TRANSKRIPT 🔗
WEBVTT
00:00:00.031 --> 00:00:03.080
It feels more now like we are living
00:00:03.080 --> 00:00:05.957
in the future. The empowerment that it's
00:00:05.957 --> 00:00:08.975
given to a lot of people to bring their ideas to
00:00:08.975 --> 00:00:11.810
life. I don't want to anthropomorphize the AI too
00:00:11.810 --> 00:00:14.675
much, but it's like a colleague that you
00:00:14.675 --> 00:00:17.845
gain trust with. We, I think all agree
00:00:17.845 --> 00:00:20.688
about the excitement, the incredible
00:00:20.688 --> 00:00:23.706
speed that things are occurring at, but what are
00:00:23.706 --> 00:00:26.874
people underestimating? I feel like the bad guy
00:00:26.874 --> 00:00:29.700
when you and Kurt are saying you're more
00:00:29.700 --> 00:00:32.118
optimist. I'm a little more skeptical
00:00:47.360 --> 00:00:50.298
Yo, yo, yo, this is MK back with another
00:00:50.298 --> 00:00:53.165
episode of Ready, Set, Runtime, a
00:00:53.165 --> 00:00:56.241
podcast from AHEAD. Today
00:00:56.241 --> 00:00:58.905
I'm joined by Kurt Boden and Habeeba Mansour, and
00:00:58.905 --> 00:01:01.998
we're gonna talk about the gap between the demo
00:01:01.998 --> 00:01:05.164
and the deploy. Kurt and Habeeba, could you
00:01:05.164 --> 00:01:08.325
introduce yourselves? Yeah, I can dive in first,
00:01:08.325 --> 00:01:11.427
Habeeba. I am Kurt Boden. I'm a principal
00:01:11.427 --> 00:01:12.645
technical consultant.
00:01:13.699 --> 00:01:15.729
I've been with AHEAD for a little over a year, but
00:01:15.729 --> 00:01:18.322
I've been in the AI space for over 10 years.
00:01:19.193 --> 00:01:21.946
back before AI was called AI, I was a predictive
00:01:21.946 --> 00:01:24.276
modeler, and just followed it down the rabbit hole
00:01:24.276 --> 00:01:24.683
from there.
00:01:25.729 --> 00:01:28.199
it's been interesting to see how much progress has
00:01:28.199 --> 00:01:31.137
been made in the space, in the last 10 years, five
00:01:31.137 --> 00:01:34.779
years, one year. but yeah, I'm excited to be here.
00:01:35.844 --> 00:01:38.801
Amazing. Habeeba I'm Habeeba. I'm a senior
00:01:38.801 --> 00:01:41.660
technical consultant in our AI practice at
00:01:41.660 --> 00:01:44.431
AHEAD. I've been at AHEAD for
00:01:44.431 --> 00:01:47.187
coming up on the third year
00:01:47.187 --> 00:01:49.844
now, and, similar to Kurt,
00:01:50.836 --> 00:01:53.766
my background was primarily in machine learning,
00:01:53.766 --> 00:01:56.785
data science, and robotics. That's what I did
00:01:56.785 --> 00:01:59.972
before I joined AHEAD. and then after
00:01:59.972 --> 00:02:02.470
I joined, I started doing a lot more in the
00:02:02.470 --> 00:02:05.600
agentic space, AI SDLC, and just seeing
00:02:05.600 --> 00:02:07.939
where the AI boom takes us.
00:02:08.582 --> 00:02:10.958
Yeah, amazing. Thank you for joining us today. I
00:02:10.958 --> 00:02:11.659
appreciate it.
00:02:12.731 --> 00:02:15.689
Kurt, why don't you start, by sharing project
00:02:15.689 --> 00:02:18.694
work that you're working on presently,
00:02:18.694 --> 00:02:21.522
that demonstrates how our
00:02:21.522 --> 00:02:24.349
clients or at AHEAD we're
00:02:24.349 --> 00:02:27.340
adopting AI. A few things come to
00:02:27.340 --> 00:02:30.345
mind, and even just like high level, it's been
00:02:30.345 --> 00:02:32.905
interesting seeing the different types of projects
00:02:32.905 --> 00:02:34.560
that come through our organization.
00:02:35.681 --> 00:02:37.849
Habeeba and I have talked through this a little
00:02:37.849 --> 00:02:40.824
bit, but it's like you really have to be very full
00:02:40.824 --> 00:02:43.521
stack, as an engineer in the AI space.
00:02:43.811 --> 00:02:46.770
there are projects that I've worked on that
00:02:46.770 --> 00:02:48.788
have been very much, on-prem,
00:02:49.427 --> 00:02:52.172
standing up, local infrastructure,
00:02:52.172 --> 00:02:54.368
like, standing up the core,
00:02:54.368 --> 00:02:57.057
architecture for model serving, and
00:02:57.057 --> 00:03:00.100
deployments. I've worked a lot with like
00:03:00.100 --> 00:03:02.686
coding assistance and, accelerated
00:03:02.688 --> 00:03:05.336
development, with clients. And most
00:03:05.336 --> 00:03:08.231
recently, or very recently, I've been
00:03:08.231 --> 00:03:11.278
working through a big project for building
00:03:11.278 --> 00:03:14.198
out, an agent that does some pretty advanced
00:03:14.198 --> 00:03:17.009
operations to put together like an outside in
00:03:17.009 --> 00:03:19.977
analysis, for companies. And so just in the
00:03:19.977 --> 00:03:23.156
space, it's a pretty br-broad spectrum, I
00:03:23.156 --> 00:03:25.961
think, for us as engineers right now, just
00:03:25.961 --> 00:03:28.944
picking up a lot of very different projects. And
00:03:28.944 --> 00:03:32.020
so you can go from, bare metal all the
00:03:32.020 --> 00:03:35.018
way to building a front-end interface,
00:03:35.018 --> 00:03:38.377
almost in the same project. It's pretty amazing to see.
00:03:39.378 --> 00:03:41.956
Yeah. More than ever now, understanding the
00:03:41.956 --> 00:03:44.363
platform requirements, the infrastructure
00:03:44.363 --> 00:03:46.783
requirements for the deliveries that we're doing.
00:03:47.244 --> 00:03:50.262
In the past, I feel like, we had a very good
00:03:50.262 --> 00:03:53.306
understanding of those architectures, and they
00:03:53.306 --> 00:03:56.162
really were mostly unchanged for a long time.
00:03:56.658 --> 00:03:59.104
And now I think we need to understand the software
00:03:59.104 --> 00:04:01.436
layer and how that interacts with the hardware
00:04:01.436 --> 00:04:04.350
layer. and then-- and additionally, like you said,
00:04:04.350 --> 00:04:06.690
these things are married so closely together.
00:04:06.863 --> 00:04:08.422
Habeeba, what about you? What kind of things are
00:04:08.422 --> 00:04:08.887
you working on?
