Latest / AI for Founders with Ryan Estes / Reality check every founder needs in 2026

Transcript

Reality check every founder needs in 2026 · AI for Founders with Ryan Estes

0:00
Welcome to AI for Founders. I'm Ryan Estes.
0:02
It's late December energy. The year's basically over.
0:05
The cookies are gone, the group chats are quiet.
0:07
And this is the one moment where you're allowed to stop shipping
0:10
long enough to look back. Because 2025 was insane for AI.
0:15
Every week felt like a new launch, a new model, a new
0:18
panic, a new promise that this one would change everything.
0:21
Faster code, cheaper labor, smarter agents, louder fear.
0:24
And somewhere in the middle of all that noise, a lot of
0:26
founders quietly fell behind without realizing it.
0:30
That's why this moment matters. Not to chase headlines, but to
0:33
understand what actually shifted.
0:35
Ranor Rousey has been building software for 30 years, and what
0:39
he's seeing isn't hype, it's pressure.
0:41
In 2025, delivery timelines got cut in half.
0:45
Clients didn't ask for less, they asked for more.
0:48
Faster AI didn't end projects early, it turned the same teams
0:52
into factories. MVP shipped sooner, and instead
0:55
of stopping, they kept going. More features, better
0:58
architecture, automated workflows.
1:01
Les, we'll clean this up later. Here's the part that founders
1:04
miss. The advantage isn't the model,
1:07
it's the system around it. If your onboarding brakes at
1:10
scale, if your back end can't handle growth, or if your
1:13
workflow still assume humans for everything, repetitive AI just
1:16
exposes the weaknesses faster. That's where teams like Auto
1:20
Maze step in, acting like a technical Co founder rebuilding
1:23
foundations while AI quietly takes over the expensive, soul
1:27
crushing work you assumed required.
1:29
More bed count. And looking ahead to 2026, the
1:32
big question isn't capability, it's trust.
1:35
Agents can already do more than we're comfortable admitting.
1:38
The real winners will be the founders who learn how to
1:40
delegate to AI without surrendering judgement.
1:43
Co pilots turn into Co workers, systems get quieter, interfaces
1:47
disappear, output goes up, and the margin between leaders and
1:51
laggards gets brutal. So here's the question worth
1:54
sitting with. If your company doubled its
1:56
output overnight, would it give you freedom or would it just
1:59
raise expectations to tighten the leash?
2:02
If you're a founder, operator or builder trying to make sense of
2:05
what actually mattered in 2025 and how to position yourself for
2:08
2026, you're in the right place. This is about tools,
2:12
infrastructure, leverage, and the uncomfortable truths hiding
2:15
behind the headlines. If you want to keep thinking
2:18
this way, go to aiforfounders.co.
2:20
It's the newsletter I send when something actually changes.
2:24
Not the hype, not the noise, just what works.
2:26
And if the show helps you see the game more clearly, leave a
2:29
review for somebody like you. And one last thing, if you're
2:32
building something real and no one knows about it yet, that's
2:35
not a product. That's a problem.
2:37
It's an attention problem. Kickcaster helps founders turn
2:40
podcast guesting into repeatable growth channel by getting you on
2:44
shows your customers already trust.
2:46
We handle the strategy, the pitching, the booking and the
2:48
prep. You show up and tell your story.
2:51
1 great interview can turn into pipeline partnerships and long
2:54
term visibility. If you're ready to be heard,
2:57
Kickcaster is how you do it. That's kickcaster.com by Mobers
3:01
a AIRI all right ran welcome to the
3:19
show. Thank you so much for joining
3:20
me. Thank you for having me.
3:23
Absolutely, absolutely. My pleasure.
3:25
Really excited for this conversation.
3:27
So we talked a little bit beforehand.
3:28
We're going to do kind of a recap of the big stories in 2025
3:33
in AI. There were a lot of big stories.
3:37
I mean, it just seemed like every week was a major launch,
3:40
major headline news. So we've got a lot of ground to
3:43
cover there. And then maybe we'll get to kind
3:45
of what we're looking forward to in 2026 and kind of new year,
3:49
new me energy. But to kind of kick things off,
3:52
Ran, give us an introduction to you and to your work.
3:57
Right. So I've been in the programming,
4:02
marketing, entrepreneurship space for the better part of the
4:06
past 30 years. I started very, very, very young
4:10
and that was always been my passion, doing code work.
4:15
I'm a tinkerer. So that kind of led me to open
4:18
source. I like to view the source of
4:21
what's happening under the hood and therefore also contribute.
4:25
Over the past number of years I've been running a company
4:29
called Autumn Maze here on a T-shirt and with we're providing
4:35
ACTO and kind of technical Co founder as a service type of
4:39
thing. We work with the companies
4:41
around the world and providing development to consultancy,
4:45
architecture guidance and obviously lots of AI over the
4:52
past year or so. And yeah, I'm, as I mentioned,
4:59
I'm an open source contributor. I have created libraries for
5:04
market data in the finance space called Y Finance, if anyone here
5:09
heard of it. And overall, yeah, my open
5:12
source code is enjoying about 20 million downloads every month.
5:16
So I'm quite, quite proud of that.
5:18
And yeah, look forward to to having this conversation.
5:22
Yeah, 100%. So kind of did the consulting
5:25
work, you know, how did kind of the, the project requirements
5:28
and kind of what people were looking for in the beginning of
5:31
the year, let's say last January kind of evolved through the
5:34
year. Yeah, definitely.
5:36
So the the way that we work in general is kind of we don't do
5:42
kind of a one off projects. We work with our clients long
5:44
term and I would say that if a year or so ago when we were on
5:51
boarding a new client, it was all about let's build something
5:56
from scratch or completely revamp and older legacy system.
