineering time for higher‑value work, though quantifying the impact on the bottom line remains a challenge. - In the legal sector, Spellbook’s new “Spellbook Associate” AI agent is designed to handle complex, multi‑document matters, demonstrating the need for agentic, not just query‑based, workflows to make large‑scale AI usage practical. - The rapid launch of competing products, such as Harvey’s AI platform touting 70 % year‑long user retention, shows how fierce competition is driving rapid adoption and refinement of AI agents across industries. ## Sections - [00:00:00](https://www.youtube.com/watch?v=7qMWTXNNdOI&t=0s) **AI Automates Java Upgrades** - Amazon’s internal AI assistant, Amazon Q, automates Java 17 upgrades, saving the company an estimated $260 million and 4,500 developer years by turning a months‑long, low‑value task into a minutes‑long process. ## Full Transcript
so we're going to cover four different
AI use cases that are all monetizing
right now and I'm going to break them
down and show how there's a common
pattern here and I'll be really curious
to hear what you think so number one
this is in the developer side of things
developer workflows CEO of Amazon Andy
Jassie tweeted that Amazon had saved an
estimated $260 million in efficiency
gains because they automated almost all
of their upgrades to Java 17 with Amazon
Q which is their internal AI assistant
and so instead of something that would
previously have taken a
developer months and months to do on
large systems like Amazon not hard Q
just does it in a few
minutes and it's also not fun for
developers I have never met a developer
who enjoys upgrading their Java it's
it's got to be done there's like a
security issue if you don't do it but
it's also not something that adds
functionality it's also not something
that
you can really put on the resume like
nobody likes doing it now Q does it
jasse estimated that Amazon saved
4500 developer years as a result in
other words if you measure one year as
like a developer working for a year they
save 4,500 of those just by implementing
q and having Q do Java upgrades now
obviously a startup is not going to have
the same kind of savings because they
don't have the same kind of scale but I
think I think it does illustrate how you
can get agentic workflows for developers
into place and realize value very
quickly and I've been watching to see
when will publicly traded companies
start to talk about how AI is driving
their bottom line and this is one of the
first statements I've seen that is sort
of in that direction now it's not
actually talking about how Q is helping
Drive the Top Line it's not talking
about Q directly saving dollars and
cents this is really efficiency gain so
it's trading out the time that
developers would have spent on you know
manually moving Java 8 to Java 17 and
actually putting that time into better
use it's still real savings it's just
not savings that's going to necessarily
appear on the bottom line and I think
that's a higher bar and we need to keep
waiting for that to see when AI is going
to be accredited for that but
nonetheless it's a big deal all right
number
two this one's in law and there's two of
them actually wrapped inside this so
Spellbook is releasing Spellbook
associate which is an AI agent for Law
and that's big because you need AI
agentic workflows to work through big
multi-document legal matters otherwise
you're just asking the llm over and over
again and I've been saying for a long
time just asking the llm is cognitively
expensive it's not going to be easy to
do it won't last we're starting to see
other workflows now competition is super
tight in AI spaces so as soon as
Spellbook associate was announced the
other big competitor in the space Harvey
which also uses AI released a press
release talking about how high their
user retention is 70% over a year how
they have growing uses usage by firms
who pick up Harvey and during the year
use it more and more as they come to
trust it so we'll see what Harvey
actually releases this very much feels
like a defensive press release to me but
I think the fact that they felt they had
to release something just to punch back
at Spellbook suggests that they're a
little bit worried about the power of
agent-based workflows they're worried
about a lawyer being able to tell
Spellbook associate exactly what they
would tell any other associate and go
have them run down multi-document
research and come back and give them an
overall assessment and approach on the
case does that mean the associate is
always right no the human associate
isn't right either all the time so
there's going to be a tolerance for
error here that is probably scary if
you're Harvey and you're a competitor
underlines how fast the world is
Shifting I think we're going to see more
and more of this move from a ask the llm
type software solution to a let the
agent do it type software solution be
really curious to see how that goes all
right that's number two number three is
in sales this is actually a startup I'm
really curious what you think of it
clay.com
has data enrichment uh that they have
automated for sales leads across 75
different data sources and then they
will also automate the Outreach for you
on top of that so all you bring is like
a list of email addresses and they will
take care of turning that list of email
addresses into a complete verified
profile and they will also make sure
that you are not being charged saslik
fees across all 75 of these sources you
have sort of a tokenized pay as you go
system what's interesting about this to
me is that there is some AI there
there's a large language model element
to that Outreach for sure but there's
also just a traditional bundle and save
play and it's reminding me again there
is money to be printed in these spaces
if you are combining smart AI use cases
with ordinary best practice business
value that we've been able to build for
a long time bundling together a bunch of
different services that would
individually cost a lot and making sure
customers can get access to all of them
and save is as old as TV bundling like
we've had that for a long time it's
probably older than that the point is
it's not particularly new it doesn't
take Ai and it still works really well
because it was a good idea that solved a
real problem we'll see more of those two
all right the last one I want to call
out is that perplexity is starting to
rumor their plans for monetization this
hit I think it was CNBC and they are
charging a lot so so Ju Just for the
like background you can charge a couple
of bucks for display Impressions right
like $2 cpms Are Not
Unusual they're charging 50 they're
charging
$50 for a
CPM for search appearances in
perplexity we don't really know what ads
in llms look like they have been demoed
they look in context like part of the
answer as far as I've seen
I think that people are making the case
if you work in in sales at those
organizations if you work in sales at
perplexity or sales at open AI that if
it appears in context it's going to be
more powerful and influential and
therefore justifies the price maybe
we'll see but I tell you what can you
imagine the impact on the market if the
market is suddenly willing to pay $50
cpms wild I I have no idea if they're
going to be able to pull that off uh but
just seeing the price point is reminding
me that AI is not free AI is going to
monetize and wow the numbers are are
popping and that leads me to sort of my
last reflection these agent-based
workflows are not really designed for
the software pricing model that we have
today the pricing model we have today is
really like you have a person they can
do a job they can do their job with your
software
done what we have coming is have the
agent do it for you and that's great
it's almost like a virtual employee
which I know that there was a big
release about um just a few months ago I
think it was bamboo that did the virtual
employee
and they talk about the space for that
in an HR System like bamboo but they
obviously hadn't built it and now that
we start to see it it's reminding me
that we haven't priced it well I don't
know what you charge for that because
the savings is tremendous pretty much
whatever you charge and we're starting
to see some eye popping numbers because
the savings is so high so even with Clay
which is not just AI it's also just
bundling together manual research
hours it's and this is the uh the sales
one I talked about
earlier it is charging hundreds of
dollars a month because that's vastly
cheaper than paying someone to do it on
a monthly
basis and so we're going to start to see
some big numbers we're going to start to
see software subscription
Topline Revenue numbers that we haven't
seen in a long time maybe ever because
what they're going after is not
replacing the software
economy which is in the hundreds of
billions of range but they're eating
into the costs that business allocates
for compensation right for for paychecks
and that's in the trillions in fact I
think the global estimate is $10
trillion and so you're going to start to
see really eye popping numbers if this
catches on and agents are able to be
priced correctly ly and I'll be really
curious to see how people do it what did
I miss what did you think you're going
to see for agent based workflows