Latest / The Punch List / The Honest Truth About AI in Business Right Now

Transcript

The Honest Truth About AI in Business Right Now · The Punch List

0:00
Construction is probably one of the most complex endeavors we do as human beings.
0:07
There used to be 30 to 50 startups in construction tech popping up every year. That
0:12
number has tripled up. What can you really do as a small, medium sized firm?
0:20
I am less bullish on this having a positive impact on how we do business in the real world
0:26
business, but I am very bullish on AI companies making a lot of money and continuing to grow
0:32
because as a builder, I care about building a building, not buying a software. And we're
0:38
back with the number one podcast for architecture, engineering, construction in the entire world.
0:45
The Punch List podcast with myself, John Wright, Cone of and TC.
0:52
How's everybody doing? Good. It's the good, busy day of the week. Bummed out. Back to back
0:59
meetings. Let's hit it, guys. Yeah, well, we've got a couple topics today. Uh, number one, AI
1:06
demo burnout. Is that happening? We're hearing from our friends in the construction in the
1:12
architecture world that they're getting bombarded with, uh, pitches and demos. We're going to talk
1:19
about that. We're going to talk about what you should do if you are in a construction,
1:25
development, engineering or architecture firm right now to prepare for wider adoption of AI.
1:30
What's that look like for you? What can you do now? Uh, you know, without a big budget
1:37
three AI hype cycle where maximum AI hype cycle probably, or approaching
1:44
it. What are some of the real gains that you know, we're experiencing as business owners and the
1:49
industry is experiencing from AI and then forth? Uh, cybersecurity is on the rise. What does it mean
1:55
for the industry? There's an article that was hacked recently. So we're going to talk a little
2:00
bit about that. So let's start with let's start with demos and um think of and TK
2:07
you're probably uniquely positioned here because you're, you've got an AI startup targeting
2:13
construction industry. What what are you hearing out there. Yeah, I mean, I can start that first, uh,
2:21
and then the first part is they used to be 30 to
2:28
50 startups in construction tech popping up every year. Back in the day. This was like
2:35
2010 till 2021, I think 22 and three onwards that
2:41
number has tripled up. We're seeing like 151, 80 startups popping up every year.
2:48
Uh, there's three times of more startups coming in. The industry is one is one
2:55
angle. I think the second angle is. Anyone
3:03
can feel comfortable about AI helping in construction, so let's assume
3:10
someone doesn't know how to estimate cost of construction. They pop it up in AI, and AI
3:18
pops out some estimates. The person who is not skilled in that field will say, oh,
3:24
that sounds like a good output. You can estimate the cost is breaking down into all the
3:30
subdivisions and concrete and masonry. Oh, that's that's good stuff. But if you show it to an expert,
3:36
they're like, what the heck is this? This is not good. This is not a good output. You need to churn
3:41
it more. You need to improve it more. It's got the wrong assumptions. It's got, you know, ten things
3:46
wrong. So there's a lot of unfinished product churning out in the industry
3:53
and when on the consuming side of it.
4:00
Suddenly we are seeing construction companies, architecture firms, engineering firms
4:08
getting inbounds from like ten, 12 startups a week, doing like 3 or 4 demos a month.
4:15
Uh, they, they get really excited about the promise. But when they try out their
4:22
own real data, their real drawings, their real spec documents, the real project stuff into it,
4:29
it just starts to fail. They they think, you know what is not there.
4:36
Six months later, they try it again. Again. They they put the documents in, it doesn't
4:42
work. And then they get burned out. And then and then they start to believe AI doesn't work. AI is
4:49
full of crap. So that's the that's the overarching I'm oversimplifying the story for for all of
4:56
us, but that's the Overarching arc. And I empathize with them. You know
5:03
you are. You know, I came from healthcare. And healthcare is
5:10
one of the best industries for vendors, software vendors because healthcare cost is very high in
5:16
the US, which is bad for us as consumers or people who are paying for it. But
5:23
it is high because of the administrative spend, which is there a lot of administrators sitting in
5:28
not doing the actual job of of coding a patient? What is the job? The job is to evaluate software.