6:03
I would say that these days, the first, the expectation is that
6:09
with AI, everyone expects us to deliver more, which which we
6:14
are. So there's definitely this type
6:18
of expectation from clients from something that was very
6:21
reasonable to take 3-4 months a year or so ago.
6:25
The expectation now is for it to take at least at most half of
6:29
that time. Yeah.
6:31
So that's, that's been kind of kind of at least we, we, we to
6:36
be honest, we weren't kind of sure how to look at it.
6:41
So what we were going to have less clients because we are
6:43
there expecting us to finish the the job faster.
6:46
But what had happened is that we just ship more.
6:50
It's it's not like we're ending contracts any sooner.
6:54
We just ship more. We are finish up with MVPS a lot
6:59
sooner and go continue with developing more features and
7:05
enhancing internal processes. And so from our perspective, we
7:11
work the same. We just produce a lot more with
7:14
the with the help of AI. I know you know, that's what
7:17
they're the, the goal was like, hey, we can 10X our workflow, so
7:20
we get all this time off, but it just turns out we just ship more
7:24
and work. Exactly.
7:27
I love it. Exactly right.
7:31
This episode is brought to you by Moberst, the AI first growth
7:35
engine trusted by iconic startups like Calm, Robin Hood,
7:39
Reddit, Uber and Google. Moberst is a digital, full
7:43
service, mobile first marketing agency that helps companies to
7:46
scale and become category leaders.
7:49
Whether you're finding product market fit or ready to
7:51
accelerate growth, Moberst delivers creative, data-driven
7:55
solutions that help with scale and efficiency and impact.
7:59
At MO Burst, we're not just guessing what works, we're
8:02
predicting it. As an AI first agency, MO Burst
8:05
uses cutting edge tools to analyze user behavior, adapt
8:09
creative in real time, and uncover hidden growth
8:12
opportunities. The result?
8:14
Marketing that's faster, smarter, and built to convert
8:17
from early stage startups to global brands.
8:20
We've helped hundreds of companies overcome complex
8:23
challenges, unlock new audiences, and optimize every
8:26
metric that matters. At Moeburst, we believe that
8:29
creative thinking, advanced technology, and data are the
8:33
core drivers of success at scale.
8:35
If you're serious about scaling your business with precision and
8:38
purpose, it's time to talk to Moeburst.
8:41
Visit moeburst.com and let's build something category
8:44
defining together. OK, I've got a confession.
8:49
As a podcaster and founder, I'm constantly meeting incredible
8:52
people, guests, listeners, or other entrepreneurs.
8:55
But here's my little secret. I'm terrible at following up.
8:58
My inbox is basically a graveyard of missed
9:01
opportunities. Does that sound like you?
9:03
You meet someone amazing, exchange emails, have great
9:06
intentions, and then life happens.
9:08
Six months later, you're kicking yourself thinking I should have
9:10
stayed in touch with that person.
9:12
This is exactly why I started using Warm Star dot AI and
9:16
honestly, it's transformed how I think about my network.
9:18
Instead of letting relationships go cold, Warm Start analyzes my
9:23
existing contacts and helps me identify who I should be
9:26
reaching out to and why. It even writes the outreach
9:29
messages for me. But here's the key.
9:31
I review everything before it goes out.
9:33
I'm not just talking about this because they're a sponsor.
9:36
I'm talking about it because watching a healthcare executive
9:39
use this to build his entire Advisory Board in one weekend
9:42
with four CEO sign ups made me realize I was leaving money on
9:45
the table every single day. Your network is your net worth,
9:49
but only if you actually nurture it.
9:52
Check out warm start dot AI. Use code AI IFF to get 50 bucks
9:57
off your first month. Stop letting your relationships
10:00
turn into miss revenue. And you know, so as a fractional
10:04
kind of CTO, talk about the teams you work with.
10:07
Am I safe to assume these are kind of designers, biz biz OPS
10:11
people, maybe investors that are like, hey, we, we need somebody
10:15
to ship something. So we're calling RAN.
10:18
Is that kind of how it worked mostly?
10:20
Well, we have two, I would say 2 main types of of clients.
10:26
Ones are that either have their own business already running.
10:32
That was actually our first customer that we still work to
10:35
this day, to this day with. They had kind of growing pains,
10:41
but they had good problems. More and more users were signing
10:44
up, but the server and architecture just couldn't
10:46
handle it. So they had to bring in someone
10:49
with more experience and a good quality team to kind of take a
10:53
look at their existing architecture and see how it can
10:57
be improved. So I would say that that
10:59
belonged to Type 1 of customers, existing businesses, digital
11:04
businesses and they're looking to to fix what's not working or
11:11
convert their existing application to something that
11:13
can scale. On the other end of the
11:16
spectrum, we have founders that are just turning up starting off
11:21
with projects. So we're building the MVPS for
11:24
them. We're helping them kind of
11:26
decide what the product would be, what's reasonable to have in
11:31
a, in an MVP, what's reasonable to have in AV1, ETC.
11:35
And in, in that type of environment, you can see that or
11:40
that type of, of client, it completely varies.
11:44
So we work with, with anything from founders of dating apps to
11:51
casual gaming through health clinics and insurance companies.
11:56
So it's really all over the all over the place.
11:59
And yeah, definitely over the past year or so it's been
12:05
heavily inspired, but by what AI could bring to the table.
12:10
So we completely automated processes with the insurance
12:14
companies and clinics. The things that I wouldn't say
12:19
were impossible before would would have been very expensive,
12:22
that the project would probably wouldn't have taken off or
12:26
wouldn't even started. Totally, Yeah, I would say one
12:30
feature of 2025 was really the rise of AI augmented programming
12:36
tools, you know, the the leaps and bounds and improvements this
12:40
year. So I'd love to hear about your
12:42
tech stack, you know, one what, what are you using LLMS for and
12:45
which ones? And then what is kind of your
12:47
favorite programming tool? Are you a cursor guy?
