Definitely Not Talking AI with David Demaree

WEBVTT
NOTE
This file was generated by Descript
00:00:07.552 --> 00:00:10.049
Chris Enns: Hello, and welcome to
Dailies for Friday, October 16th, 2026.
00:00:10.065 --> 00:00:12.493
I'm recording an episode every day
for this, my 50th year on the planet.
00:00:12.828 --> 00:00:16.684
As I mentioned in the previous episode,
uh, which I just recorded minutes
00:00:16.684 --> 00:00:20.521
ago, so this is very confusing to my
brain, that messes with the timelines.
00:00:20.579 --> 00:00:24.423
Uh, I'm gonna have guests on for
Saturday episodes, which in theory
00:00:24.423 --> 00:00:27.526
would give me more time to edit and
produce and whatever, except that
00:00:27.533 --> 00:00:30.361
this happens to be on Friday, which
is the day before Saturday, so it's
00:00:30.362 --> 00:00:31.365
not a whole bunch of lead time.
00:00:31.366 --> 00:00:34.905
But anyways, my guest for this episode,
as you can see if you're watching
00:00:34.907 --> 00:00:36.833
the YouTube video, is David Demari.
00:00:36.881 --> 00:00:37.083
Dema-
00:00:38.464 --> 00:00:38.760
David Demaree: Demarie.
00:00:39.169 --> 00:00:39.801
Chris Enns: Demery,
00:00:39.971 --> 00:00:40.791
David Demaree: We literally just…
00:00:40.976 --> 00:00:41.555
It's okay.
00:00:41.610 --> 00:00:44.786
It- people have been saying Demarie
for my 46 years, so you're fine.
00:00:44.786 --> 00:00:45.303
Chris Enns: Yes.
00:00:45.332 --> 00:00:47.987
I that is the problem with…
00:00:48.152 --> 00:00:51.785
I have a, a thing in my brain that's
like, uh, when it's a left or right one
00:00:51.785 --> 00:00:56.804
and zero, then I- it's wh- I- my wife can
attest to, I will never remember it again.
00:00:56.814 --> 00:01:00.039
I have a note on my Apple phone, on
my iPhone, that's like, "Remember
00:01:00.040 --> 00:01:02.650
shit," and it's just, like, literally
those kinds of things where it's
00:01:02.651 --> 00:01:03.620
like, "How to pronounce this?
00:01:03.659 --> 00:01:05.266
Which farmer's sausage to buy at co-op?"
00:01:05.626 --> 00:01:06.837
Yeah, those important things.
00:01:06.845 --> 00:01:08.438
So I'll add your, your
last name to the list.
00:01:08.849 --> 00:01:11.051
I know David through the
Shop Talk Show Discord.
00:01:11.051 --> 00:01:13.988
Shop Talk Show is a client I work
on, a podcast I work with, and
00:01:13.988 --> 00:01:16.637
they have a Discord, and in that
Discord, there's people we talk…
00:01:16.675 --> 00:01:19.344
who talk and share and joke and et cetera.
00:01:19.539 --> 00:01:23.445
David is in that Discord, uh,
and so, and, uh, presumably also
00:01:23.445 --> 00:01:24.365
a listener to Shop Talk Show.
00:01:24.366 --> 00:01:28.302
But in addition to that is, uh,
software product leader, web
00:01:28.302 --> 00:01:29.459
designer, developer, author.
00:01:29.607 --> 00:01:32.444
You run your own consulting/design agency.
00:01:32.444 --> 00:01:32.740
Is that…
00:01:33.004 --> 00:01:34.730
What's the human version of
00:01:35.400 --> 00:01:38.738
David Demaree: Yeah, it's a,
we, it's a consultancy/agency.
00:01:38.770 --> 00:01:39.965
See, it's hard to do without the slash.
00:01:39.975 --> 00:01:42.609
I'd say a consultancy in
that we, but we build stuff.
00:01:42.613 --> 00:01:46.507
So, um, helping people, um,
mainly especially with CMS.
00:01:46.515 --> 00:01:50.543
It's like it's hard to get people to kind
of be like, um, separate CMS from website.
00:01:50.558 --> 00:01:53.319
So like oftentimes it's
like, "We build websites."
00:01:53.338 --> 00:01:56.189
"Oh, you mean like you work on
like, uh, Squarespace or Webflow?"
00:01:56.198 --> 00:02:00.402
Like no, the good, the good stuff
with like, you know, uh, like with
00:02:00.413 --> 00:02:04.583
backends and, and CMS and like
more, you know, content operations.
00:02:04.593 --> 00:02:08.467
So kind of more like, you know, uh,
fancy, uh, business to business,
00:02:08.566 --> 00:02:09.799
uh, professional websites.
00:02:10.471 --> 00:02:13.181
Chris Enns: Yes, with real budgets
and not just, "Hey, can you help me
00:02:13.183 --> 00:02:15.321
on the weekend while I mow my lawn?"
00:02:15.344 --> 00:02:15.933
kind of thing.
00:02:16.046 --> 00:02:16.411
Yes.
00:02:16.827 --> 00:02:20.082
Very nicely designed websites
and really well done, I would say
00:02:20.084 --> 00:02:21.184
from what I've seen you share.
00:02:21.225 --> 00:02:25.991
As well as your blog is on
your website, D-E-M-A-R-E-E.me
00:02:25.998 --> 00:02:29.370
is what I'm just gonna say on the,
on the podcast anyways, and the links
00:02:29.370 --> 00:02:30.355
will be in the show notes, of course.
00:02:30.453 --> 00:02:33.105
Uh, we're going to not talk about AI,
but we're definitely gonna talk about
00:02:33.105 --> 00:02:35.084
AI in this episode, so just a heads up.