00:04:10.233 --> 00:04:12.850
Yeah. So what comes to mind for more research
Load full transcript
00:04:12.850 --> 00:04:16.037
projects that I've been working on is really we've
00:04:16.037 --> 00:04:19.102
seen an uptick in AI SDLC. A lot
00:04:19.102 --> 00:04:22.283
of companies have bought all these
00:04:22.283 --> 00:04:25.260
tools, and they really want to see
00:04:25.260 --> 00:04:28.217
the benefits of them. They wanna see efficiency go
00:04:28.217 --> 00:04:30.504
up. They want to see their engineers and their
00:04:30.504 --> 00:04:33.227
talent start using these tools, and those are two
00:04:33.227 --> 00:04:36.038
of the projects that I was recently on. And I
00:04:36.038 --> 00:04:39.109
think in particular with all of the security
00:04:39.109 --> 00:04:42.107
vulnerabilities that are happening day to day
00:04:42.107 --> 00:04:45.114
now that we hear every week happening with AI, a
00:04:45.114 --> 00:04:48.021
lot of companies are looking to use AI SDLC to
00:04:48.021 --> 00:04:50.993
enhance their security postures. I think,
00:04:50.993 --> 00:04:53.677
again, with Anthropic coming out with
00:04:53.677 --> 00:04:56.741
Mythos, that's really caused a lot of companies
00:04:56.741 --> 00:04:59.648
to prioritize their security findings and
00:04:59.648 --> 00:05:01.735
enhance their security findings. That's one
00:05:01.735 --> 00:05:04.543
project I'm on now, where they're really trying to
00:05:04.543 --> 00:05:07.578
use the AI SDLC process to remediate and find any
00:05:07.578 --> 00:05:09.809
security vulnerabilities they have.
00:05:10.769 --> 00:05:13.104
It's so interesting that we're returning first to
00:05:13.104 --> 00:05:16.046
the language of the SDLC, because for at least 10
00:05:16.046 --> 00:05:19.055
years, people had just started saying agile,
00:05:19.374 --> 00:05:21.579
to replace sort of the SDLC idea.
00:05:22.657 --> 00:05:25.164
and I like the return to the language. I've been
00:05:25.164 --> 00:05:27.924
saying XDLC because I think it's bigger than
00:05:27.924 --> 00:05:30.106
software. I think it's about the overall IT
00:05:30.106 --> 00:05:32.926
delivery. when each of you think
00:05:32.926 --> 00:05:35.974
about your own use of AI over the
00:05:35.974 --> 00:05:38.980
last 18 months Do you s- find that your
00:05:38.980 --> 00:05:42.140
own adoption has changed over the last 18 months?
00:05:42.537 --> 00:05:45.639
Do you start to have, a cohort of agents
00:05:45.639 --> 00:05:48.764
that are well-defined that you are leaning on?
00:05:50.510 --> 00:05:53.475
are you starting to understand, a
00:05:53.475 --> 00:05:55.393
tool set and a stack?
00:05:56.442 --> 00:05:59.377
So in the last 18 months, I think the landscape
00:05:59.377 --> 00:06:02.434
has changed so significantly that, yes, for
00:06:02.434 --> 00:06:05.496
sure my tool adoption has changed. Along
00:06:05.496 --> 00:06:08.227
with it, I think 18 months ago, we
00:06:08.227 --> 00:06:11.340
didn't have skills, for example. We didn't
00:06:11.340 --> 00:06:14.042
have a lot of these things that are now considered
00:06:14.042 --> 00:06:16.835
just day-to-day and everyone's using.
00:06:17.149 --> 00:06:19.726
And again, 18 months ago, you had to be a lot more
00:06:19.726 --> 00:06:22.572
prescriptive with AI. You had to really review the
00:06:22.572 --> 00:06:25.489
output a lot more, add a lot more guardrails, a
00:06:25.489 --> 00:06:28.600
lot more definitions. Whereas now, if
00:06:28.600 --> 00:06:31.504
you have a good setup, you have…
00:06:31.504 --> 00:06:34.469
You can be a lot less hands-on
00:06:34.469 --> 00:06:37.494
with AI. I've personally, more recently,
00:06:37.494 --> 00:06:40.632
just been using agents that can be deployed in
00:06:40.632 --> 00:06:43.555
Cursor or in Claude Code, and then
00:06:43.555 --> 00:06:46.561
just using skills that, again, with
00:06:46.561 --> 00:06:48.912
the open source community, I think that's been one
00:06:48.912 --> 00:06:51.377
of the best things to see. There's so many people
00:06:51.377 --> 00:06:54.103
that spend a lot of time creating these amazing
00:06:54.103 --> 00:06:57.034
skills, and you don't want to reinvent the wheel.
00:06:57.206 --> 00:07:00.227
Just using what, again, the community has
00:07:00.227 --> 00:07:02.528
provided has been really great,
00:07:02.914 --> 00:07:05.930
again, with any AI tool that you use. And
00:07:05.930 --> 00:07:08.427
then honestly, with the space where we work, one
00:07:08.427 --> 00:07:11.075
of the nicest things is that we get to experiment
00:07:11.075 --> 00:07:13.595
with different tools and switch when one gets
00:07:13.595 --> 00:07:16.381
better than the other, and that's been great to,
00:07:16.381 --> 00:07:19.333
to experience just switching between Anthropic or
00:07:19.333 --> 00:07:22.409
Claude or Cursor or Windsurf and seeing how these
00:07:22.409 --> 00:07:25.535
different, products and tools evolve. I'm gonna
00:07:25.535 --> 00:07:27.459
follow up on the tools in a minute, but Kurt, I'm
00:07:27.459 --> 00:07:30.095
curious also about your, current experience with
00:07:30.095 --> 00:07:32.476
how things have changed over the last year and a
00:07:32.476 --> 00:07:32.914
half or so.
00:07:34.316 --> 00:07:37.362
Yeah. To Habeeba's, call-out, or
00:07:37.362 --> 00:07:40.460
one of the call-outs, it feels
00:07:40.460 --> 00:07:43.318
like it is, and I don't wanna
00:07:43.318 --> 00:07:46.282
anthropomorphize the AI too much, but it's like a
00:07:46.287 --> 00:07:49.477
colleague that you gain trust with.
00:07:49.510 --> 00:07:52.595
And once you use it over, and over, again,
00:07:52.595 --> 00:07:55.443
and at first you start out, I think everybody
00:07:55.443 --> 00:07:57.541
starts out with, the coding assistance where you
00:07:57.541 --> 00:08:00.568
have to, click yes or, with the, the
00:08:00.568 --> 00:08:03.272
original patterns that these big coding
00:08:03.272 --> 00:08:06.040
assistants, the coding assistant
00:08:06.040 --> 00:08:09.168
platforms came out with, it's allow or I agree
00:08:09.168 --> 00:08:12.331
or yes. And a- as you gain trust, you
00:08:12.331 --> 00:08:15.190
allow all, and as you, gain more
00:08:15.190 --> 00:08:17.866
trust, it's like you give it bigger and heavier
00:08:17.866 --> 00:08:20.171
lifts that it needs to do. And there are a couple
00:08:20.171 --> 00:08:22.101
of different things that, of course, the models
00:08:22.101 --> 00:08:24.091
have made significant progress over the last 18
00:08:24.091 --> 00:08:26.980
months. The harnesses are… It's wild how
00:08:26.980 --> 00:08:29.751
much more complex and interesting those have
00:08:29.751 --> 00:08:32.427
become. There are, there is a big
00:08:32.427 --> 00:08:35.155
difference now, I think, between good
00:08:35.159 --> 00:08:38.265
harnesses and great harnesses. and
00:08:38.265 --> 00:08:40.376
so just seeing how much those have progressed in
00:08:40.376 --> 00:08:43.551
the last 18 months is wild. but I think it just
00:08:43.551 --> 00:08:46.702
boils down to, trust. I've used these things now
00:08:46.702 --> 00:08:49.421
like crazy, pushed them in a number of different
00:08:49.421 --> 00:08:51.510
ways, and I have a strong
00:08:51.510 --> 00:08:54.000
understanding of, what they'll
00:08:54.000 --> 00:08:56.604
actually come back with now,
00:08:56.604 --> 00:08:59.402
after using these for, 18
00:08:59.402 --> 00:09:02.738
months, 10 months, six months.