5:34
The job is to architect processes is always to make them more efficient. So there are people who
5:41
have full time jobs to evaluate software in healthcare, in construction, in in development and
5:48
architecture engineering. They're very lean admin teams. They don't have enough administrators. So
5:54
they are taking out their crucial time from actual projects and trying out the software, and
6:00
at the end they get they get burned out by by the software not working out. So I empathize with them.
6:06
When you hear from them, then they're getting burned out by AI not working. But it is not the AI
6:13
not working. It is a bunch of software vendors who are not putting the right
6:19
rigor and discipline on AI not working, and or they don't have access
6:26
to the real data to even put that discipline that will AI work or not. So vendors who are
6:33
handicapped with the right data, or they are handicapped with the right rigor of discipline,
6:38
are burning the market out. And that's what we're heading over to you. Yeah.
6:45
I mean, the only point I would add to this is, I mean, there's a. Famous psychological concept,
6:52
right? Which is a illusion of explanatory depth which which basically means, you know, a little bit
6:59
about a topic and then you feel like, oh, I know this, right. And construction is one of those
7:05
industries. As a, as a observer looking at software people in general, I'll categorize them as one
7:11
people as I talk to a lot of them as I'm on the startup side, I, I see this again in again
7:18
tendency to understand few things and feel like, oh, I, I understand how construction works. I know
7:24
what exactly the market needs. And then they rush to building a product and shipping it. Right.
7:30
Whereas, I mean, I say this again and again, construction is probably one of the most complex
7:35
endeavors we do as human beings. Every building has every form of engineering in it,
7:42
like it's got mechanical, it's got electrical, it's got software engineering. I mean, John, last
7:47
building you and I did, we had great outcome, except the software that managed the building
7:52
management system was in operating properly. And we had to hear a earful from our client. So. So
7:59
every branch is involved in a building. All that means is I just want companies as
8:06
they're coming in and building tools for this industry, to have that humility and take the time
8:11
to understand how a building really works and what the people building the building need.
8:16
Because as a builder, I care about building a building not, you know, buying a software, so to say.
8:23
Um, and I will not all put it on the software guys side as well. I see it on the other side too, where
8:29
construction guys play around with Claude and think like they are AI experts now. So and they
8:36
start, you know, building their own tools, whereas maintaining a software tool have it working for
8:42
you long term takes effort in some expert involvement too. So a little bit on the both sides.
8:47
But but I'm more seeing just people looking at the industry outside and oversimplifying the
8:54
problem and shipping out a product too early. John, what's your take? You have been a real estate
8:59
developer. I'm sure you're getting a bunch of inbounds from this software. Yeah, experts? Yeah. I
9:06
mean, I have two minds about this. One is it's exciting and it's exciting that, um,
9:13
AI is is generating this type of enthusiasm and interest to solve problems in our industry. And
9:20
so I'm probably an outlier in the sense that I will take those demos just to see.
9:28
Um, I would say, uh, a couple things that I'm noticing. One is I think the
9:34
current startup companies are tackling kind of the low hanging fruit first. So managing an
9:41
acquisition process, um, site search diligence, aggregating
9:48
public data around parcels and information around cities That's, you know, helpful and
9:55
usable early in the process of identifying property to buy and making a decision to buy it
10:00
and and what to put on there. And look those offerings have been around even pre AI. I think
10:07
they're just getting better and faster. Um, and I'm not dismissing them as
10:14
unhelpful, but I'm finding increasingly I'm looking at these demos and
10:20
just literally thinking I can, you know, with the pace at which the labs are moving with the models,
10:27
I can do this by myself without actually a lot of pain. Um,
10:34
unless, to your point, unless they have a proprietary data set that's really
10:41
hard to aggregate, like public data is is public data, but it's not easy
10:48
to get and and easy to get in a clean version that's readable by agents. And so if they have
10:54
that, I'll pay a little more attention because I know that they have gotten domain experts, civil
11:00
engineers, attorneys to do that and pull that together. Or I'm talking civil in the context of
11:06
like, most of the stuff I'm seeing is like site search and diligence. Um, so that's worth looking
11:13
at. But the actual layers on top of like a parcel that helped me make a
11:19
decision. I'm finding I can do that pretty simply with a Claude Mack subscription. And I'm
11:26
non-technical. So am I getting a little burned out with that? Yeah, I am, I, I'm actually feeling
11:33
sorry for the teams that are putting a lot of effort into putting this together and putting the
11:38
software together, because I'm seeing the, you know, the kind of like models catch up to them really
11:42
fast. So, uh, that's kind of that's the state of the market right now.