00:02:35.288 --> 00:02:39.261
Because the, the three f- first guests
are Shop Talk related, the three,
00:02:39.263 --> 00:02:43.775
there's the Discord obviously, we're
all swirling around this like, "Hey,
00:02:43.775 --> 00:02:47.040
it's fun to play with AI, but it also
is kind of gross to play with AI."
00:02:47.476 --> 00:02:50.167
And that was the-- my episode
yesterday was about that idea.
00:02:50.189 --> 00:02:51.377
And so also…
00:02:51.434 --> 00:02:56.396
I know everybody else, I feel like, is
sick of hearing people talk about AI
00:02:56.877 --> 00:03:00.184
David Demaree: Two, I think the, the
thing that people maybe struggle with in
00:03:00.185 --> 00:03:03.515
general in life, and especially with AI,
it's not special, it's like that when
00:03:03.517 --> 00:03:04.905
two things can be true at the same time.
00:03:05.378 --> 00:03:09.477
I stopped myself from, like, going totally
down, 'cause I, I decided instead of just,
00:03:09.478 --> 00:03:13.434
like, randomly ranting on social media
this morning to kind of be, like, put
00:03:13.434 --> 00:03:15.909
out, like, a newsletter and do some things
that were a little bit more focused.
00:03:16.430 --> 00:03:18.249
And so, um, there…
00:03:18.250 --> 00:03:22.106
But I did see somebody, and I was very
surprised that they had published, um…
00:03:22.119 --> 00:03:27.144
It was, like, another web designer, agency
owner type person, and he said, um, "I'm
00:03:27.144 --> 00:03:29.514
now removing all AI from my practice.
00:03:29.601 --> 00:03:33.025
Um, I really hadn't any way gotten
past using cursor completion."
00:03:33.029 --> 00:03:37.185
So basically they had, h- hadn't been
doing really much with AI, and is now
00:03:37.204 --> 00:03:39.995
doing less 'cause they didn't really
find the value in what they were doing.
00:03:40.379 --> 00:03:45.037
Uh, which I think is, you know, if it
works, your clients are happy, that's…
00:03:45.066 --> 00:03:46.886
I'm not even like the, "Oh,
you're gonna be left behind."
00:03:46.894 --> 00:03:49.410
Like, no, the measure of whether
you're gonna be left behind is are you
00:03:49.410 --> 00:03:53.025
fulfilling your purpose and are you
creating unnecessary work for yourself?
00:03:53.327 --> 00:03:58.003
It wouldn't shock me if this guy has
had to receive AI, like someone has sent
00:03:58.004 --> 00:04:02.789
over, like, a Claude doc or something
as part of the brief for a website,
00:04:03.189 --> 00:04:04.684
uh, or if he hasn't, he's going to.
00:04:04.696 --> 00:04:08.696
So it's like he probably
not completely AI ascetic.
00:04:09.197 --> 00:04:11.841
But that was really interesting, and
the thing I kind of s- you know, stopped
00:04:11.841 --> 00:04:15.573
myself from saying, again 'cause I didn't
need to, like, put anyone on blast, was,
00:04:15.627 --> 00:04:18.375
yeah, the fact that, like, two things can
be true, but also, like, I think where
00:04:18.375 --> 00:04:22.207
we're getting to, 'cause this is, like,
the same day that, like, Meta and Apple
00:04:22.218 --> 00:04:26.984
had some, partnership or, like, s- at
least like a co-marketing moment where
00:04:27.228 --> 00:04:31.489
Meta has a new open weight model that's
gigantic and you can run it by tethering
00:04:31.489 --> 00:04:36.870
together four of the most expensive
$12,000 Mac Studios, and look, local AI.
00:04:38.602 --> 00:04:40.895
So that's, like, both, it's, like,
the exception that proves the rule.
00:04:40.896 --> 00:04:44.757
It's like the issue isn't
necessarily AI, it's cloud AI.
00:04:44.772 --> 00:04:45.843
Or let's say that's one of, like…
00:04:45.914 --> 00:04:47.228
Obviously slop is an issue.
00:04:47.249 --> 00:04:48.945
People, like, turning off
their brains is an issue.
00:04:48.945 --> 00:04:49.620
That's a whole thing.
00:04:49.955 --> 00:04:56.055
But some of the, like, sustainability,
power, water, uh, you know, land use,
00:04:56.058 --> 00:04:59.458
various things and, you know, like
surveillance, privacy, slurping up all
00:04:59.459 --> 00:05:05.276
the data in the world is more specific
to cloud AI and the big two labs or,
00:05:05.278 --> 00:05:06.684
like, you know, the big hyperscalers.
00:05:06.735 --> 00:05:10.710
And just inevitably, open
source always finds a way.
00:05:10.745 --> 00:05:12.549
Like, the open w- like, we…
00:05:12.639 --> 00:05:16.451
The things that even that I do with AI and
that any of the, many of the folks we know
00:05:16.451 --> 00:05:22.133
are doing with AI don't need a 300, 400
billion parameter model that's running in
00:05:22.139 --> 00:05:23.746
the biggest supercomputer ever created.
00:05:23.746 --> 00:05:24.717
Like, they just don't need that.
00:05:25.070 --> 00:05:29.961
So if the funding for everything were to
just collapse, okay, no one can, like,
00:05:30.080 --> 00:05:33.569
solve all of the hard math problems in
the world using a supercomputer, but
00:05:33.569 --> 00:05:37.416
we can continue to auto-complete HTML
tags, and we can probably do it locally
00:05:38.087 --> 00:05:39.933
with a much lower power envelope.
00:05:39.961 --> 00:05:41.005
And then everything, like…
00:05:41.577 --> 00:05:44.372
I mean, what I ca- where it came down to
on it is, like, I don't think LLMs are
00:05:44.372 --> 00:05:49.350
going away But the whole other thing that
we're actually objecting to, who knows?