00:09:04.188 --> 00:09:05.915
it's been pretty amazing to see how much progress
00:09:05.915 --> 00:09:06.399
they've made.
00:09:07.580 --> 00:09:10.344
Think of your own control plane, in the
00:09:10.344 --> 00:09:12.978
absence of other governance boundaries or
00:09:12.978 --> 00:09:15.334
guardrails, because that's a whole conversation,
00:09:15.334 --> 00:09:17.713
right? How do we ensure the
00:09:17.713 --> 00:09:20.723
risk-managed good use of AI in
00:09:20.723 --> 00:09:23.908
the enterprise? But when you think about
00:09:23.908 --> 00:09:27.087
your own, let's say, personal cohorts that you're
00:09:27.087 --> 00:09:30.074
creating, do you also think about a control plane
00:09:30.074 --> 00:09:32.894
kind of locally for yourselves, or do you rely on
00:09:32.894 --> 00:09:35.825
the enterprise to provide that? I think
00:09:35.825 --> 00:09:38.870
it's so s… I don't wanna say easy. Nothing is
00:09:38.870 --> 00:09:40.837
easy. Everything is harder than it seems, and
00:09:40.837 --> 00:09:42.646
maybe that's a future question or a future
00:09:42.646 --> 00:09:45.838
conversation that we have. But, you can go
00:09:45.838 --> 00:09:48.648
so fast now in standing up
00:09:48.648 --> 00:09:51.663
anything custom. And so for, the
00:09:51.663 --> 00:09:54.629
people with experience in the space or,
00:09:54.629 --> 00:09:57.764
maybe for people that are, they just feel more
00:09:57.764 --> 00:10:00.774
comfortable or confident with software in general,
00:10:00.774 --> 00:10:03.813
you can really stand up, bespoke solutions and
00:10:03.813 --> 00:10:06.797
whether it's our own tools or tooling that
00:10:06.797 --> 00:10:09.840
we w- we use every day or, So
00:10:09.840 --> 00:10:12.375
I think I'm less reliant
00:10:12.375 --> 00:10:15.547
maybe, on the enterprise to come out with
00:10:15.547 --> 00:10:18.475
something that is geared for me, and I am much
00:10:18.478 --> 00:10:20.841
more empowered as an individual to just, go out
00:10:20.841 --> 00:10:23.656
and build the thing that I need, build, craft the
00:10:23.656 --> 00:10:26.635
tool or craft the skill that I need, for a
00:10:26.635 --> 00:10:29.693
particular, situation. But, yeah, I think this
00:10:29.693 --> 00:10:32.705
goes back to Habeeba and the shout-out to the
00:10:32.705 --> 00:10:34.848
open source community.
00:10:35.740 --> 00:10:38.492
like you can learn a lot if you go out and look up
00:10:38.492 --> 00:10:41.188
things online. And, yeah, we're very much
00:10:41.188 --> 00:10:44.342
empowered to unblock ourselves and create cool
00:10:44.342 --> 00:10:47.342
solutions on our own, I think. Yeah. I think
00:10:47.342 --> 00:10:49.827
that's great. when I think about it, when you
00:10:49.827 --> 00:10:52.497
think about the enterprise, we're in a spot right
00:10:52.497 --> 00:10:55.093
now where we have to give leeway to our
00:10:55.093 --> 00:10:58.220
engineering teams to discover, build, and
00:10:58.220 --> 00:11:01.365
implement, tool sets. But at some point
00:11:01.365 --> 00:11:03.701
there needs to be a rationalization of that,
00:11:03.701 --> 00:11:06.707
right? And I think especially from a risk
00:11:06.707 --> 00:11:09.847
standpoint and an efficiency standpoint, if
00:11:09.847 --> 00:11:12.836
Habeeba builds a great PR agent, let's all
00:11:12.836 --> 00:11:15.874
use Habeeba's PR agent, right? I think that's
00:11:15.874 --> 00:11:19.031
the simplest thing to talk about. But we've always
00:11:19.031 --> 00:11:20.807
said this when we had service-oriented
00:11:20.807 --> 00:11:23.714
architectures, is, was that there was an
00:11:23.714 --> 00:11:26.869
idea that we'd have a catalog of,
00:11:26.869 --> 00:11:28.632
endpoints or APIs.
00:11:29.889 --> 00:11:32.781
and I'm not sure we achieved that
00:11:32.781 --> 00:11:35.918
end the way we wanted to. I used
00:11:35.918 --> 00:11:38.713
to always think in the enterprise, I would build
00:11:38.713 --> 00:11:41.262
an API to something and I would think, "Hasn't
00:11:41.262 --> 00:11:43.669
three other people built this connector here?"
00:11:44.788 --> 00:11:47.737
What I wonder really is, with the dramatic
00:11:47.737 --> 00:11:50.370
delivery of all of these new tools,
00:11:50.370 --> 00:11:53.540
especially open source, how would you
00:11:53.540 --> 00:11:56.510
guide young engineers in picking
00:11:56.510 --> 00:11:58.790
and choosing tool sets?
00:12:00.024 --> 00:12:03.076
Without the concern of, this might, this one might
00:12:03.076 --> 00:12:05.461
be obsolete, because that just might happen.
00:12:05.500 --> 00:12:07.655
That's not a bad thing. It's just… But
so how
00:12:07.655 --> 00:12:10.194
would you encourage a young engineer to think
00:12:10.194 --> 00:12:12.958
about what they choose to include in their
00:12:12.958 --> 00:12:15.849
stack? Habeeba? Yeah, I think this is a
00:12:15.849 --> 00:12:18.521
great question. And honestly, just
00:12:18.521 --> 00:12:21.599
because they're engineers, and as an
00:12:21.599 --> 00:12:24.796
engineer myself, what works best is trying it. you
00:12:24.796 --> 00:12:27.395
are… Someone can try to sell you something and
00:12:27.395 --> 00:12:29.645
tell you it's the best tool ever, and as an
00:12:29.645 --> 00:12:31.746
engineer you're not going to believe them until
00:12:31.746 --> 00:12:34.646
you try it for yourself and you actually start
00:12:34.646 --> 00:12:37.547
using it. even with all these tools coming out,
00:12:37.547 --> 00:12:40.574
the best way for me to evaluate them is by using
00:12:40.574 --> 00:12:43.278
them to code, using them to do my job.