11:49
Um, and I can see setting in an architecture firm, an engineering firm, a construction firm where you
11:56
don't have, as kind of said, like people whose sole job it is to evaluate
12:02
regulated software. Regulated industry software. Uh, you're doing that as the project manager or
12:09
the project captain or the architect or whatever. And so you start to get a little bit picky
12:16
about, uh, how, you know, am I going to spend my time on this other startup that wants to show me
12:22
something? Even taking the demo call, zoom is kind of becoming like, uh, send me the
12:29
login. I'll tell you whether I think it's a good idea or not. Yeah, I think that's
12:35
absolutely right. And I think the contrarian view I have for
12:42
or some feedback I have for. These firms right is.
12:51
AI is not deterministic. AI will not give you a binary answer. I will give you a
12:58
gray answer which is not white and black, right? So I get it. It's tough to evaluate
13:05
AI where it's gray and not white or black. Uh, but just because
13:12
of that reason, you need to put more time and effort in building some
13:19
kind of evaluation tests to evaluate these vendors, because the more
13:26
you know, I get this, uh, why is there a burnout? Right. Let's let's take a real scenario.
13:33
Let's say a company comes in, a software AI company comes in and says to a construction firm,
13:40
hey, I can I can do X for you, I can do I can cost estimation or I can do design
13:46
reviews or. Streamline your RFI. Modifies whatever. Right. I can do X for you. Great. They give
13:53
you a nice, cute demo. You love the demo. They're like, oh, this demo is amazing. Obviously it's a
13:58
canned demo. Like right now the next step is either
14:05
you get it through the wringer of evaluation within your phone, like you show it to ten more
14:10
people. You get approvals from your CFO, you negotiate on the pricing, you do all the
14:17
legal stuff and do all that hassle. So then try out
14:24
on one real project and it all fails. And then you're like, damn, I mean, what did I do,
14:31
right? So you went through the four months of ringer to evaluate them, get them in, and then you
14:36
realize it failed. And that's one scenario. Second scenario is you do a quick
14:43
evaluation. You say, hey, can you do like a one month evaluation? Can you try on a real project? If
14:49
you do that, you fail within one month. Great. But what if you already had
14:57
some real documents? Let's say in the case of rfe RL, it can help you,
15:05
uh, file RF files. You are a construction company. It can help you file. RFE is pretty fast. Or draft.
15:11
RFE is pretty fast. Great. Well, what's a good test? Here is a drawing set. Here is a spec
15:17
document. A draft RFE is for me. I know all the answer keys. Now,
15:24
one might say it's. It's not confidential and you can't pass it on to a vendor. Yeah,
15:31
let's redact it. Let's remove all the all the confidential stuff from there. It's a one time job.
15:38
You have a drawing certain respect document. Maybe two, maybe three. And anyone that comes in and says
15:44
I can make RFE. Like, really? Can you like. Here is a set of two documents show it to me,
15:52
like they will show it to you in a day or two if they show it to you. If they take more time to
15:56
show it to you, then obviously it's not AI, right? So that can really streamline your process of
16:02
evaluation within two days versus one month versus four months. So the key is do you have
16:09
enough test with you that you can give it to those vendors? And that's that's the way to
16:16
get less burned out over time. So okay
16:25
we're back to the scenario of the lean team. You're what do what do
16:31
construction development real estate development groups, engineering firms, architects do now
16:38
to prepare for this wider adoption? If it doesn't mean buying the next SaaS, uh, let's set aside
16:45
giving everybody a clot or an open AI account. Or maybe you should do that. What can you really
16:51
do? As a small, medium sized firm in the real world to prepare for AI adoption?