00:05:49.369 --> 00:05:50.491
Maybe that's not sustainable.
00:05:50.491 --> 00:05:52.500
But I think it's interesting,
like, like how I like to think
00:05:52.510 --> 00:05:55.697
about it is like, yeah, you know,
it doesn't have to be absolute.
00:05:55.743 --> 00:05:58.762
Two things can be true, but it
is helpful to also tease apart,
00:05:58.773 --> 00:06:00.467
like, what are we actually mad at?
00:06:00.783 --> 00:06:05.370
Um, as it happens, the people who are
doing large language models and the
00:06:05.373 --> 00:06:10.038
people who are doing other nonsense
are the same people at the same time.
00:06:10.451 --> 00:06:13.615
But, you know, other people are then
taking what they've put out in the
00:06:13.616 --> 00:06:17.430
world, adapting it to new purposes,
uh, much cheaper, faster, better.
00:06:17.432 --> 00:06:21.997
And so a few years from now, modulo any
like, you know, whatever economic crash
00:06:22.003 --> 00:06:25.617
might come out of it, um, the technology
may be left behind, and then some people
00:06:25.617 --> 00:06:28.829
might come around to it being like, "All
right, now that it is not, you know,
00:06:29.175 --> 00:06:32.927
threatening people's water supply, I can
try it out and it might be useful or not.
00:06:33.257 --> 00:06:34.382
It's, you know, it's fine."
00:06:35.826 --> 00:06:39.536
Chris Enns: It's like the comparison
that m-made me think of is like the, you
00:06:39.536 --> 00:06:42.845
know, vehicle man-manufacturers right
now, and I'm not a car expert by any
00:06:42.846 --> 00:06:46.748
means, but just like we have the Rivian
electric vehicle and that seems to be
00:06:46.748 --> 00:06:52.105
like pushing towards like usability and
also taking risks and, uh, like, but
00:06:52.105 --> 00:06:58.712
not, uh, at the expense of being a, I
don't know, owned by a crazy nut job,
00:06:58.714 --> 00:06:59.770
let's say for lack of a better word.
00:07:00.258 --> 00:07:00.463
And then
00:07:00.463 --> 00:07:00.734
you still
00:07:00.745 --> 00:07:03.251
David Demaree: we may find out
that RJ Scaringe is also a, a crazy
00:07:03.252 --> 00:07:04.691
nut job, but it doesn't seem so.
00:07:04.731 --> 00:07:07.932
It seems like his craziness is,
you know, promising an e-bike that
00:07:07.932 --> 00:07:10.406
no one can actually buy because
they don't have enough parts, um
00:07:10.435 --> 00:07:10.709
Chris Enns: Right.
00:07:11.999 --> 00:07:15.082
And that's where like, yeah, once
someone gets a, a billion that adds
00:07:15.082 --> 00:07:17.860
the zeros to become a billionaire,
then it's, it feels like that, that
00:07:18.132 --> 00:07:19.591
tends to flip a switch in their brain.
00:07:19.594 --> 00:07:21.180
And, but that can be true.
00:07:21.181 --> 00:07:24.375
Like, we can have vehicles that are
good and healthy and good for the
00:07:24.375 --> 00:07:28.255
environment as well as the ownership
of said vehicles could be less crazy,
00:07:28.255 --> 00:07:32.672
and we can also still have these evil
megalomaniac people who also happen
00:07:32.672 --> 00:07:34.127
to build cars, and both are true.
00:07:34.177 --> 00:07:35.992
And, um, I feel like I…
00:07:36.029 --> 00:07:37.645
That, that does prescribe like a…
00:07:37.655 --> 00:07:42.355
Or the vision of, like, a future where
AI kind of, kind of recedes back to being
00:07:42.357 --> 00:07:46.879
this, like, big ticket thing we have to
fight about and can be like, "Here's a
00:07:46.879 --> 00:07:50.543
useful way to, like, find all the pictures
of your dog in your photo library,"
00:07:50.543 --> 00:07:54.078
which is kind of what most people are
gonna use it for, I feel like, once we
00:07:54.078 --> 00:07:57.944
get past this, like, it's so crazy that
I can make myself look like I was in
00:07:57.946 --> 00:08:00.330
an '80s sitcom, you know, slop stuff.
00:08:00.702 --> 00:08:03.170
David Demaree: have you ever seen
or heard of, there's a show, like
00:08:03.170 --> 00:08:06.181
a series of documentary miniseries
from Ewan McGregor and his best
00:08:06.187 --> 00:08:07.607
friend where they ride motorcycles
00:08:07.744 --> 00:08:09.310
Chris Enns: I've s- I know
what you're talking about.
00:08:09.310 --> 00:08:09.726
I haven't watched it, but
00:08:09.780 --> 00:08:11.211
David Demaree: And like the most
recent one, they were, like, going
00:08:11.212 --> 00:08:12.526
through Europe, and we just…
00:08:12.556 --> 00:08:17.535
Our family is just finishing up the
one before it from, like, 2020, 2021,
00:08:17.567 --> 00:08:19.593
uh, you know, 2020 called Long Way Up.
00:08:20.101 --> 00:08:24.718
And that one is really interesting
because they, as, like, an experiment…
00:08:24.886 --> 00:08:27.483
And bear in mind, this is, like,
before the Rivian trucks and
00:08:27.483 --> 00:08:28.972
stuff, like, fully came to market.
00:08:29.004 --> 00:08:31.386
So, like, they were in development,
but they weren't, like, out, out.