00:12:43.603 --> 00:12:46.151
That's how you learn what these tools are good at,
00:12:46.151 --> 00:12:48.438
what they're not good at, what do I like about
00:12:48.438 --> 00:12:50.668
them, what do I not like about them, and that's
00:12:50.668 --> 00:12:53.616
really worked well for me, even in the space of,
00:12:53.616 --> 00:12:56.295
using agents that the open source community has
00:12:56.295 --> 00:12:58.231
come out with, or using skills that they've come
00:12:58.231 --> 00:13:00.614
out with. 'Cause again, people will recommend
00:13:00.614 --> 00:13:03.201
certain things, but I think what works best is to
00:13:03.201 --> 00:13:06.275
just trying them for, yourself and seeing if
00:13:06.275 --> 00:13:08.927
you like them or not, and then based on that,
00:13:08.927 --> 00:13:10.933
making your judgment and just gaining more
00:13:10.933 --> 00:13:13.256
exposure. So if you like something, don't just
00:13:13.256 --> 00:13:15.863
stick to it and try other things. The more things
00:13:15.863 --> 00:13:18.670
that you try, the better of an idea that you have
00:13:18.670 --> 00:13:20.907
on, what you like and what you don't like and what
00:13:20.907 --> 00:13:23.028
works best for you. Totally agree.
00:13:24.114 --> 00:13:27.211
Build and build anything that you, can
00:13:27.211 --> 00:13:30.352
think of or imagine. and you just
00:13:30.352 --> 00:13:33.517
learn a lot by trying. And there… I'm a big
00:13:33.517 --> 00:13:36.458
proponent of, the let the rocket ship
00:13:36.458 --> 00:13:39.038
blow up on the launchpad and just keep, building
00:13:39.038 --> 00:13:41.295
and deploying- Yeah … and getting things out there
00:13:41.295 --> 00:13:44.397
into the world. And, I think we learn a ton, by
00:13:44.397 --> 00:13:47.529
doing that. And maybe to piggyback off on, what,
00:13:47.529 --> 00:13:50.069
another thing that Habeeba had mentioned,
00:13:50.293 --> 00:13:53.305
it's like the, the- There is
00:13:53.305 --> 00:13:55.902
still a lot of the human
00:13:55.902 -->
00:13:59.005
factor, I think, in all of this,
00:13:59.005 --> 00:14:01.194
where we might get advertisements or you're
00:14:01.194 --> 00:14:03.455
bombarded with ads for, the new latest and
00:14:03.455 --> 00:14:06.489
greatest tool out there, but it's still very human
00:14:06.489 --> 00:14:09.301
with what I select. I go to a lot of
00:14:09.301 --> 00:14:12.346
talks, in, in San Francisco. The,
00:14:12.346 --> 00:14:15.007
the AI community is wild, and they're like, you
00:14:15.007 --> 00:14:17.551
could go to 10-- you could pick between 10 or 15
00:14:17.551 --> 00:14:19.818
different talks a day out here it seems like.
00:14:20.602 --> 00:14:23.619
but, I am much less likely to pick
00:14:23.619 --> 00:14:26.546
up a tool if I see it advertised to me on
00:14:26.546 --> 00:14:29.640
social media. But if some random person that
00:14:29.640 --> 00:14:32.397
I just meet one time in person at one of these
00:14:32.397 --> 00:14:35.314
events, or even internally, we have a bunch of
00:14:35.314 --> 00:14:38.218
great conversations at AHEAD that are, like,
00:14:38.218 --> 00:14:41.306
people just talking about some new thing that
00:14:41.306 --> 00:14:43.972
they've tried, and it's cool, and this is how I'm
00:14:43.972 --> 00:14:46.891
using it. And those water cooler conversations,
00:14:46.891 --> 00:14:49.690
they have a huge impact on me. And I'm like, "Oh,
00:14:49.690 --> 00:14:51.714
if this person's excited about it," and I get a
00:14:51.714 --> 00:14:54.760
chance to read that, like I'm gonna go out and try
00:14:54.760 --> 00:14:57.618
this new tool. And so it's still like a big
00:14:57.618 --> 00:15:00.748
human process, I think, at the end of the day.
00:15:01.513 --> 00:15:03.694
with all the big advancements in AI, it's still
00:15:03.694 --> 00:15:05.622
very human, I think. The, the work component's
00:15:05.622 --> 00:15:06.291
still very human.
00:15:07.485 --> 00:15:09.330
I can't resist this question just 'cause we're
00:15:09.330 --> 00:15:12.438
together. Is there a tool, a specific tool or a
00:15:12.438 --> 00:15:15.530
category of tool that you've been really surprised
00:15:15.530 --> 00:15:18.147
about in the last however amount of time where you
00:15:18.147 --> 00:15:20.648
were like, "This is the coolest thing. It's really
00:15:20.648 --> 00:15:22.692
made a big difference in how I think about
00:15:22.692 --> 00:15:26.874
delivery"? I'd say just shout out Cursor.
00:15:28.833 --> 00:15:31.789
that's something… I've been using Cursor now for
00:15:31.789 --> 00:15:34.872
probably, over a year. and it's, of course,
00:15:34.872 --> 00:15:37.851
the adoption, like how much you use it and where
00:15:37.851 --> 00:15:40.907
you use it, s- being strategic and
00:15:40.907 --> 00:15:43.721
making sure to, do, the sanity
00:15:43.721 --> 00:15:46.304
checks, and the tool has gotten significantly
00:15:46.304 --> 00:15:48.891
better. But it's just been unbelievable to see how
00:15:48.891 --> 00:15:51.418
much progress has been made with the coding
00:15:51.418 --> 00:15:54.111
assistance. It's really unbelievable. And there
00:15:54.111 --> 00:15:57.115
are some smaller tools maybe that, fit
00:15:57.115 --> 00:16:00.188
into some of my processes that are really
00:16:00.188 --> 00:16:03.232
powerful, but just it blows my mind to see
00:16:03.232 --> 00:16:05.475
what these coding assistants can do now.
00:16:06.633 --> 00:16:09.045
it's just unbelievable. That has been the biggest
00:16:09.045 --> 00:16:11.899
impact. And there was this talk that I saw
00:16:11.899 --> 00:16:15.077
recently of, the, the person who invented
00:16:15.077 --> 00:16:18.039
Ruby on Rails, talking about, it's
00:16:18.039 --> 00:16:20.071
the end of engineering.
00:16:20.842 --> 00:16:23.854
and, like I am retired now as
00:16:23.854 --> 00:16:26.666
an engineer, and now I am a
00:16:26.666 --> 00:16:29.796
creator of things. And, he changed
00:16:29.796 --> 00:16:32.820
his perspective of, if we're using these
00:16:32.820 --> 00:16:35.978
tools, you can just build products, in a
00:16:35.978 --> 00:16:38.472
day or… I don't wanna set any crazy expectations
00:16:38.472 --> 00:16:41.015
for anyone. I love that it's a Ruby guy. It's so
00:16:41.015 --> 00:16:42.920
hard. Everything's- Yeah. It's just funny to me
00:16:42.920 --> 00:16:45.585
that it's a Ruby story because that's, they're,
00:16:45.585 --> 00:16:47.976
they have always been ambitious that way, right?