12:50
Are you a clawed code guy? I'm a Droid guy actually.
12:57
I've been using factory AIS Droid heavily over the past few
13:02
months. Kind of switched from about a
13:06
year ago from cursor to clawed code when it came out, and I've
13:11
noticed a better results from droids using the same model.
13:16
So I'm using Opus 4.5 as the as the LLM with the agenda kind of
13:23
wrapper, if you wanted to call it that way of Droid.
13:28
They seem to have a better memory after a compression.
13:33
So after they compress the session, they have better memory
13:37
and as to what we were working on.
13:39
So I kind of like that. As I mentioned, I started like
13:43
30 years ago. So it's very I'm using to me
13:47
that we're kind of went full circle back to the terminal and
13:52
I'm feeling right at home there. This is where I started.
13:56
So it's very it's very nice to to be in that environment again.
14:01
But yeah, I would say that a genetic wise that's Droid with
14:06
Opus, but in general tech stack and obviously it depends on what
14:12
the project is, but my go to would be kind of a hybrid of
14:18
gold Python And JavaScript with databases.
14:25
It's Postgres click house. These are kind of my go to's.
14:32
But obviously everything is based on what the project
14:36
requires. Yeah, we're not going to over
14:39
bloat a project unnecessarily or under deliver unnecessarily.
14:45
So it's it. It really is a custom tailor fit
14:48
for per project. Amazing.
14:51
You know, and I am an amateur programmer.
14:54
So when you talk about returning to the terminal, what do you
14:57
mean? So those black screens where
15:02
everything happens, kind of the Unix style of of coding.
15:05
Monochromatic. Monochromatic.
15:08
Yeah. Well, now they have, they have
15:10
colours, but it's still the same environment.
15:14
It was always kind of a favorite place for for me to to to
15:21
decode. And it's also the most
15:25
efficient. It it almost no memory
15:27
consumption or CPU consumption. So it was always nice to to be
15:34
in that environment again. But yeah, Windows users probably
15:37
know this. Remember that as the DOS, the
15:41
old DOS operating system, and the PowerShell Mac users and
15:48
Linux users knows it as the terminal.
15:51
But yeah, that's basically where where the metric matrix happens.
15:55
So yeah, that's basically it. Cool.
15:59
OK, so I've got a series of big stories this year and I'm going
16:04
to throw them at you and just kind of get your take.
16:06
You know, we're right now, it's what, December 23rd?
16:10
We're getting ready for Christmas.
16:11
Everyone's kind of hibernating. It's my favorite time of the
16:14
year. You know, you, you eat too many
16:16
cookies, you take a nap, you wake up, you watch football, you
16:20
crack a beer, you go back to bed.
16:22
It's wonderful. Laid back times, yeah.
16:25
It's a good time to reflect. Obviously, we're coming into the
16:28
new year setting new goals. You know, Q1 is upon us, so
16:32
let's kind of reflect on 2025 and I will start with a question
16:36
or not a question, but just the rise of Chinese AI
16:40
breakthroughs. You know, I think American
16:44
supremacy and technology has hasn't really been tested since
16:47
maybe the space race. You know, we've basically led
16:50
the way and, and this year in particular with maintenance and
16:53
some other tools, China's really stepped up.
16:56
If you just look at the the videos on YouTube of their
16:58
robotics and their drone technology, they seem to have
17:02
surpassed the United States in certain kind of ways.
17:05
And I'll include Europe in that as well, basically the West.
17:09
So kind of what do you see as the implications?
17:12
Sleeping, but yeah. Yeah.
17:14
What do you see the implications of a strong and mighty Chinese
17:19
techno reality? I mean that's I always take what
17:28
I see coming out of China with a grain of salt.
17:33
That being said, assuming that it's 70% true, it's amazing of
17:41
of their achievements. And I think the main reason that
17:45
they've been able to really challenge the the US and the
17:49
West in that regard is that they see it as a national security or
17:56
a national priority. And they're kind of putting all,
18:01
all their chips, including some chips that should go on
18:05
improving the welfare of their citizens.
18:09
But I mean, that being said, you see the models that are coming
18:15
out, you see the robotics that are coming out.
18:17
On one hand, it's not very surprising that they are able to
18:21
achieve that. They've been, you know, the, the
18:24
factory of the world for the past 20-30 years.
18:27
So they definitely have the technology of the know how the,
18:30
the, the extremely talented and smart people.
18:33
So it's not like anyone should be surprised by that.
18:38
But I, I think that their, to me the most surprising thing is
18:44
their approach that all the LLM models that we're seeing coming
18:48
out of China, they're all open source.
18:50
So that to me the most surprising thing and that might
18:55
be their edge going into 2026 because you know, open source,
19:03
none of what of the technologies that we have today would have
19:06
been here with without open source contributors.
19:11
And it's just a scientific method.
19:13
You know, we have, you have thousands of people looking at
19:15
your code and trying to hack it and improve it and finding
19:19
security flaws or just way of ways of optimizing it.
19:23
So I really think that that's going to be their edge in the
19:26
fact that from the US at the very moment only Meta has an
19:31
open source model that's that's kind of that.
19:37
I think that that's going to be an issue.
19:40
Yeah, it's interesting that the Meta is kind of AI is kind of
19:43
falling off a little bit. You know, they haven't had a big
19:46
announcement in some time, especially because it is open
19:49
source. So when you're thinking about,
19:50
but I wouldn't count Zuckerberg out.
19:52
So you know, when you're thinking about like Edge with
19:55
open source, ultimately it's just like adoption, right?
19:58
Like if people are going to want to to use an open source
20:02
framework. Right, for multiple reasons.
20:06
One, they want they want to know exactly how it works.