00:08:31.867 --> 00:08:35.693
And so they decided they were gonna
do this whole, like, 15,000-mile ride,
00:08:36.171 --> 00:08:39.758
uh, from the very tip of South America
all the way to LA, but they were
00:08:39.758 --> 00:08:44.437
gonna do it on electric motorcycles
with the support crew in Rivians.
00:08:44.797 --> 00:08:48.274
And these were all, like,
experimental things.
00:08:48.318 --> 00:08:51.462
And certainly, like, taking an
electric motorcycle, like, across
00:08:51.463 --> 00:08:54.547
the US is probably a little bit of
an adventure, and certainly it was.
00:08:54.857 --> 00:08:58.616
To go through, like, the desert and
stuff in South America was like…
00:08:58.616 --> 00:09:00.270
There, there wasn't power.
00:09:00.282 --> 00:09:04.004
There wa- they had to, like, have, like,
things airlifted occasionally from Rivian
00:09:04.005 --> 00:09:05.470
and Harley-Davidson on the equipment.
00:09:05.470 --> 00:09:07.912
It's, like, really f- it's really,
really interesting because it's both,
00:09:07.939 --> 00:09:11.887
like, you know, a travelogue of going
through, uh, some beautiful parts of the
00:09:11.887 --> 00:09:16.863
world, but also the ongoing experiment
of trying to get from place to place
00:09:16.911 --> 00:09:20.930
when they really just do not know if
their bikes and cars are gonna work.
00:09:21.165 --> 00:09:23.771
Chris Enns: Yeah, that's a, actually
a good, uh, reminder of just checking
00:09:23.772 --> 00:09:27.143
that out 'cause it's always one of
those things where it feels, um, I want
00:09:27.148 --> 00:09:31.564
to watch and then I go to Apple TV or
whatever and end up s- because it's…
00:09:31.692 --> 00:09:35.251
I want fantasy or fiction or take
me away from real life, and so…
00:09:35.456 --> 00:09:37.714
David Demaree: I, I will say like
the, like from the like pragmatic
00:09:37.715 --> 00:09:39.571
thing, like So this morning, like,
00:09:39.959 --> 00:09:41.538
just to kind of anchor it
in, like, a real story.
00:09:41.538 --> 00:09:45.937
So, like, my wife is, like, the education
director for, um, our local yarn store,
00:09:45.939 --> 00:09:49.920
which is literally called The Local
Yarn Store, and this weekend, literally
00:09:49.921 --> 00:09:53.544
tomorrow, is the Rhinebeck Sheep and
Wool Festival in upstate New York.
00:09:53.573 --> 00:09:57.799
So people from New York and New Jersey,
especially, like, in our community,
00:09:58.078 --> 00:10:00.630
they'll, like, buy their tickets, but
that'll include, like, a bus trip.
00:10:00.631 --> 00:10:04.761
So everyone will, like, gather together
from the local yarn store community,
00:10:04.787 --> 00:10:07.496
get on a bus to go do that, and then
come back at the end of the day.
00:10:07.498 --> 00:10:10.239
And so she's coordinating that,
and as part of that, she's got
00:10:10.240 --> 00:10:13.266
a Google sheet with a bunch of,
um, you know, email addresses.
00:10:13.275 --> 00:10:17.283
And, you know, a lot of small businesses,
this is like, you know, I'm not even
00:10:17.283 --> 00:10:19.727
gonna pretend to pitch my business,
but, like, one of the things that's,
00:10:19.727 --> 00:10:23.022
like, concerning to me, even when I
was in big tech or now that I do an
00:10:23.027 --> 00:10:26.949
agency thing, is for all the people
who are like, "I've got this perfect
00:10:26.949 --> 00:10:30.427
HubSpot system that's wired into all the
different things, and it's automated,
00:10:30.428 --> 00:10:33.391
and there's an agent or something,"
most people are like, if they've got
00:10:33.391 --> 00:10:36.984
a Google sheet that's well-organized,
they are ahead of the curve, man.
00:10:37.388 --> 00:10:42.642
So she's got all these email addresses,
but how do you get a comma-separated
00:10:42.642 --> 00:10:45.679
list of email addresses to paste
into the BCC field of Gmail?
00:10:45.708 --> 00:10:46.766
And one of the first things that…
00:10:46.767 --> 00:10:53.047
So I asked ChatGPT, I asked my ChatGPT
dot, my agent, uh, in part because it
00:10:53.049 --> 00:10:56.935
is so much easier to just, like, frame
a question, have it go off and maybe
00:10:56.935 --> 00:11:01.153
do multiple searches, and especially if
it's, like, part of a whole other thing,
00:11:01.169 --> 00:11:02.416
like your cloud coding or something.
00:11:02.419 --> 00:11:04.574
Basically, if you're not, like,
going out of your way, but you're
00:11:04.576 --> 00:11:07.848
kind of, like, using up the same
pool of credits as other stuff.
00:11:07.848 --> 00:11:08.066
I don't know.
00:11:08.066 --> 00:11:10.878
It seems like it's, you know,
you're already kind of in for a
00:11:10.879 --> 00:11:12.273
penny, in for a pound at that point.
00:11:12.274 --> 00:11:13.039
But it can…
00:11:13.220 --> 00:11:17.044
Usually, it used to be very bad, like,
AI would hallucinate or give you bad
00:11:17.044 --> 00:11:20.367
information, and they've gotten a lot
better in that, and especially you can
00:11:20.368 --> 00:11:22.029
tell because the answers will be short.
00:11:22.332 --> 00:11:23.868
So it was like, you have two options.
00:11:23.870 --> 00:11:27.594
The short one is here is the formula you
put into Google Sheets to take a list of
00:11:27.595 --> 00:11:31.486
columns and output, um, a comma-separated
result, which ended up being the solution.