00:16:47.976 --> 00:16:50.718
It's gonna just be incredibly fast and
00:16:50.718 --> 00:16:53.714
that's great. I think that's exactly the
00:16:53.714 --> 00:16:56.391
way, we're all thinking. Habib, any tools that
00:16:56.391 --> 00:16:59.379
come to mind that you would wanna talk about? I,
00:16:59.379 --> 00:17:02.138
Kurt mentioned my favorite one, and it feels like
00:17:02.138 --> 00:17:05.040
now we're pushing Cursor too hard, but
00:17:05.040 --> 00:17:07.386
it really is, so good
00:17:07.386 --> 00:17:10.266
compared to its peers that
00:17:10.266 --> 00:17:12.828
it's hard to think… And
00:17:12.828 --> 00:17:15.943
again, I think a year ago, if
00:17:15.943 --> 00:17:18.746
you used, Cursor and maybe
00:17:18.814 --> 00:17:21.785
Cognition products, you were like, they're both
00:17:21.785 --> 00:17:24.707
the same, or even Claude Code. But now with
00:17:24.707 --> 00:17:27.491
Cursor, there is such a significant difference
00:17:27.491 --> 00:17:29.970
between it, and even just using the same exact
00:17:29.970 --> 00:17:32.906
model. If you use it in Cursor, it… you
00:17:32.906 --> 00:17:35.928
save so many more tokens, the processing is
00:17:35.928 --> 00:17:39.109
significantly faster. The way that they manage
00:17:39.109 --> 00:17:41.931
context is really good as well. And so
00:17:41.931 --> 00:17:45.049
that's just been, like, a great tool to be
00:17:45.049 --> 00:17:48.060
able to use because the productivity
00:17:48.060 --> 00:17:50.531
gains from Cursor are, like,
00:17:50.531 -->
00:17:53.724
amazing. And, again, compared to the other tools,
00:17:53.724 --> 00:17:56.400
I think they're really outdoing themselves. I
00:17:56.400 --> 00:17:58.676
think we get an update now, every three days, they
00:17:58.676 --> 00:18:01.492
have an update to the application. You
00:18:01.492 --> 00:18:04.485
brought up a really interesting point about token
00:18:04.485 --> 00:18:07.386
visibility. How much of your day-to-day are you
00:18:07.386 --> 00:18:10.019
thinking about your token use? And then similarly,
00:18:10.019 --> 00:18:11.929
when you're building something, when you're
00:18:11.929 --> 00:18:14.509
architecting something, are you also thinking
00:18:14.509 --> 00:18:17.685
about the token cost of running that platform? Are
00:18:17.685 --> 00:18:20.609
those starting to be, like, important daily
00:18:20.609 --> 00:18:23.618
conversations? The first point, in
00:18:23.618 --> 00:18:26.602
personal usage, I personally am aware
00:18:26.602 --> 00:18:28.946
of my token usage just because we all have the
00:18:28.946 --> 00:18:30.882
limits that we're bound to and how many tokens
00:18:30.882 --> 00:18:33.475
we're consuming. And so you always want to be
00:18:33.475 --> 00:18:36.137
cognizant of that to a certain extent of how many
00:18:36.137 --> 00:18:38.655
tokens are you consuming. If you're consuming too
00:18:38.655 --> 00:18:40.637
many, is there a more efficient approach that you
00:18:40.637 --> 00:18:43.610
should take with the models that you're selecting
00:18:43.610 --> 00:18:46.633
or the type of task that you're asking AI to
00:18:46.633 --> 00:18:49.669
execute on? And so I'm definitely cognizant
00:18:49.669 --> 00:18:52.458
of that in my personal use, again, because someone
00:18:52.458 --> 00:18:54.776
is footing the cost at the end of the day, either
00:18:54.776 --> 00:18:57.396
I am or AHEAD is. With clients, what I've found is
00:18:57.396 --> 00:18:59.714
that if they're really trying to get something out
00:18:59.714 --> 00:19:02.045
there and get people to like using it and want to
00:19:02.045 --> 00:19:04.715
keep using it and get adoption out, they are less
00:19:04.715 --> 00:19:07.287
concerned about token usage just because they want
00:19:07.287 --> 00:19:09.731
people to use it. And then once people are
00:19:09.731 --> 00:19:12.626
actually using it in their day-to-day and you have
00:19:12.639 --> 00:19:15.377
people that like using it, that's when they really
00:19:15.377 --> 00:19:17.652
start focusing on tokens. So if we're, like,
00:19:17.652 --> 00:19:19.777
building a POC or just building a pilot, they
00:19:19.777 --> 00:19:21.693
usually don't care that much about tokens. You
00:19:21.693 --> 00:19:23.467
just want to get a really good product out there.
00:19:23.467 --> 00:19:26.202
And then when you know what that looks like,
00:19:26.202 --> 00:19:29.357
that's when you start focusing on tokens, how can
00:19:29.357 --> 00:19:32.135
I minimize my token usage, and what's the best way
00:19:32.135 --> 00:19:35.308
to architect this? Might actually give the
00:19:35.308 --> 00:19:36.298
inverse answer.
00:19:37.475 --> 00:19:40.251
I am-- I try not to think about the
00:19:40.251 --> 00:19:43.105
token usage, internally, and maybe I
00:19:43.105 --> 00:19:46.288
should, and now I feel, guilty or awkward saying
00:19:46.288 --> 00:19:49.068
this. It's being recorded. Somebody's gonna reach
00:19:49.068 --> 00:19:51.516
out to me about my token budget now internally.
00:19:52.513 --> 00:19:55.035
I'm, I am not a proponent of just token maxing for
00:19:55.035 --> 00:19:58.206
the sake of, like- Yeah you just use tokens just
00:19:58.206 --> 00:20:01.363
to use tokens. and it seems like that's a-- I
00:20:01.363 --> 00:20:04.377
can see how that is a flawed metric. just
00:20:04.377 --> 00:20:06.575
encourage peop- or encouraging people just to
00:20:06.575 --> 00:20:09.305
build anything. We definitely wanna build things
00:20:09.305 --> 00:20:12.335
that are valuable or, important. But, I think
00:20:12.335 --> 00:20:15.488
we're just still learning a lot about
00:20:15.488 --> 00:20:18.467
the new development process, and so
00:20:18.467 --> 00:20:21.332
putting caps on, token usage right
00:20:21.332 --> 00:20:24.242
now, it just seems like we shouldn't be
00:20:24.242 --> 00:20:26.749
doing that. We should just be allowing people to
00:20:26.749 --> 00:20:29.251
play with it, monitoring it, of course, to make
00:20:29.251 --> 00:20:32.451
sure that nothing goes crazy or, there aren't some
00:20:32.451 --> 00:20:35.162
crazy outliers that could, sink our group or
00:20:35.162 --> 00:20:38.114
something with our overall budget. But, we still
00:20:38.114 --> 00:20:40.618
need to learn a lot more about, what the
00:20:40.618 --> 00:20:43.643
development process looks like now and,
00:20:43.643 --> 00:20:46.347
so letting people use it, and
00:20:46.347 --> 00:20:49.218
adapting, the rest of how a group comes
00:20:49.218 --> 00:20:50.763
together to actually stitch together the
00:20:50.763 --> 00:20:53.838
components that they're building or, yeah, product
00:20:53.838 --> 00:20:56.857
development I think is shifting enough. We
00:20:56.857 --> 00:20:59.500
just kinda should be unblocking people to go out
00:20:59.500 --> 00:21:02.172
and try things and explore, with some
00:21:02.172 --> 00:21:05.047
guardrails, in place. I think what you're both
00:21:05.047 --> 00:21:05.779
saying is right on.
00:21:08.107 --> 00:21:11.089
I think- It's good to be aware of your
00:21:11.089 --> 00:21:14.289
token usage, just because I think it's gonna be
00:21:14.289 --> 00:21:16.861
an ongoing conversation. I also think you need to
00:21:16.861 --> 00:21:18.757
identify people in your organization who you're
00:21:18.757 --> 00:21:21.029
saying, "Listen, I'm not thinking about your token
00:21:21.029 --> 00:21:23.472
usage," because you've gotta really be
00:21:23.472 --> 00:21:26.781
ambitious, and I can't limit your ambition.