20:10
They want to try to improve it. They may need a smaller version
20:13
of it that they can hack and run on the phone, or they can just
20:19
buy a $10,000 PC and run the LLM at home without paying monthly
20:23
fees. So there's various reasons why
20:27
people would want to use an open source model and kind of, I
20:32
mean, the the American companies obviously get it because you see
20:35
that their entire focus is on getting developers on board.
20:40
So it's not like they're blind to the fact that whoever kind of
20:46
owns quote UN quote developers will will own the market.
20:50
But they're kind of missing it with not going open source or at
20:55
least open source. Your previous model, it's fine,
20:59
you know, just open source clawed 3.5 or GPT 4.
21:05
Just at at least give the open source community something to
21:09
work on. Yeah, did you see the videos of
21:13
the Chinese robots that they taught karate to?
21:17
And I've seen some of them as backup dancers to to like a a
21:21
band as well. It's just it's quite cool.
21:24
Yeah, it's really cool. I I love.
21:27
I know I have mixed feelings like all of us.
21:29
I guess it's it's cool and creepy at the same time.
21:33
Yeah, they, they can't be teaching them martial arts.
21:37
Like, what are we doing? Like, do not teach them martial
21:40
arts. It's a terrible idea, you know,
21:43
Yeah, for a lot of reasons. But it's really cool to see him,
21:45
you know, And I saw the backup dancer video and, you know, I,
21:49
I'm contemplating seriously like doing like a, a robot clothing
21:54
line. And the first thing I noticed
21:56
was the outfits for the robots were terrible.
21:58
I was like, dude, this is going to take off.
22:01
So you get whenever Elon rolls out his robot army, that might
22:05
become kind of the standardized size and shape.
22:08
I think that's when I I. And you'll have probably like
22:10
3-4 sizes, so it's going to be easy adjustments.
22:15
Yeah, give them little outfits, you know why not?
22:17
I think it'll be cool, you know, so, so the rise of China is.
22:22
Probably going to be army uniforms first.
22:25
That's my suspicion. Yeah, if they, if they start
22:28
launching these things, they're camouflage and you know what I
22:31
mean? Like we're over and we'll get to
22:34
that 'cause I think that kind of AI deployment in government and
22:39
military is a really interesting question.
22:40
But I want to stick on kind of China for a second, largely
22:44
about their ability to move quickly because it's a national
22:47
issue. And there really there's only
22:50
one source of truth. You know, you, you kind of
22:53
mentioned that Europe is kind of asleep a little bit with
22:55
technology innovation. And a lot of that really has to
22:58
do with regulation and compliance that's more partly
23:02
enforced in Europe. But you know, the United States,
23:05
particularly California had legislation that was passed for
23:08
AI transparency, requiring risk disclosure, safety frameworks,
23:13
basically they're putting roadblocks in to slow us down.
23:16
So when we're looking at this this race, some people warn that
23:19
we're running headlong into our own destruction to be really on
23:23
the outside of the main and the doom doomsayers.
23:27
But at this point, it looks like there there really can't be any
23:30
slowing down because it's it's another arms race whether we.
23:35
Like it or not. So kind of what's your take on
23:38
like maybe the Europe's approach to regulation and maybe what the
23:41
United States should start to be thinking about?
23:44
Well, definitely not what Europe is doing.
23:47
So that's definitely not what the US should be thinking about.
23:50
I mean, there are obviously a lot of talented people in in
23:53
Europe. It's not like the continent is
23:55
is cursed or something. But the over regulation kind of
23:58
hinders entrepreneurship and and creativity.
24:02
And the one thing that's great about the US, you start with a
24:08
yes almost everywhere you go. And let's see how we can work
24:11
this out as as opposed to Europe that you start with a no and you
24:15
need to see how you can navigate that maze of bureaucracy.
24:21
But regarding the California thing, I'm I'm I'm OK with with
24:29
disclosure. Disclosure is fine.
24:31
No one is against the privacy policy, but the the problem is
24:34
when a privacy policy becomes like a, a giant pop up that
24:39
tells you that you're we're using cookies like it's a
24:41
surprise. So, so that's where it it really
24:45
becomes a roadblock. I think as you rightly mentioned
24:52
that it is an arms race and it's not necessarily an arms race
24:58
between the US and China, but it's literally W versus East, I
25:03
guess. And this is how the, the US need
25:10
to look at it. And I, it seems, again, I'm not
25:13
too heavily involved in, in AUS politics, but it seems like the
25:17
recently the administration came out with kind of an AI policy
25:24
that they're seeing it as a, as a national emergency or
25:30
something like that. It seems like someone's at the
25:33
top kind of understands that. On the other hand, you probably
25:39
don't want to say if you are an AI company, you pay no taxes.
25:44
So you need to find the kind of the balance between what will
25:48
give a lot of entrepreneurs kind of the the desire to go and
25:54
innovate and not be stopped by regulation.
26:00
Something also should also probably mention that something
26:04
needs to be done in order to open the door for smaller
26:08
start-ups. I don't know how I really don't
26:12
have the answer because just lifting regulation won't just
26:16
give them 10s of billions of dollars to invest in
26:18
infrastructure. But yeah, that's the fact that
26:25
you need that much capital to create a new model.
26:32
Yeah. It's it's an issue that I don't
26:35
think a government can solve. Yeah, it's yeah.
26:37
Yeah, I mean Bytedance alone has has capital commitments in 2025
26:44
of 23 billion and that's insane. It's just them.
26:48
So you know, a, a lot of this regular regulation has come
26:51
from, you know, Sam Altman and the Elon's where all of them
26:54
want to basically scare everybody to create a regulatory
26:59
Moat and be like, hey, everyone should be really afraid of these
27:02
AI companies. The government should just trust
27:04
us and we should have, you know, total domination in this market.