00:11:31.858 --> 00:11:35.625
I learned something, and it, like, took
what was kind of a, like, how do we do
00:11:35.625 --> 00:11:40.083
this, how do we even know what question to
ask, and got to a real solution for a real
00:11:40.084 --> 00:11:42.263
person in, like, five minutes or less.
00:11:42.454 --> 00:11:46.116
Um, and it's, like, hard to
totally complain about that.
00:11:46.410 --> 00:11:51.011
it is fair to say, like, you know, is
it worth the compute and energy costs?
00:11:51.031 --> 00:11:54.216
Um, I do think that things like that
are almost, like, around the margins.
00:11:54.216 --> 00:11:56.738
Like, they're spending trillions of
dollars building what they think is
00:11:56.738 --> 00:12:01.339
super intelligence to take us to the
stars, and so, you know, be concerned
00:12:01.339 --> 00:12:04.626
about all the compute they're throwing
at solving the math problem so that
00:12:04.626 --> 00:12:06.215
we don't need mathematicians anymore.
00:12:06.593 --> 00:12:12.638
Uh, you know, the whatever two kilowatts,
two nanowatts to solve my, uh, spreadsheet
00:12:12.640 --> 00:12:16.655
problem I'm, I'm happy to, you know,
take a little bit out of, uh, Sam
00:12:16.657 --> 00:12:18.517
Altman's pocket or, uh, Dario's pocket
00:12:19.086 --> 00:12:19.327
Chris Enns: Yeah.
00:12:21.069 --> 00:12:21.642
Yes, exactly.
00:12:21.680 --> 00:12:24.734
Yeah, and that's, I think that's where
the argument against using it is gonna
00:12:24.734 --> 00:12:28.618
be really hard 'cause similarly, like
for my, my wife in the, her job at an
00:12:28.629 --> 00:12:32.947
office, a regular office that isn't--
They're not like AI-focused or AI,
00:12:32.981 --> 00:12:35.257
anti-AI, like they're just doing stuff.
00:12:35.659 --> 00:12:39.971
And the trouble is the tools, like
they're very Office 365 or whatever
00:12:39.971 --> 00:12:41.103
Microsoft's calling it these days.
00:12:41.508 --> 00:12:45.609
And so it's there, and then they have
to use it almost, and then it just
00:12:45.609 --> 00:12:47.130
gets used and that becomes the norm.
00:12:47.132 --> 00:12:49.799
That becomes just like find and
replace was for us, you know,
00:12:49.801 --> 00:12:51.050
previous generation or whatever.
00:12:51.479 --> 00:12:55.082
And so there isn't even a-- There's
not debate about AI usage or not.
00:12:55.086 --> 00:12:58.041
It's just that's the tool that's in
front of me that I have to use to do the
00:12:58.043 --> 00:12:59.587
spreadsheet thing that I need to get done.
00:12:59.587 --> 00:13:04.038
And, and unfortunately for if
there's any a- anti-AI crowd
00:13:04.055 --> 00:13:07.198
involved in that, it's really hard
to move that rock out of the way.
00:13:07.198 --> 00:13:08.347
I feel like it, it's just
00:13:08.367 --> 00:13:09.150
David Demaree: Yeah.
00:13:09.358 --> 00:13:14.409
Well, and th- this is where like, um, in
mostly bad ways a lot of AI folks are…
00:13:14.424 --> 00:13:17.471
Uh, this is like, almost less an AI thing
and more like a user experience, 'cause I,
00:13:17.510 --> 00:13:19.558
I don't remember when I first heard this.
00:13:19.897 --> 00:13:22.362
I'd love to think I invented this thought.
00:13:22.576 --> 00:13:26.248
The, like, real anecdote that's closest to
it is I remember reading years ago where,
00:13:26.249 --> 00:13:29.719
like, there was some anecdote where the
head of admissions, dean of admissions for
00:13:29.722 --> 00:13:34.565
MIT was speaking at some kind of event,
and then some students, um, who might
00:13:34.808 --> 00:13:38.798
wanna go to MIT came up after and talked
about the difficulties that they'd had
00:13:38.894 --> 00:13:41.849
with the, um, with the application blank.
00:13:41.849 --> 00:13:43.447
Like, at that time it was a paper form.
00:13:43.962 --> 00:13:48.078
And the concern was that they wouldn't
be able to say everything they needed
00:13:48.079 --> 00:13:52.072
to say in response to an essay in
the specific space on the page.
00:13:52.638 --> 00:13:56.024
And this was really confusing to the,
to the dean because she was like, "Well,
00:13:56.025 --> 00:13:57.842
but, no, you can add additional pages.
00:13:57.863 --> 00:13:59.168
You can do whatever you need to do.
00:13:59.178 --> 00:14:00.325
You can say 'see attached.'"
00:14:00.681 --> 00:14:03.855
But because the, like, what you
can see is the blank, they assume
00:14:03.859 --> 00:14:05.102
like, no, it has to be here.
00:14:05.102 --> 00:14:07.975
And it's similar to, like,
if you give someone a button.
00:14:08.192 --> 00:14:10.336
I think this has, like, come up
in the context, especially like
00:14:10.336 --> 00:14:13.898
B2B software design or any kind
of like, you know, control plane,
00:14:13.908 --> 00:14:15.305
whether it's like Microsoft Word.
00:14:15.318 --> 00:14:17.976
Like, why, why is the
Microsoft Word ribbon horrible?
00:14:18.656 --> 00:14:22.226
It is convenient if you actually
do need even, like, 50% of
00:14:22.229 --> 00:14:23.185
the things that are on there.
00:14:23.198 --> 00:14:27.143
But it also creates this pressure
that if you're not using those
00:14:27.148 --> 00:14:29.230
things, you must be bad at software.