00:21:27.930 --> 00:21:30.507
and I al- because I've worked in large
00:21:30.507 --> 00:21:33.492
enterprises, I'm always… I would prefer us to
00:21:33.492 --> 00:21:36.289
give the most powerful tools to our most trusted
00:21:36.289 --> 00:21:38.823
technology partners. And so I don't
00:21:38.823 --> 00:21:41.942
necessarily wanna do that for everybody in, on the
00:21:41.947 --> 00:21:45.141
business side, but also for them, they need to
00:21:45.141 --> 00:21:48.144
have some people who are similarly doing
00:21:48.144 --> 00:21:50.914
something like that, who have the
00:21:50.914 --> 00:21:54.110
ability to just explore. ROI is
00:21:54.110 --> 00:21:56.849
important. In the age of real
00:21:56.849 --> 00:21:59.837
innovation, you have to balance ROI a
00:21:59.837 --> 00:22:02.918
little bit. You have to have an i- an appetite
00:22:02.918 --> 00:22:05.794
to recognize that some of your spend may not
00:22:05.794 --> 00:22:09.375
have an ROI because you're in an innovation cycle.
00:22:10.600 --> 00:22:13.638
when we think about the, the title of this
00:22:13.638 --> 00:22:16.513
session, which is From Demo to Deploy,
00:22:16.513 --> 00:22:19.503
are you finding that we're able to
00:22:19.503 --> 00:22:22.664
get a better view from that start to
00:22:22.664 --> 00:22:25.285
finish because of the tools we're using?
00:22:25.287 --> 00:22:27.890
Because, maybe I'm a product owner or
00:22:27.890 --> 00:22:30.766
business user and I'm using the tools to
00:22:30.766 --> 00:22:33.849
build UIs, and so now I'm demoing
00:22:33.849 --> 00:22:37.023
to my teams, non-working software,
00:22:37.023 --> 00:22:39.874
but UIs that demonstrate what it would do.
00:22:40.739 --> 00:22:43.240
and then those become stories for the
00:22:43.240 --> 00:22:46.158
engineers, and those engineers build those
00:22:46.158 --> 00:22:49.129
features, and then we get to deploy.
00:22:50.042 --> 00:22:52.915
Are you starting to see the quality of what
00:22:52.915 --> 00:22:55.871
we're doing improve over that period of
00:22:55.871 --> 00:22:58.783
time? And the converse, are we seeing
00:22:58.783 --> 00:23:01.890
anything that is promoting bad habits, or the
00:23:01.890 --> 00:23:05.225
speed creates a disproportionate negative effect?
00:23:06.523 --> 00:23:09.358
I think it's always been show, don't
00:23:09.358 --> 00:23:12.429
tell, in terms of just, if somebody can
00:23:12.429 --> 00:23:15.244
see a UI, they can understand a bit
00:23:15.244 --> 00:23:17.981
more about what's going on and how
00:23:17.981 --> 00:23:20.895
quickly we can get to not even
00:23:20.895 --> 00:23:23.809
just, a demo, but to a working
00:23:23.809 --> 00:23:26.818
prototype. It's r- just getting there and
00:23:26.818 --> 00:23:29.639
getting that in front of someone, they have a much
00:23:29.644 --> 00:23:31.972
better understanding of what's going on. And so
00:23:31.972 --> 00:23:34.599
just starting almost, the testing
00:23:34.599 --> 00:23:37.347
cadence is-- it's
so much earlier in a
00:23:37.347 --> 00:23:39.931
project, and you almost wanna start it on, day
00:23:39.931 --> 00:23:42.819
two. let's actually just start testing something.
00:23:43.124 --> 00:23:46.301
And it's been even really interesting to see how
00:23:46.301 --> 00:23:48.948
much faster… building demos and building, a
00:23:48.948 --> 00:23:51.332
working demo six months ago even or at the
00:23:51.332 --> 00:23:54.187
beginning of the year, it was much more powerful.
00:23:55.542 --> 00:23:58.714
and, just, getting a, a simple toy demo out
00:23:58.714 --> 00:24:01.678
there was much more powerful, and now
00:24:01.678 --> 00:24:04.599
the bar is much higher. Even, eight, 10
00:24:04.599 --> 00:24:07.702
months later, people are like, "Oh, I
00:24:07.702 --> 00:24:10.769
don't even wanna see a just, toy demo
00:24:10.769 --> 00:24:13.313
anymore. I wanna talk about, the
00:24:13.313 --> 00:24:16.069
production process. what are all of the
00:24:16.069 --> 00:24:18.411
nitty-gritty things that are gonna come back and
00:24:18.411 --> 00:24:21.216
bite me?" I think people's levels of,
00:24:21.216 --> 00:24:24.094
just, understanding of what quality is
00:24:24.123 --> 00:24:26.966
and, the bar is much higher, and it's
00:24:26.966 --> 00:24:29.908
happened very quickly from, "Wow, a, a
00:24:29.908 --> 00:24:32.805
quick toy demo is really powerful," maybe 10
00:24:32.805 --> 00:24:35.803
months ago to now it's like, "Nope, we wanna
00:24:35.803 --> 00:24:38.654
see even more. We wanna, see more out of the
00:24:38.654 --> 00:24:41.479
tool," or, "We wanna hear more about what you all
00:24:41.479 --> 00:24:43.653
are doing." but it's interesting seeing that
00:24:43.653 --> 00:24:46.418
process too. Yeah, and to echo
00:24:46.418 --> 00:24:49.456
Kurt's point, I think even Google implemented a
00:24:49.456 --> 00:24:52.060
policy recently where if you are a
00:24:52.060 --> 00:24:54.838
product owner, you need to yourself
00:24:54.838 --> 00:24:57.254
develop what you envision your product to look
00:24:57.254 --> 00:25:00.289
like before you go out and ask engineers to, build
00:25:00.289 --> 00:25:03.110
a prototype for you. Just, giving the
00:25:03.110 --> 00:25:06.213
ownership of that now to people who have a less
00:25:06.213 --> 00:25:09.229
technical role and traditionally would have had to
00:25:09.229 --> 00:25:12.024
rely on engineering to bring their ideas to life.
00:25:12.024 --> 00:25:14.696
It's a lot easier now for people to do that.
00:25:14.696 --> 00:25:17.826
Standing up a UI now is really easy with all
00:25:17.826 --> 00:25:20.691
of these tools, showing people what your
00:25:20.691 --> 00:25:23.787
vision is, what you're thinking of is a lot
00:25:23.787 --> 00:25:26.688
easier. I think then at that point is
00:25:26.688 --> 00:25:29.544
when a lot of times that visionary or the
00:25:29.544 --> 00:25:31.908
product owner hands it off to engineering teams,
00:25:31.908 --> 00:25:34.971
and they work on, fully
productionalizing
00:25:34.971 --> 00:25:38.153
that demo. But just to stand up a demo is really
00:25:38.153 --> 00:25:40.758
fascinating to see the progress of we went from
00:25:40.758 --> 00:25:43.388
"Oh, I need a month to be able to create something
00:25:43.388 --> 00:25:46.085
that I can show to other people," to if you just
00:25:46.085 --> 00:25:48.898
give me 24 hours, you can turn something around
00:25:48.898 --> 00:25:51.808
that looks pretty good. And again, that speed has
00:25:51.808 --> 00:25:54.289
also shown itself in engineering teams who can
00:25:54.289 --> 00:25:56.530
then take that and run with it and create
00:25:56.530 --> 00:25:59.393
something that is more productionalized and ready
00:25:59.393 --> 00:26:02.381
to be used by other people. But again, it's
00:26:02.381 --> 00:26:05.438
been a fascinating change to see. And again, the
00:26:05.438 --> 00:26:08.080
empowerment that it's given to a lot of people to
00:26:08.080 --> 00:26:11.226
just honestly, bring their ideas to life.