27:08
And so, you know, a lot of that like AI will take our jobs kind
27:12
of came from that hysteria. But what they're doing really is
27:15
just playing to the fear tactics in order to try and get some
27:18
regulations securing their position as #1 which which is,
27:23
which just sucks, honestly. And hopefully in 2026, people
27:26
will start to see beyond that. Now, I think as far as
27:29
innovation on, you know, commercial or whatever kind of
27:33
means, like we really have to like remove that regulation, my
27:36
personal opinion, so that we can keep pace.
27:39
If there's some way that start-ups can compete, maybe
27:42
there should be some like $20 billion endowment to, yeah,
27:46
startups to help. That's.
27:48
One way of doing that, yeah, or if you're a monopoly, you have
27:51
to give 10% of your TPU powers to to startups or something can
27:57
be arranged along those lines. By the way, you I suspect that
28:02
part of the reason why the Chinese models are open source
28:05
is to increase trust and companies in the US to use those
28:11
models. So you'll have at least have the
28:14
ability in theory to take a look and take a peek inside and see
28:19
that whether or not you're being spied on.
28:21
Yeah, 100%. And, you know, and maybe this is
28:25
a place where we need kind of unilateral, you know, China
28:28
cooperation for certain kind of regulations that I think will
28:32
put a lot of people at ease. You know, people are afraid of
28:35
using maintenance, but they're happy to go on TikTok.
28:38
So, you know, you can't tell public opinion.
28:41
But when we're talking about AI adoption for warfare, I think
28:45
there needs to be some kind of like summit, you know, and maybe
28:48
Trump's the guy to get Xi Jinping in and like to the table
28:52
to say like, hey, look, we don't want, you know, a cloud of
28:56
drones flying over a town and eliminating anybody who doesn't
29:01
have 10,000 YouTube subscribers or whatever kind of weird AI
29:04
qualification they can take. You know, I think that kind of
29:08
Black Mirror episode is what nobody wants.
29:11
We don't want the Kung Fu robots knocking on our door and saying,
29:15
hey, we're looking for Rand because he's using the open
29:19
source competitor that of our rivals, you know, right.
29:23
And I haven't seen a lot of that talk.
29:25
You know, there's posturing over Taiwan, there's posturing on all
29:28
over the place, but really a place just to come together and
29:31
be like, hey, let's have closed doors talks about like where
29:34
where we're not going to go with with AI because it's just it's
29:38
too threatening and too scary. Well, you're, I agree with you,
29:43
I just don't think it's ever going to happen in even if they
29:46
have this kind of summit and agreements, each country will
29:51
just serve its own interests behind the scenes.
29:54
So realistically speaking, I just don't see it happening,
30:00
sadly. It would be great if everybody
30:02
could agree on this is, you know, this is ethical.
30:07
This is isn't ethical. But different cultures have
30:10
their own moral barometers. So I don't I don't see this this
30:16
happening. If you look at you don't want
30:20
robots coming at your door. One of the things that kind of
30:26
concerning is on the one hand, if you go to like in a battle in
30:32
an open field, a robots versus robots, I don't see the point in
30:37
that. I mean, my robots are better
30:40
than your robots. I mean, I don't there's really
30:43
no not much sense in that. But once they get where the
30:47
citizens are, and I guess I'm more concerned about the AI
30:53
rather than I am of terminators working around.
31:00
Although when when we get to that level, maybe I'll be scared
31:02
of them as well. But yeah, AI, the main issue
31:08
with it that we all should be concerned of is that it's a non
31:14
deterministic software. You really cannot predict in
31:22
certainty what it's going to do giving a set of parameters.
31:29
And the more advanced it gets and the more life experience it,
31:36
it, it gains just from being an aging model, it's less and less
31:44
deterministic. So unlike software you where you
31:47
can say, oh, I know that if the software sees, I know this many
31:52
drones in the sky, it's going to launch a missile, You don't know
31:58
that the AI will make the same decision.
32:01
So that's that, that's the tricky part.
32:03
I mean, there's some probabilities that you can guess
32:07
that it's going to do this or that, but unlike software, it's
32:10
not a workflow with a set a set number of steps.
32:15
Yeah, that's a major problem. And I guess the work around
32:19
they've had is, is just kind of the model releases, because
32:22
that's also been a major story this year is every, you know,
32:25
every couple of weeks, you know, open AI step takes a step
32:29
forward and then Google brings something with Gemini and then,
32:33
you know, Anthropic drums up here and then Claude's jumping
32:36
up here. All of the model releases are
32:38
really have been like huge leaps, I mean like huge leaps.
32:42
But we're starting to see those those gaps kind of narrow.
32:45
Do you think that's because the the models themself are starting
32:47
to deteriorate or do you think we're kind of starting to hit a
32:50
wall, so to speak of like what they're capable of?
32:57
I wouldn't necessarily say that we're starting to hit a wall.
33:00
I mean, obviously, one, once we get to 100% software bench,
33:05
software engineer benchmarks, then yeah, that you can get it
33:09
cannot get better than that. But we're still far away from
33:13
from this point. And I, I think that we're going
33:19
obviously we're going to see the, the differences kind of
33:26
becoming smaller and smaller between the different models.
33:28
And very funny to me to see on X people saying I just cancelled
33:32
my cloud subscription because Gemini is amazing.
33:36
And three days later I just cancelled my Gemini subscription
33:39
because cha GPT has the best model and just I'm I'm not going
33:45
to switch a model I'm comfortable with because of half
33:48
of the percent improvement. It's it's just not not going to
33:53
be the case. I am playing around with all the
33:57
models I'm subscribed to all of them.
34:00
That's sadly for my wallet, but but but yeah, at the end of the
34:06
day, you have your you have your go to's.
34:10
Yeah. So going back to the actual
34:12
question, if we're going to hit some sort of a wall, I don't
34:17
think so, not in the next, I would say 2 years, but the the
34:24
exponent is going to slow. Yeah.