16:59
Yeah, I think there's three things. Whenever a new wave of
17:05
technology is coming, you should always do three things one. Swim in it.
17:13
You know, use AI or self like use cloud OpenAI yourself. Because once you
17:20
swim in that technology, you kind of know how the waters are, you know, how much can it do where it
17:26
fails, where it stretches, you know its limitations. So I always recommend people to swim in that, that
17:32
technology and experience experience it themselves. That's the most important thing.
17:39
Once you swim in that technology, you know its limitations. You build a test
17:46
like around that. You know where. What's an easy test? What's a tough test? What's a barium test?
17:52
So make a nice, tough test. Uh, to evaluate vendors very fast. Right? And
17:59
then third. And the most important one is AI is as good as the data and the context it gets.
18:08
The least you can do is organize your own data using AI. I mean, I see
18:16
construction companies having thousands of spreadsheets and word documents and PDFs sitting
18:21
in one project folder, and everyone's sharing the link on Teams and slack about which is the latest
18:28
drawing, which is the latest spec, which is the latest, you know, uh, cost estimate. But
18:35
once the project is done, you ask the third person to go in and look at that project folder. They
18:40
will have no idea. You don't give them slack. You don't give them teams. They will have no idea what
18:45
what folder, what file is what. So the least you can do is organize your own data and say, hey, this
18:52
is the latest spec. This is the latest drawing. This is the latest estimate. This is the latest,
18:57
you know, uh, budget. Like if you do all that, you are preparing yourself to
19:04
use AI down the road. Yeah. I've, I've never been more excited
19:11
or motivated to document processes, thinking, decision analysis
19:18
as I have today because I know that is going to go somewhere and live somewhere
19:25
and be useful in the future, both for my own work and for the future of the business and
19:31
organization that I'm creating now, I, I find that if you're to to amplify
19:38
that, if you're sitting in a firm right now wondering what to do to prepare for wider
19:44
adoption, the the the action you can take right now that's agnostic to
19:50
whichever model takes the lead next week, whichever harness, whether it's cloud code, whether
19:55
it's crockpot Codex, takes the lead or goes away is reducing the,
20:03
you know, your firm's knowledge, moat and analysis and analytical framework that
20:10
puts it ahead and and is its edge reducing that to a a
20:17
written record that can be read by computers and agents. And it's not to your
20:23
point a set a dry file with 1000 files that we hand over to people when we try to
20:30
do transitions for projects, it's just that's worthless. It needs AI now gives you the ability
20:36
to take all of that, summarize it, contextualize it, build that transition memo. You know,
20:43
we'd all I can remember, like people would roll off a project, or you would hand over a project to
20:49
an owner or a client, and you'd write this long transition memo, and that takes forever. And you
20:54
wonder, whatever comes of that, uh, sometimes you doesn't get done. You know, AI can now help you
21:01
do that. Yeah. Um, and then when they when you do that and wiki, it ties your data.
21:08
That becomes the power of the gains you'll get from that, the returns you'll get from that,
21:14
from, you know, plugging that into a model with a harness is exponential. Um, so and
21:21
it doesn't matter which SaaS vendors coming to you with the new whiz bang or which model gets
21:27
better, that is going to help you in the future. Yeah, there's only one question there, which is
21:34
AI is not a magic wand. So be humble about. So I'll give you one example.
21:41
I asked AI last year. It's been 12 months now. So the the model was pretty old. And I say last
21:48
year you have my emails, you have my calendar, you have all the stuff.