00:14:29.310 --> 00:14:33.788
Because, like, if you're an airplane
pilot, presumably you're doing something
00:14:33.788 --> 00:14:37.224
with all the switches and all the,
the sticks and all the gauges that
00:14:37.225 --> 00:14:38.819
are in an airplane cock- cockpit.
00:14:39.124 --> 00:14:40.083
You … Maybe you're not.
00:14:40.095 --> 00:14:43.089
Maybe you're mainly just working with,
like, two or three, and then they're
00:14:43.092 --> 00:14:46.138
there in case you need it, 'cause you're
in midair and you need robustness.
00:14:46.613 --> 00:14:49.257
Uh, you're not flying an airplane
if you're in Microsoft Word.
00:14:49.313 --> 00:14:53.523
Um, and you could probably go root
around in menus if you needed to.
00:14:53.536 --> 00:14:59.300
So the design, the object of good design
is what are the opinionated, like,
00:14:59.305 --> 00:15:03.445
top 10% of the things that need to be
visible, and then how, how do you kind
00:15:03.445 --> 00:15:06.138
of shift into and find the other stuff?
00:15:06.142 --> 00:15:08.918
And, like, the thing with a lot
of Microsoft and a lot of, like,
00:15:08.924 --> 00:15:13.568
enterprise software design is a complete
unwillingness to have an opinion, because
00:15:13.569 --> 00:15:18.758
if someone somewhere in their customer
ecosystem says, "But where's this thing?
00:15:18.831 --> 00:15:22.163
I find it too hard to click through two
menus," they might be the only person
00:15:22.164 --> 00:15:25.204
with that opinion, and they might
not even be from the biggest account.
00:15:25.427 --> 00:15:29.316
But that might also be the only data
point that designer or product manager
00:15:29.316 --> 00:15:32.425
got, and so they're gonna be like,
"Well, we don't wanna get anything
00:15:32.425 --> 00:15:34.181
that might indicate losing an account.
00:15:34.182 --> 00:15:35.958
So we're just gonna put it all up here.
00:15:36.397 --> 00:15:37.846
We're gonna put it all
in front of everyone."
00:15:37.864 --> 00:15:40.868
And you can game that by saying like, "All
right, we're gonna put stuff in front of
00:15:40.868 --> 00:15:44.544
everybody so that you feel like you have
to reason about it as part of your work.
00:15:44.544 --> 00:15:47.164
So why don't we give you
a big old Gemini sidebar?"
00:15:47.173 --> 00:15:52.149
Now, when you try typing in it and saying,
"Hey, Gemini, can you redo all the fonts?
00:15:52.195 --> 00:15:55.532
Every place that I'm currently using
Arial, can you change it to Public Sans?"
00:15:55.836 --> 00:15:56.962
No, it'll refuse to do it.
00:15:57.044 --> 00:15:57.801
It, it can't do it.
00:15:58.213 --> 00:16:02.563
Um, but you will have engaged with
the Gemini chatbot, which made their
00:16:02.563 --> 00:16:03.984
product manager's metrics go up.
00:16:03.984 --> 00:16:05.803
So from their perspective, job done.
00:16:06.092 --> 00:16:08.700
Chris Enns: There's probably some,
uh, joke or irony in Microsoft
00:16:08.700 --> 00:16:12.049
also making their flight simulator
app just in your analogy there.
00:16:13.372 --> 00:16:14.026
The, uh,
00:16:14.081 --> 00:16:15.925
David Demaree: if you look at
Microsoft Flight Simulator,
00:16:15.925 --> 00:16:17.448
does it have any extra controls?
00:16:17.511 --> 00:16:17.854
No.
00:16:17.965 --> 00:16:19.380
It is, it is a well-designed app
00:16:19.861 --> 00:16:22.656
Chris Enns: there's no ribbon across
the top of the cockpit probably.
00:16:23.071 --> 00:16:26.115
yeah, the, the idea that like because
it's there then it gets used and
00:16:26.116 --> 00:16:28.688
then so because it's being used
then it means it's good and that
00:16:28.690 --> 00:16:30.697
sort of cycle that also transpire.
00:16:30.723 --> 00:16:34.005
And I, I feel that frustration
of like how AI gets inserted into
00:16:34.029 --> 00:16:35.610
every app in an obnoxious way.
00:16:35.639 --> 00:16:40.116
Like the Riverside that we're in
right now, the editor side prompts me
00:16:40.164 --> 00:16:45.040
every second I open any new tab inside
the app, "Hey, here's the chatbot.
00:16:45.047 --> 00:16:45.649
Here's the thing.
00:16:45.651 --> 00:16:46.233
Here's how you…"
00:16:46.477 --> 00:16:47.853
And it's like it's useless.
00:16:47.880 --> 00:16:49.443
It doesn't do the thing you need it to do.
00:16:49.586 --> 00:16:53.160
I don't know who's using it to actually
edit video 'cause it's awful in broad…
00:16:53.186 --> 00:16:56.419
Like if you're trying to do anything
remotely professional obviously
00:16:56.419 --> 00:16:59.136
if you're just trying to generate
something quick you can do some of that.
00:16:59.141 --> 00:17:02.010
But, and that's, I assume that's
just being played out over and
00:17:02.010 --> 00:17:03.102
over in offices all over the
00:17:03.244 --> 00:17:03.597
David Demaree: yeah.
00:17:03.955 --> 00:17:07.057
Chris Enns: the AI thing is like
it works for 2% of the situations
00:17:07.057 --> 00:17:10.738
and then it's sitting there wasting
credits for 98% of the time.
00:17:11.027 --> 00:17:11.515
yeah, we got
00:17:11.537 --> 00:17:12.766
David Demaree: having
a similar thought too
00:17:12.861 --> 00:17:13.883
Chris Enns: without talking about AI.