00:26:11.226 --> 00:26:13.125
That's what I liked the most about computer
00:26:13.125 --> 00:26:15.335
science when I first started in it, and now it
00:26:15.335 --> 00:26:18.054
feels like everyone has access to I can just write
00:26:18.054 --> 00:26:20.923
something and then I get something out. Yeah, I
00:26:20.923 --> 00:26:23.926
started building software, over 20 years ago, and
00:26:23.926 --> 00:26:26.336
we really focused on user center--
00:26:26.336 --> 00:26:29.269
user-centered design, right? And I
00:26:29.269 --> 00:26:32.125
always wanted to have UI and UX
00:26:32.125 --> 00:26:34.624
people on my team to really
00:26:34.624 --> 00:26:37.516
emphasize the importance of a good experience, and
00:26:37.516 --> 00:26:39.712
that the good experience would translate to
00:26:39.712 --> 00:26:42.157
revenue efficiency, risk management.
00:26:42.941 --> 00:26:45.155
But that was always hard to get budget for.
00:26:46.177 --> 00:26:48.232
I've worked in environments that don't-- didn't
00:26:48.232 --> 00:26:50.336
value that, and their software looks like
00:26:50.336 --> 00:26:53.158
complicated forms that you need to have
00:26:53.158 --> 00:26:55.680
secret handshakes to understand how to get
00:26:55.680 --> 00:26:58.606
through, right? And now we're
00:26:58.606 --> 00:27:01.419
making that more of a, a
00:27:01.419 --> 00:27:04.600
democratic ability. We can all
00:27:04.600 --> 00:27:07.551
build UI and UX now. And if you're a
00:27:07.551 --> 00:27:10.196
product owner, you have to be thinking that way.
00:27:11.551 --> 00:27:14.669
I still think, it requires Smart
00:27:14.669 --> 00:27:17.787
database strategy, because a lot
00:27:17.787 --> 00:27:20.522
of times connecting the UI to the
00:27:20.522 --> 00:27:23.419
database functionality still needs
00:27:23.419 --> 00:27:26.435
critical architecture. But to start, I can
00:27:26.435 --> 00:27:29.613
get buy-in and excitement so much faster for
00:27:29.613 --> 00:27:32.218
an opportunity, and then when we talk about the
00:27:32.218 --> 00:27:33.440
cost of delivering it,
00:27:34.801 --> 00:27:37.180
the audience that's approving that budget has a
00:27:37.180 --> 00:27:39.559
better sense of what the outcome looks like, and
00:27:39.559 --> 00:27:41.802
so they feel more confident giving that budget.
00:27:41.872 --> 00:27:44.636
And so I really like the… where we are
00:27:44.636 --> 00:27:47.573
because it has become a reminder to people that we
00:27:47.573 --> 00:27:50.556
can show how this will work, and people will
00:27:50.556 --> 00:27:53.655
intuitively understand what the value is as it
00:27:53.655 --> 00:27:56.789
moves forward. What do you think that
00:27:56.789 --> 00:27:59.972
people are underestimating right now? we, I
00:27:59.972 --> 00:28:02.934
think, all agree about the excitement,
00:28:02.934 --> 00:28:06.015
the incredible speed that things are occurring
00:28:06.015 --> 00:28:09.125
at. But what are people underestimating? What
00:28:09.125 --> 00:28:10.912
are the challenges that are coming up?
00:28:12.543 --> 00:28:15.401
I'll take a stab at this first. I think that
00:28:15.401 --> 00:28:18.095
the issues that aren't being
00:28:18.095 --> 00:28:21.162
thought about now, and again, it
00:28:21.162 --> 00:28:23.467
makes sense, people just wanna try, they wanna
00:28:23.467 --> 00:28:26.361
experiment, and they wanna be creative, is really,
00:28:26.361 --> 00:28:29.517
the governance and operations around all of
00:28:29.517 --> 00:28:32.159
these agents and all of these things that are
00:28:32.159 --> 00:28:34.950
being deployed. Y- if you have some of our clients
00:28:34.950 --> 00:28:37.709
that are more advanced, and they have, hundreds of
00:28:37.709 --> 00:28:40.586
agents that are out there running in the cloud,
00:28:41.007 --> 00:28:43.633
operationalizing something like that and the
00:28:43.633 --> 00:28:46.475
governance around it is a nightmare. And a lot of
00:28:46.475 --> 00:28:49.274
companies do not think about that before. They're
00:28:49.274 --> 00:28:51.415
like, "Okay, let's have agents. We want agents
00:28:51.415 --> 00:28:54.404
everywhere," and people are using agents. Because,
00:28:54.404 --> 00:28:57.169
again, you get the security incidents that we hear
00:28:57.169 --> 00:29:00.268
about. And then with the models changing so
00:29:00.268 --> 00:29:02.898
often with agents and harnesses, if
00:29:03.253 --> 00:29:06.159
Anthropic retires a model, what does your agent
00:29:06.159 --> 00:29:09.093
look like now that you've trained it to use that
00:29:09.093 --> 00:29:11.770
particular model? And how do you keep
00:29:11.770 --> 00:29:14.473
evaluating your agent when all of these
00:29:14.473 --> 00:29:17.025
different models come out, you've tailored it to
00:29:17.025 --> 00:29:19.694
these different models? So the governance around
00:29:19.694 --> 00:29:21.908
that, like, how often are people using these
00:29:21.908 --> 00:29:24.386
agents? hey, maybe this was useful six months ago
00:29:24.386 --> 00:29:26.667
when everyone was using it, but now no one's using
00:29:26.667 --> 00:29:28.784
it. And if you're not tracking, if you're not
00:29:28.784 --> 00:29:30.972
tracking all of those requests, if you're not
00:29:30.972 --> 00:29:33.630
tracking who's calling what agent, you end up with
00:29:33.630 --> 00:29:36.367
really agent sprawl, and then everyone is, making
00:29:36.367 --> 00:29:39.265
their own agent for every small thing. You're not
00:29:39.265 --> 00:29:42.460
reusing agents that might already exist. And
00:29:42.460 --> 00:29:45.413
then the ownership of that is unclear, and I think
00:29:45.413 --> 00:29:48.180
that happens in a lot of organizations where
00:29:48.180 --> 00:29:51.180
individuals are deploying agents, but then who
00:29:51.181 --> 00:29:53.615
actually is responsible for this? If there is a
00:29:53.615 --> 00:29:56.612
security event, who is responsible for ensuring
00:29:56.612 --> 00:29:59.407
that doesn't happen and governing that
00:29:59.407 --> 00:30:02.283
entire platform? I think clients are grappling
00:30:02.283 --> 00:30:03.394
with that right now.