21:55
Draft a travel policy for me, like how do I travel? What are my
22:02
choices? What is my what are my loyalty numbers and everything to give it to my to my admin.
22:09
Now, it did a good job doing the research, but I had to still review that document. The output it
22:15
gave me refine it a bit. But that small refinement that I did was important. But once
22:22
refined, then I'm sure there will be an external AI. Right now I can trust on booking
22:29
my flights, but there will be an as soon I can trust on booking my flights and I will give that
22:34
dog. I will point that document to that AI, right? So you still need a small human in the loop?
22:41
Uh, but at least the effort was not that three page document I would have
22:47
never written myself if I would not have done the first pass. Exactly. Then
22:54
your time is is is refining and adding and checking and editing. Yeah. I mean, one my
23:01
thought on this would be, I mean, to me, the buying decision, whether you're buying an AI tool, I don't
23:07
think it's very different from how we used to gauge softwares before, which is the approach
23:13
always needs to be processed first. I mean, at least that's what I thought always worked well
23:17
for me. And when I say process was just kind of diving a little deeper there, there is an
23:24
outcome we want. Whatever we are doing, there is an outcome we want. We pull back from it and say,
23:29
these are the steps and processes I'm taking, whatever tool I'm buying. Is it making that
23:35
efficient or is it leapfrogging something? Right. Like when we talk about takeoff, am I going to
23:40
leapfrog doing the takeoff or am I going to make my take off faster. The
23:47
biggest pitfall I see is, is the. Is that like that thing we heard in that documentary Supersize Me,
23:54
right? Like, oh, for $0.50 more, I'm gonna give you double the chocolate shake or whatever, right? Do
23:59
you need that much chocolate? Like, do you need. Do I need 1000 other documents being summarized and
24:05
fed to me too, for me to do my job? Is that something I'm adding on? Never let tools die. Drive
24:11
the process. I think that's a big pitfall I have seen in construction. And I'll give you a very
24:17
specific example. Not taking software companies names, but we keep talking about how construction
24:24
efficiency has fallen compared to where it was in the 40s. Well, technology has improved. How the hell
24:31
did it fall? Like, how did that even happen? And I think one of the core reasons is we have let
24:37
tools drive some of our processes, and we are copy pasting a lot because computers
24:44
allow us to do that rather than thinking through the details as we're drawing it. I mean, that's one
24:49
of the biggest culprits of why I think efficiency is drop. I just hope to God that AI doesn't become
24:55
a contributor to that. And this time it's different. Like, we got to have a process first
25:01
thinking this time around, we just got to. The question is, do you when you implement an AI tool,
25:07
are you just doing your existing process faster but yet still keeping those
25:14
steps? Or is there an opportunity to reduce, eliminate, streamline or refinement. Or leapfrog
25:21
it? Yeah. Or leapfrog it and leapfrog it? Yeah. If you're just using it to do your existing process
25:27
faster, is that actually getting you what you want or or is it really an opportunity to do
25:34
something completely different? And that's kind of the I would say after you reduce your whole
25:40
company moat and knowledge mode and process mode. To writing then you. The next
25:47
step is evaluating all your current processes. Yeah. I mean, step one is knowing your
25:54
company process. What what has happened in construction? Because of fragmentation, we have
25:59
divided our processes into so many smaller chunks that the person making the buying decision really
26:04
needs to make the effort of understanding the process end to end before you, before you go make
26:10
a buying decision first of all. Uh, second thing is, yeah, you got to categorize a tool by, like you
26:17
said, John, is it making things efficient or is it allowing me to leapfrog and then factor in what
26:22
kind of was saying? It is not deterministic. If I'm putting a GMP together, I want deterministic
26:29
tools. I cannot, you know, because I have accurate piece of drawings. I need to stand behind it.
26:36
If I'm doing a conceptual estimate, it doesn't have to be deterministic, because guess what? I'm
26:41
doing this anyways, you know, like it's a lot. A lot of conceptual estimating. Is that so? So can AI.