00:17:14.245 --> 00:17:15.131
David Demaree: I, yeah, we can stop.
00:17:15.133 --> 00:17:15.811
I will say like the…
00:17:15.888 --> 00:17:17.834
I'm sorry, 'cause it keeps
coming up with ideas.
00:17:18.360 --> 00:17:21.041
'Cause like one thing I had spent
on my mind, 'cause like us…
00:17:21.080 --> 00:17:26.297
Okay, so alongside Meta having their
new open weight model that will run on
00:17:26.303 --> 00:17:30.288
$40,000 worth of equipment in the same
way the, the, the other big models will
00:17:30.288 --> 00:17:34.282
run on millions of dollars of equipment,
uh, they also have their closed model, M-
00:17:34.332 --> 00:17:37.451
Muse Wonder, and they put out a bunch of
benchmarks, and the benchmarks say, "Oh my
00:17:37.452 --> 00:17:41.007
gosh, Muse Wonder is as good as Fable, as
good as Astra, as good as the other ones."
00:17:41.332 --> 00:17:42.836
And then you like go into…
00:17:42.846 --> 00:17:46.132
'Cause I, at this point, I follow a
few, like, AI content creators, like
00:17:46.133 --> 00:17:51.765
Claire Vo is one, Simon Willison, who
was like a social, uh, data hacker
00:17:51.766 --> 00:17:55.054
before he was an AI person, uh, and
like the Everyblog folks, and, you
00:17:55.054 --> 00:17:56.359
know, they'll have the vibe check.
00:17:56.657 --> 00:17:57.464
And so that's always good.
00:17:57.465 --> 00:17:58.219
Like, it…
00:17:58.389 --> 00:18:00.128
Where it came around from is like,
you know, if you see a bunch of
00:18:00.130 --> 00:18:03.230
reporting around AI benchmarks,
don't trust the benchmarks.
00:18:03.252 --> 00:18:06.664
Because at this point, the companies are
really good at like teaching to the test.
00:18:06.699 --> 00:18:10.531
And so, you know, it's not that
it like, not that the Muse models
00:18:10.541 --> 00:18:15.144
or Gemini 4 Argon or GPT 7.2
00:18:15.191 --> 00:18:19.562
Copacabana, whatever, are bad models,
but they don't tell you whether
00:18:19.564 --> 00:18:22.139
they're actually going to like
be versatile enough to help you.
00:18:22.414 --> 00:18:26.424
And so you'll try to say like do a vanilla
website task with Muse Wonder, which
00:18:26.424 --> 00:18:30.014
is supposed to be as good as Fable, and
it will completely redo your design.
00:18:30.015 --> 00:18:31.440
This literally happened to Claire.
00:18:31.902 --> 00:18:35.553
Uh, redo your design, it'll make it
vanilla, it'll ruin your words, it'll go
00:18:35.554 --> 00:18:39.130
outside of the brief, and it's because
the me- whatever benchmark they use
00:18:39.131 --> 00:18:42.950
to say this is a good model, it has a
narrow list of things that it's evaling
00:18:42.953 --> 00:18:44.667
for, and anything else isn't gonna work.
00:18:44.679 --> 00:18:48.353
And so, but the, the difference
there is like, okay, why then at the
00:18:48.355 --> 00:18:51.120
same score do you have like Opus 5.5
00:18:51.122 --> 00:18:53.976
from Anthropic, which people seem
to now think is the gold standard
00:18:53.977 --> 00:18:57.033
model, and then these other ones that
score just as well but are horrible?
00:18:57.034 --> 00:19:00.196
And I think it's because anytime
you're in Claude Code, every once in
00:19:00.196 --> 00:19:02.751
a while they'll put up that little
bar that says, "How are we doing?"
00:19:04.871 --> 00:19:09.927
not that it's ever, like, perfect, but
when you've been out in the world putting
00:19:09.927 --> 00:19:13.620
this in front of people for three or four
years and getting some data of just even,
00:19:13.621 --> 00:19:18.655
like, that little, like, you know, um,
like hit three if you're, if you're happy.
00:19:19.163 --> 00:19:22.625
Um, even if they're not, like, retaining
a bunch of data, they can still get
00:19:22.626 --> 00:19:25.509
some, like, "All right, this is what
good looks like," without us having to,
00:19:25.510 --> 00:19:27.320
like, do a whole bunch of extra research.
00:19:27.330 --> 00:19:28.912
They're ca- getting ongoing signal.
00:19:29.300 --> 00:19:31.641
So that makes you just, like, you can't…
00:19:32.087 --> 00:19:34.918
First of all, like, you know, like
any other things with UX, like, you
00:19:34.918 --> 00:19:37.417
get good signals and you actually
listen to them, you're probably
00:19:37.417 --> 00:19:38.450
gonna make a better product.
00:19:38.954 --> 00:19:44.135
Um, and even if the way you do that with
AI is through this arcane post-training
00:19:44.152 --> 00:19:48.236
reinforcement learning, like, kind of
academic thing, data is still data.
00:19:48.251 --> 00:19:51.508
And if you're kind of, like, prioritizing
what is gonna make people feel like
00:19:51.509 --> 00:19:55.397
they're getting stuff done, then you
can start to optimize in that direction.
00:19:55.403 --> 00:19:59.393
But then you can also cheat on the
test and appear to have really high
00:19:59.398 --> 00:20:02.759
benchmarks, and the proof is gonna
be when you ask this to do something,
00:20:02.995 --> 00:20:04.239
is it going to completely screw up?