34:26
So yeah. Well, and let's let's talk a
34:29
little bit about the major players in the LLMS in the
34:31
business models, you know, because this is also kind of a
34:34
sustainability kind of capability, you know, so you,
34:40
you have, you know, the grandfather of them all open AI
34:44
that's raised unbelievable cash and and Sam's looking to raise
34:48
more at some diabolical evaluation of his company.
34:53
But you know, there's also pressing litigation and lawsuits
34:57
as far as like what they've trained their data on and what
35:00
kind of liability and what kind of money they're going to have
35:02
to pay for that. And we can probably safely
35:05
assume they've already digested the entire Internet 10 times
35:08
over and so everybody. They can just have them pay
35:13
money. Yeah, that's that's an easy
35:15
problem to fix. You you bet they'll, they'll pay
35:18
off everybody. But then at the end of the day,
35:20
all the money, all the valuation, they're not going to
35:23
be able to, you know, survive on $20 monthly subscriptions.
35:27
And if we're thinking about, you know, if we have settlements in
35:31
these lawsuits of like for data and information moving forward,
35:35
then they're at a a very big disadvantage to say Grok or to
35:39
Google, obviously, even to Apple, Apple doesn't want to do
35:43
any kind of like open LLMS. But between Apple, Google,
35:47
Amazon, they have mountains of cash, you know that ultimately
35:51
they're going to need to deploy and why why not deploy it in
35:54
this in this way? So when you're looking kind of
35:57
as business models, it's hard to bet against Google right now.
36:01
But what's your take? You know, where do you see kind
36:03
of the big LLM companies kind of panning out of for supremacy?
36:09
Well, financially and it's a very hot take, I guess I think
36:17
that open the eye is the one at the most disadvantage here.
36:25
Yeah, pretty much obviously I'm I'm long Google the fact that
36:28
they've first they own I think like 10% of them Tropic and so
36:35
they they have some hand in that as well.
36:38
The fact that they so they they they went with a really appelly
36:44
type of move here where they own the entire chain.
36:48
They have their TP us, they have their data, they have their
36:52
their distribution, they control everything.
36:55
So they are in the best position possible moving forward there.
37:03
Someone there was smart enough to say, you know what, we need
37:07
to make our LLM have the same API as as open AI just to like
37:13
to make it easier for, for adoption.
37:17
Because yeah, it was really horrible before that with Google
37:20
is known for having bad AP is with their systems.
37:24
But someone who's smart enough to say, not in this case, we're
37:27
just going to be open AI compatible because that's what
37:30
developers are already using. So Google definitely is above
37:35
anyone else here regarding Amazon, I'm not even sure that
37:41
they are going to play the LLM game.
37:44
I think they're just at least it seems that they are saying, you
37:50
know, we're going to be the the infra, we have Bedrock, just
37:53
bring your models and we're going to be the marketplace.
37:58
Entropic has it's kind of the the favorite amongst developers,
38:05
which positions them in probably the best position moving forward
38:12
in terms of adoption and usage within software.
38:15
And just the fact that they are the favourites among developers
38:21
while being 20 times more expensive than Gemini or four or
38:26
five times more expensive that GPT just tells you how good
38:30
their models are. People wouldn't just volunteer
38:34
to pay 20 times what they can get from Google.
38:38
So GPTS when open AI despite the fact that they were here 1st and
38:45
they have amazing models. It's not like they are behind on
38:49
on a technological aspect at all is just financially.
38:54
I think that they're least independent, even their deals is
38:59
NVIDIA is investing in open AI. They need to use that money to
39:04
buy chips from NVIDIA. Same kind of deal with Amazon.
39:08
Amazon is now investing and they're going to give that
39:11
investment as AWS credit. So it's all kind of just money
39:16
moving around. It's, it's not like money going
39:20
into the the company's pockets. And Apple, I don't know what
39:23
Apple is doing. I hope it's they're going to
39:26
surprise everyone, but I've no idea what what they're doing.
39:30
Man, Siri is still so dumb. I I mean, I got in an argument
39:34
with Siri the other day. I was like, it, it forgot a
39:36
name. And I was like I said, like my
39:37
cousin's name. And it's like, I don't have that
39:40
contact. I was like, since when you've
39:42
had it for four years? Like, what?
39:44
What are we doing? Yeah, yeah, yeah.
39:47
And OK, cool. And Alexa as well.
39:50
He's just talking to an idiot. You know, it was a technological
39:53
marvel a few years ago. And now he's just, Oh my God,
39:57
you're so stupid. Yeah.
39:59
Would you like to subscribe to dog food?
40:01
It's like, Alexa, I don't have a dog, you know?
40:04
Like what? You should know that Amazon.
40:07
Exactly. OK, cool.
40:08
I want to talk kind of talk about big questions coming in at
40:11
2026. You know, you know, new year,
40:14
new May, new AI, new, new, where we going?
40:18
So one thing I wanted to bring up is, you know it this summer,
40:22
you know, everybody shipped AI agents and now kind of the job
40:26
reports are coming back that very few actually replace jobs.
40:29
So do agents in 2026 become autonomous Co workers or do they
40:34
remain kind of powerful Co pilots?
40:39
That's an interesting one. I think that they're gonna
40:45
become powerful Co workers towards the second-half of the
40:49
year as they first you need to build a trust.
40:55
Even if they're already there in terms of capabilities, the trust
41:00
is not there that they're going to complete the job.
41:03
But as more and more companies will start using these agents
41:07
and see that they're actually delivering on the promise, I
41:12
think that they will become very capable Co workers.
41:15
I'm not subscribed to this whole fear mongering of AI is going to
41:20
take our jobs. I think, hey, definitely going
41:23
to change the way we work, but I, I, I don't, I don't think
41:29
that is going to take anyone's job.