26:48
Help me there. Yeah. By far and quite a lot like on conceptual drawings. Putting an estimate together,
26:53
putting a schedule together. I think AI would would do a good job. Obviously a human being still
26:59
needs to review it. But if you're doing a GMP, I don't know. Take a closer look at what AI is
27:04
giving you. I'll give you a good framework. Right. What is the risk profile of a job,
27:12
and how much of that job is maintained versus not? If the job
27:18
is a lot mundane and the risk profile is low, AI is the best bet for you right now.
27:25
Immediately. Like if you have to read spec documents cover to cover, that's 1000 pages
27:32
of mundane ness. But if you if the risk profile of reading that document for some for getting
27:39
something out is is low. Like air is the best bet for that, right? Because
27:46
humans will make more mistakes in cleaning that mundane thing than AI will. And the risk profile
27:51
is low anyway. So that's the easiest low hanging fruit for AI to help you out.
28:01
So if we're in, you know, either in or approaching the height
28:08
of the AI hype cycle. Yeah, at least this current version of the AI hype cycle,
28:15
this this initial version of this first version of the AI hype cycle, where are
28:22
you seeing real gains in the industry? Now concrete feedback.
28:29
What are you seeing? First, let's define the hype cycle. Okay so. Although
28:36
I'm biased and I'm I am an investor in anthropic but At the same time, I will say
28:44
the hype has started because anthropic.
28:51
S1 is saying there their dam is $32 trillion. It's more than the
28:58
US GDP. That's what anthropic is saying. That time is time is a total addressable
29:05
market, which is how much can they eat? Anthropic thinks they can eat up to $32 trillion.
29:13
I call it BS and it's a big hype. I'm not saying it's a it's not a
29:20
good company. It's it's an amazing company. It's going to do wonders. And that's why I put my money
29:27
behind them. But I am biased to it as well. So fii but it's
29:34
not it's not $32 trillion. That's a hype. Uh, so.
29:41
Let me ask you guys, have you seen any out? There are AI producers and AI
29:47
consumers, the people who are producing AI and Nvidia, all the infrastructure, semiconductor layer,
29:54
data center, construction layer, uh, the semiconductor layer, the the racks, the
30:00
GPUs, the, the AI model companies, those those are all the AI app companies. They're all AI
30:07
producers. And then there are AI consumers who are not producers, who are just consuming it. Have
30:14
you seen any AI consumers having any real dollar gains?
30:21
I don't know from all from people I know. I mean, and I'm talking more construction industry. I
30:27
would say no. Not measurable. Outside of hey, we are building a data
30:33
center. So that's going to obviously construction companies are making money more for from building
30:39
the physical infrastructure required to support AI. So yes, there. But I
30:46
don't think that's the question I think. Kind of just to clarify, the question you're asking is if
30:51
I am using AI in my daily life today, can I increase my margin?
30:58
Am I doing something I would need 20 people for? With ten people?
31:06
I would say no. If you remove all and I don't have a chart will be right now, but I can get it up
31:13
in some time. If you remove all the AI producers from the S&P 500 list and look at
31:20
how much net margin has been increased, it's going down. If we
31:27
look at what is the sales growth increase, it's going down or flat. So if your sales
31:34
growth is flat, if your net margin is not improving, like what the heck are we talking about
31:39
here in terms of dollar gains. Yeah, I mean, I think that's true for
31:46
the large enterprises. Um, I would you know, for me, I think,
31:52
um, as I have a small business that was started
31:58
post lem adoption AI native, if you will. I'm seeing two. Absolutely seeing
32:05
two, um, areas where of improvement. One is on cost and the other is revenue.