00:20:04.239 --> 00:20:07.253
I think this is true a lot of the time
with, like, Copilot, which is weird
00:20:07.253 --> 00:20:11.023
because Copilot stuff has been maybe
one of s- some of the first stuff in
00:20:11.024 --> 00:20:15.618
the market, will rarely do what you
want it to do and doesn't really have
00:20:15.619 --> 00:20:19.217
a way for you to tell it when it's
screwed up, and yet it's still there.
00:20:19.222 --> 00:20:23.662
So Microsoft has had myriad opportunities
to make this better, and this is, like,
00:20:23.684 --> 00:20:26.948
for better or worse, one thing you can
say about OpenAI and Anthropic is they've
00:20:27.084 --> 00:20:31.258
listened and they've continually tried
to, like, meet people where they are and
00:20:31.259 --> 00:20:34.420
serve their use cases as opposed to just
being like, "Well, here's a benchmark.
00:20:34.456 --> 00:20:35.687
Give us a trillion dollars."
00:20:36.938 --> 00:20:37.117
Chris Enns: Yeah.
00:20:38.637 --> 00:20:42.953
Sounds like a good business model to
me, but the I honestly have no idea
00:20:42.968 --> 00:20:44.619
David Demaree: must be really
great to be Alexander Wang.
00:20:44.619 --> 00:20:45.004
I'll just say that.
00:20:45.369 --> 00:20:45.566
Chris Enns: Yeah.
00:20:46.995 --> 00:20:51.343
I have no idea if half the model names you
referenced there were jokes or serious.
00:20:51.370 --> 00:20:53.485
That's the one, my one takeaway from this.
00:20:53.569 --> 00:20:54.853
And, uh, and
00:20:54.855 --> 00:20:55.340
actually I'm just
00:20:55.343 --> 00:20:57.491
David Demaree: need to troll people
with fake model names more often.
00:20:57.748 --> 00:20:59.593
I only, I only invented two just
00:20:59.645 --> 00:20:59.875
Chris Enns: yeah.
00:21:01.324 --> 00:21:04.373
That's the, uh, the pop quiz at the end of
the episode is guessing which model names
00:21:04.374 --> 00:21:06.943
were real and which ones weren't, and
you get 10 points if you get them right.
00:21:07.242 --> 00:21:07.481
Yeah.
00:21:07.513 --> 00:21:11.117
There might be some like follow-up where
it's just an episode of me, getting you
00:21:11.117 --> 00:21:12.629
to define term, or I guess I could just,
00:21:12.629 --> 00:21:14.785
David Demaree: That'd be like th-
this might, this might be my pitch.
00:21:14.785 --> 00:21:17.231
It'd actually be on Shop
Talk with, with the guys.
00:21:17.325 --> 00:21:20.135
Uh, it's just being like, "All right,
just start asking me about the AI terms
00:21:20.135 --> 00:21:21.786
and I will explain them for 45 minutes."
00:21:23.019 --> 00:21:25.809
Chris Enns: Which then, like a r- this
is the funny thing of like somehow
00:21:25.810 --> 00:21:28.710
that ends up, you know, filtering
back into the AI machine and then,
00:21:28.772 --> 00:21:31.582
you know, in, in a six months time
someone's like, "What does this mean?"
00:21:31.586 --> 00:21:36.454
And it's like references David's
appearance on Shoptalk show as the source
00:21:36.460 --> 00:21:39.547
for this joke AI name, and we-- the cycle
00:21:39.552 --> 00:21:40.080
David Demaree: Be like, yeah.
00:21:40.081 --> 00:21:43.796
And, and eval, it's only an, a, uh,
it's only an eval if it comes from
00:21:43.801 --> 00:21:46.464
the, uh, Evaloque region of France.
00:21:46.537 --> 00:21:48.576
Otherwise, it's a
sparkling acceptance test
00:21:49.443 --> 00:21:49.631
Chris Enns: All right.
00:21:49.631 --> 00:21:50.157
Thanks, David.
00:21:50.237 --> 00:21:52.535
Uh, where can folks find you,
follow you, you know, all those
00:21:52.535 --> 00:21:54.025
things before we sign off?
00:21:54.334 --> 00:21:59.700
David Demaree: Uh, definitely, um, my
website, uh, Demaree, D-E-M-A-R-E-E.me.
00:22:00.194 --> 00:22:02.402
Uh, and at the very, very
bottom, you'll find a link to
00:22:02.428 --> 00:22:03.971
my business, Bits and Letters.
00:22:04.037 --> 00:22:08.290
Um, and, uh, there's a blog on my
personal site, blog on the website.
00:22:08.308 --> 00:22:09.584
There's my social media.
00:22:09.616 --> 00:22:13.747
I'm on, uh, Elon's blog site, uh,
the formerly known as Twitter.
00:22:13.748 --> 00:22:17.163
I'm on Bluesky, I'm on LinkedIn,
I'm on all the places, uh, and
00:22:17.163 --> 00:22:18.022
you can find my links there
00:22:18.341 --> 00:22:20.798
Chris Enns: um, as always, you can
feel free to reply to this post on
00:22:20.798 --> 00:22:22.762
social media, wherever you saw it,
or in the comments on YouTube, or
00:22:22.763 --> 00:22:24.735
you can email me chris.enns@gmail.com
00:22:24.805 --> 00:22:27.649
you're not already subscribed,
visit daily.chrisenns.com/subscribe
00:22:27.692 --> 00:22:30.381
to see all the ways you can subscribe
to the show, or just look for "Dailyish"
00:22:30.396 --> 00:22:31.326
wherever you get your podcasts.
00:22:31.647 --> 00:22:32.388
Thank you for listening.
00:22:32.435 --> 00:22:33.339
I hope you'll be back tomorrow.
00:22:33.342 --> 00:22:33.987
Have a great day.
00:22:34.035 --> 00:22:34.485
Bye.