41:34
You know, if you account for some minor pivoting within your
41:37
job role, then you're probably going to you're probably going
41:41
to keep your job. I used to think like early in
41:44
the year, Oh the the junior developers are kind of screwed
41:47
because why would anyone hire a junior developer when they can
41:51
just give a senior developer another 1000 bucks a month in AI
41:55
tools and they can output a lot more.
42:00
But I since changed my mind. We hire a lot of junior
42:03
developers in autumn Maize as well and we taking them to a
42:06
training program and we use AI to actually make them senior
42:10
developers much faster. Obviously there's no earned
42:14
knowledge that you need to just be around for a while and
42:19
experience things, but the teachable knowledge, the
42:22
teachable experience that that can extremely be be accelerated
42:26
with, with AI. And if you don't hire juniors,
42:30
you're end up with no senior developers that they're not
42:34
just, you know, fresh up out of college.
42:37
You, you get senior developers, you have to have the juniors.
42:42
So and we've seen that in the news every time that a company
42:46
goes through a layoffs. People like to blame AI, but if
42:52
it's it's not really the case, as I mentioned, even with with
42:58
Autobots, we just ship more. It's not like we are ending the
43:01
contracts sooner. We just ship more and it's to
43:05
everyone's benefit. For the late night hours, you
43:08
know it just now the days never end.
43:10
I can't wait to get back into the saddle.
43:13
You know I like that you mentioned trust 'cause that's a
43:16
that's a big topic for 2026. I think.
43:19
You know I I probably abandoned 3/4 of the agents I built.
43:23
Even when they are functioning properly.
43:25
It just that level of automation was starting to be irritating
43:29
and so I left it. Now you can see, I think where
43:32
the the trust building and where the tools for this agentic
43:35
support in 2026 are going because everyone's going to the
43:38
browser. I think just clawed today
43:41
launched their kind of Chrome extension for the browser, which
43:45
is to say that the predictive elements of AI tools is really
43:49
what's going to be fantastic agentically because it can ask a
43:53
question. So if it sees like if I do this
43:56
e-mail and then I go to air table, I go to Google doc, I go
43:59
to Wednesday or Monday or whatever, you know, it can just
44:03
say, hey, I see you're doing onboarding.
44:05
Would you like me to complete that task?
44:06
In some cases I'd be like, yes, but in other cases it's like,
44:10
no, there's actually minute customization that even
44:13
explaining it to you is going to slow me down and be annoying.
44:16
So like as we have this integration across browsers and
44:19
across applications where it can predictably know what you're
44:22
doing before you do it and then kind of ask those follow-ups.
44:25
That's what I'm really hoping to see with agentic support in
44:28
2026. But also as far as human trust,
44:33
there's, they have a lot to overcome.
44:36
I mean now we have there's AI girlfriends, there's deep fakes,
44:40
there's synthetic voice clones, there's kind of AI generated
44:44
persuasion across the board. And who knows what Nigeria is
44:49
going to come up with for scams, you know?
44:52
No disrespect to actually get a Nigeria call of from the
44:56
Nigerian Prince now. Yeah, exactly.
44:58
It'll be, it'll be me and it'll be calling my grandma and being
45:01
like, oh, Ryan's in trouble. You know, it's just it, it's
45:05
really scary. So how do you feel like kind of
45:07
AI and the people that are skeptical for good reason, Like
45:11
how do, how do we as developers and, and businessmen, you know,
45:15
how do we build trust in the, in the tools and the companies
45:17
behind them? I think transparency is is key
45:24
so it's tricky. I actually read a few days ago a
45:29
report on Google ads. Some ads were generated using
45:35
AI, some ads were generated using AI and disclosed that AI
45:40
generated and some were just regular ads.
45:43
And the AI generated ads generated about 30% if I'm not
45:48
mistaking more click throughs. But once they disclosed that
45:54
they're using AI, the click throughs were actually 20% less
45:59
than non AI ads. So it's it's really shows you
46:03
about the trust issue and not necessarily the product itself.
46:07
Yeah. So it's like, well, everyone's
46:11
fine with your doctor going up to an encyclopedia, a medical
46:17
encyclopedia that he has on on the shelf behind him and just
46:21
read through to help diagnose the system, but will be
46:24
suspicious if he's going to use Google or AI for that.
46:28
So it's really about trust, even though it's essentially the same
46:31
tool. What we need to overcome is not
46:38
the non deterministic aspect of AI that that's fine.
46:44
You want it to be more human, let's say in that regard, but
46:48
you want to trust that it's not lying to you.
46:53
They're not hallucinating. And you know, everybody can make
46:59
mistakes, but you don't want it to just make shit up, you know,
47:04
and even when they make stuff up, they're very persuasive
47:08
about it. So that's the main thing that
47:11
needs to be resolved that that level of trust.
47:16
I my own kind of experience is that I asked every time that I
47:23
asked to judge APT or claw that I'm used mostly for information.
47:30
I always ask, please Fact Check and include sources.
47:36
Otherwise, yeah, so it takes a few minutes longer to get an
47:40
answer because they're going all over the web instead of just
47:43
using statistics to to give me an answer.
47:46
But at least I'm able to trust the answer more.
47:51
So I think that that's part of what needs to happen, that when
47:56
they give you an answer, they need to tell you why they gave
47:58
you this answer. It's just it's not just, you
48:01
know, research says that, which research says just show me the
48:05
research, just give me a a reference.
48:10
Yeah. So I think that that's the main
48:13
issue. So once we solve that, that you
48:16
can trust their data that they providing you with and that you
48:21
trust the output, that it's predictable enough that you can
48:27
just trust it to work. I think that you're going to see
48:29
it actually becoming a Co worker.