32:12
So on the cost side, I absolutely have not hired
32:18
additional at least two people because of automations and efficiency
32:25
gains, largely administrative. Uh, that
32:32
I'm using the models for, uh, or I'm using existing
32:38
automations and software capabilities that the models have allowed a non-technical person to
32:44
like me to draft and implement. The second would be on the revenue side. I have been able to
32:51
pursue more business a faster because
32:58
of the AI tools. I mean, whether it's putting together bids and proposals
33:05
to win business, uh, or, um, you know, drafting,
33:12
you know, critical documents that have allowed me to advance projects faster, that's happening, that
33:19
has happened. I have documented proof of that. And so I think smaller, leaner
33:26
AI native are seeing more than existing incumbent large enterprise who haven't fully adopted these
33:32
tools, but that, that that is coming that they're going to do that and they're going to start
33:39
seeing that, and we're going to start seeing it more than just the on the margins or with the
33:44
with the administrative level.
33:52
I mean, I will say, like, for me, the biggest thing is I think the biggest impact it has had is on
33:57
the software industry. And maybe that's where it's harder to differentiate, um, between consumer and a
34:04
producer because I'll, I'll say this as a domain guy. The biggest handicap I had dealing with
34:10
softwares was I didn't know Python, I didn't know the the language a computer speaks to, to,
34:17
to explain to it what automation automations I want to build. But thanks to Claude or, you
34:24
know, I mean, I use Claude, but whatever people use, um, I don't I can just talk to it in natural
34:30
English and have it do things I needed to do, which talks to our first point, which
34:37
is the influx of AI startups, because, you know, a lot of people can can build a
34:44
reasonably good tool, which would have taken a lot more effort in the SaaS days. But then I keep
34:50
going back to process on the. So first of all, John, I want to acknowledge on for a small business,
34:56
actually I do see value because there are probably accounting dashboards. And like there are
35:02
a lot of these, these little things that you would have need to needed to hire a consultant or
35:08
someone to kind of put it together for you, or you would spend a bunch of your own man hours that
35:13
Claude excel can can probably help you. But again, again, I go back to the process
35:20
and see how if I am running my process in construction, what measurable
35:26
positive impact has AI had? Um, I haven't seen that here. That is not to say it can't happen
35:33
in the future. Yeah, well. and if you're, uh, a project manager, a
35:40
project engineer, and a large construction company right now, the way you can advance the ball for
35:46
your firm and for yourself is becoming as efficient as possible by
35:53
using AI tools, whether it's ones you're coding yourself, whether it's just simply using,
36:00
uh, copilot or whatever they're giving you, uh, or, you know, maybe you're even paying the 20 bucks
36:07
yourself. And as long as that you're compliant with the the data security, uh, at your firm,
36:14
that's that is where the gains are going to come from is, you know, individual efficiency as well. So.
36:22
Okay. Yeah. Kind of final thoughts here on this AI gains. Yeah I think you're on
36:29
to something which is, uh, the companies that were built in the post AI
36:35
era are having good gains from here on the productivity side, and
36:43
that's going to take time to trickle to the enterprises and S&P 500. We might see those gains
36:50
in 27. 28 is my guess, or even Q4 of 26, who knows. But
36:58
are those gains justifiable to the 1.2 trillion of CapEx, or the third
37:05
32 trillion of Tam being estimated by this AI companies? I think that's a hard one to match.
37:12
And that means there will be a we're at the top of the hype. Everyone saying it
37:19
now, even the guys who are invested in this are saying it's a hype now. So there
37:25
will be a correction. I just want everyone to be
37:33
have as strong of balance sheets as possible, and have as as strong of patience
37:40
and lean spend possible for the next two years before this happens. I think that's my
37:47
advice to everyone. Words of wisdom from a serial founder. Here's what I think. I think we are not at
37:54
the peak of the hype, I think. I am less bullish on this having a
38:00
positive impact on how we do business, that real world business. But I am very bullish on AI
38:07
companies making a lot of money and continuing to grow because all the people with the money think
38:13
that's what's going to happen. And that's what CEOs of most large companies are pushing for.
38:19
Whether it's going to be societally a net positive impact that's yet to be seen. That's
38:24
that's my thought on this. Um. A lot of meat there for a future episode. Thank you. T.K.. All
38:31
right, boys, we'll see you next time.