48:31
And if we look maybe 2-3 years down the line, I think that it's
48:36
also in a lot of aspects, especially for knowledge
48:39
workers, it's going to be kind of the end of UII actually
48:45
mentioned this to someone a few days ago that I think that UI
48:50
for for humans are going to stick around for leisure
48:55
purposes, you know, do online shopping, you want to browse,
48:59
you want to, you know, go into forms and talk to people that
49:03
that's going to stay. But a lot of business
49:05
applications just going to be a database and, and a chat bot.
49:11
And when I say chat bot, the interface doesn't really matter.
49:14
It could be a chat, it could be a voice, it could be a WhatsApp.
49:18
And we'll just generate the reports you need and do whatever
49:20
it is that you need. If you think of enterprise, this
49:22
is computer show me XYZ and it just gets the data and and shows
49:29
you. So the the actual forms and
49:32
usability, all that thing is going to start to wind down over
49:37
the next few years I think. Yeah, I, I would say that
49:41
probably the most successful software product in the last
49:44
five years was the Spotify Wrapped.
49:47
You know, at the end of the year, you got to see everything
49:49
you listen to and everybody's adopted that.
49:51
And it's so cool. Like I would like to have a
49:54
refrigerator wrapped. And they'd be like, Ryan, you
49:57
ate 7321 eggs this year. Good job.
50:00
I was like, oh, my protein goals.
50:02
Let's go. You know, I like, I like your
50:06
attention to like, give me Fact Check this data.
50:10
And my favorite, you know, from ChatGPT, they've improved it a
50:15
little bit, but they would make up whole articles with URLs that
50:19
went to four O 4 pages and everything you like.
50:21
No, that's you made that up. It's kind of completely made
50:25
that up. Yeah, with code.
50:26
I noticed that a very nice trick that I'm using is what's your
50:31
confidence level in the solution that you just created?
50:36
And they they immediately said, you know what, let me think
50:39
about it. I'm only 70% sure.
50:42
So it's work to get to 100. That's a good one.
50:45
I'm going to use that. What's your confidence level of
50:47
what you're telling me? You know, I'm going to use that
50:50
with my kids. That's a good one.
50:53
Yeah, nice. Fran, kind of wrapping up here,
50:56
when you're thinking about 2026, what are you most excited for
51:00
and do you have any New Year's resolutions?
51:05
So I'm most excited about seeing the change, not the incremental
51:12
changes, but I'm looking for that iPhone moment in AI for the
51:19
one that says, you know what, we've improved incrementally and
51:24
now it's going to change everything.
51:26
And I don't know what that would look like.
51:29
It will be some sort of, I think of a life side gig type of
51:35
agents that has full control in full access to your data somehow
51:42
privately. That was the hope that of what
51:46
Apple should have, should have done and just helps you manage
51:50
your life and connect to different sources and and just
51:54
freeze your time. New Year's resolution.
51:59
I wanna, I don't know, maybe a bulk up here.
52:05
I've gained, I've lost a lot of weight this year, so it's time
52:08
to bulk up. So yeah.
52:10
I like it. You gotta get out the dead lifts
52:13
and some some bench press both. Yeah, I'll start slow, but
52:18
eventually both can just start somewhere.
52:22
You know, it's like you eat steak and and oatmeal and drink
52:26
beer, you know, lift like a lazy bear.
52:28
I love it. Nice, amazing.
52:33
Before we get out of here, I know the Make a Wish foundation
52:35
is is near and dear to your heart.
52:37
Do you do you have a a special chapter or a shout out you'd
52:40
like to make? I think that these people are
52:44
Saints, to be honest. We had a personal experience in
52:48
the family and we've we've seen how happy they can make kids
52:54
during tough times. So yeah, every good thing that I
53:00
can wish upon those people is go straight out of my heart to
53:05
them. They are the best.
53:06
They are the absolute best. Now I know you've got some
53:09
pretty big announcements coming shortly, so I suggest everybody
53:12
go to moxi.org, MUX, i.org and that's the best place to learn
53:18
about this. Yeah, it is going to be very
53:21
exciting. It's an open source project,
53:24
going to be out anytime now wrapping up the the SDKS around
53:30
it and hopefully it will bring some sort of a revolution into
53:35
developing A genetic infrastructure.
53:38
Incredible. And then of course, auto maze
53:40
dot IO, which is a UTOMAZE dot IO which is technical Co founder
53:47
and CTO as a service where they can talk directly to you.
53:51
Ran ran. What's the best place for people
53:53
to connect with you personally online?
53:56
Personally, I'm on X. My aunt is Roussi AROU Double
54:03
SI. I'm quite active.
54:06
Find me, DM me, follow me. I'll follow back.
54:11
Yeah, we're going to have a blast.
54:13
Man, this has been tremendous. I really appreciate the time in
54:15
the conversation. I'm fired up for 2026.
54:18
I think it might be bulking season for me too.
54:22
Let's hope so. Thank you very much man.
54:24
I really had a great time chatting about AI and with you.
54:27
I mean, this is really, really been an experience.
54:31
Agreed. Loved it man thank you.
54:33
Thank you very much. Hey founders, let me ask you
54:37
this. Are you building with AI or
54:40
getting left behind by it? If you're serious about scaling
54:44
smarter, faster and bolder, you need to check out AI for
54:48
founders.co. It's your go to resource for
54:51
cutting through the noise and actually using AI to grow your
54:55
startup. Each week, AI for Founders drops
54:58
a sharp, no fluff newsletter, dives deep with expert guests on
55:02
the podcast, and hosts monthly workshops where you can get your
55:06
hands dirty with the real tools and strategies.
55:10
Whether you're precede or Series B and beyond, launching
55:13
something new or just staying sharp, AI for Founders gives you
55:17
the edge you need to win in 2025 and beyond.
55:21
Hit AI for founders.co to subscribe to the newsletter,
55:25
tune into the podcast, and grab your spot in the next workshop,
55:29
and it's all free. Build smarter, move faster,
55:32
think bolder with AI for Founders.
55:35
Now back to the show.