00:30:05.626 --> 00:30:08.790
Yes. With everyone being a developer, how do
00:30:08.790 --> 00:30:11.735
you actually… It didn't… Because there's a
00:30:11.735 --> 00:30:14.524
lot of training that you get as a
00:30:14.524 --> 00:30:17.483
software engineer, for, standing up
00:30:17.483 --> 00:30:20.574
systems that are robust, making sure
00:30:20.574 --> 00:30:23.164
to monitor production systems,
00:30:23.164 --> 00:30:25.746
having, time to
00:30:25.746 --> 00:30:28.390
clean stuff up or, fix
00:30:28.390 --> 00:30:30.688
bugs or, having, an
00:30:30.688 --> 00:30:33.561
on-call that's actually,
00:30:33.862 --> 00:30:36.395
available to fix production systems.
00:30:36.786 --> 00:30:39.795
These are things that, when everyone can
00:30:39.795 --> 00:30:42.685
stand something up quickly and anyone
00:30:42.685 --> 00:30:45.795
can potentially push something into,
00:30:45.795 --> 00:30:48.835
a production environment, this stuff
00:30:48.835 --> 00:30:51.912
might be lost in the, shuffle. And, this is
00:30:51.912 --> 00:30:54.741
one of the things, it's still very hard to keep
00:30:54.741 --> 00:30:57.561
things up and running. I do think the
00:30:57.561 --> 00:31:00.126
tail of these projects has a
00:31:00.126 --> 00:31:02.957
monitoring aspect that we don't
00:31:02.967 --> 00:31:05.763
really do right now, where I have to
00:31:05.763 --> 00:31:08.710
continuously monitor the output of the
00:31:08.712 --> 00:31:11.701
model to ensure that it, continues to
00:31:11.701 --> 00:31:14.612
be aligned with what my expectations
00:31:14.612 --> 00:31:17.475
are. And I just think that's an interesting
00:31:17.475 --> 00:31:20.235
cost, and it's part of the governance strategy.
00:31:21.077 --> 00:31:23.986
and I would expect that if I build it,
00:31:23.986 --> 00:31:27.044
like my team builds it, that there might
00:31:27.045 --> 00:31:29.988
be a centralized enterprise team that is
00:31:29.988 --> 00:31:32.859
responsible for evaluating the outputs.
00:31:33.117 --> 00:31:34.441
that's the way I would do it.
00:31:37.119 --> 00:31:40.063
and I think we're gonna face and uncover
00:31:40.063 --> 00:31:42.916
lots of different challenges over this,
00:31:42.916 --> 00:31:45.858
let's call it next 18 months, as people roll
00:31:45.858 --> 00:31:48.562
out agents. I also think the idea of agent
00:31:48.562 --> 00:31:51.525
ownership, is gonna have to get clearer. the
00:31:51.525 --> 00:31:54.680
courts are gonna basically say that the companies
00:31:54.680 --> 00:31:56.974
are accountable for the agents, right? That's what
00:31:56.974 --> 00:31:59.163
the courts are gonna say. But we have to
00:31:59.163 --> 00:32:01.821
meaningfully figure out what that means from an
00:32:01.821 --> 00:32:04.162
identity and access perspective inside our
00:32:04.162 --> 00:32:07.156
organizations to have non-human entities that,
00:32:07.156 --> 00:32:10.236
that we can track. Because the thing that I've
00:32:10.236 --> 00:32:12.163
been concerned about really is the proliferation
00:32:12.163 --> 00:32:15.076
of agents permitted to build other
00:32:15.076 --> 00:32:18.175
agents. That's where I start to see the,
00:32:18.175 --> 00:32:20.885
a real loss of control. And when you look at the,
00:32:20.885 --> 00:32:23.460
other incidents that are happening, a lot of them
00:32:23.460 --> 00:32:26.321
are rooted in the ul… the permission to allow
00:32:26.321 --> 00:32:29.221
agents to do whatever they want. And I think we
00:32:29.221 --> 00:32:31.989
need to be cautious about how we think about that.
00:32:32.525 --> 00:32:35.545
We're getting into some spicy territory, I
00:32:35.545 --> 00:32:38.540
think, with, identity for AI. I'm curious if we
00:32:38.540 --> 00:32:41.490
should keep going deeper on this or if
00:32:41.490 --> 00:32:44.008
this is, I'd love to know your view
00:32:46.653 --> 00:32:49.083
I think it's wild. We're living… It feels like,
00:32:49.083 --> 00:32:51.789
even just in the last couple of months, it feels
00:32:51.791 --> 00:32:54.944
more now like we are living in the
00:32:54.944 --> 00:32:58.119
future. Like the, when I was growing up and
00:32:58.119 --> 00:33:00.790
thinking about what the world could look
00:33:00.792 --> 00:33:03.741
like, This is absolutely the
00:33:03.741 --> 00:33:06.486
future, to me or to child-
00:33:06.486 --> 00:33:09.472
childhood Kurt, Yeah … and just
00:33:09.472 --> 00:33:11.774
thinking about, yeah, where we could be or what
00:33:11.774 --> 00:33:14.765
the possibilities look like. And, to think about
00:33:14.765 --> 00:33:17.703
those sci-fi movies or books, that I
00:33:17.703 --> 00:33:20.572
grew up with, and thinking about
00:33:20.572 --> 00:33:23.094
having a real honest
00:33:23.094 --> 00:33:26.217
conversation about giving
00:33:26.217 -->
00:33:29.820
AI, an AI entity, an ID card.
00:33:31.183 --> 00:33:34.260
giving it autonomy to go operate out in the world.
00:33:34.545 --> 00:33:36.891
It's really unbelievable that we can even have
00:33:36.891 --> 00:33:39.455
this conversation seriously. this is a serious
00:33:39.455 --> 00:33:42.196
conversation to have right now. So I don't know.
00:33:42.422 --> 00:33:44.985
it's just, it still blows my mind each day.
00:33:45.931 --> 00:33:48.912
it's… I don't know. It's so fun. We live in such a
00:33:48.912 --> 00:33:51.244
cool world, I think, with, like, where we're at
00:33:51.244 --> 00:33:53.928
with technology. I'm such a big tech optimist
00:33:53.928 --> 00:33:55.810
with, we'll figure it out. Me too.
00:33:57.313 --> 00:34:00.306
it's been fascinating to hear about, the
00:34:00.306 --> 00:34:03.401
strong morals that can be picked up, by an
00:34:03.401 --> 00:34:05.739
AI model that's trained on web content.
00:34:08.570 --> 00:34:10.564
there's this really fascinating thing. I feel like
00:34:10.564 --> 00:34:13.475
this is like a, an aside, and, But, if
00:34:13.475 --> 00:34:16.199
you look at all the things that get
00:34:16.199 --> 00:34:19.380
likes online, and, there are so many,
00:34:21.630 --> 00:34:24.790
animal videos where, somebody helps an
00:34:24.790 --> 00:34:27.626
animal or, animals doing very nice
00:34:27.626 --> 00:34:30.543
things, and how many likes and hearts
00:34:30.543 --> 00:34:33.606
and loves those things get, and the fact
00:34:33.606 --> 00:34:36.602
that we have a lot of humans helping humans
00:34:36.602 --> 00:34:39.165
videos, and that gets, a lot of, attention
00:34:39.184 --> 00:34:41.982
online, and that those are the…
00:34:42.056 --> 00:34:44.656
That is essentially, this moral
00:34:44.656 --> 00:34:47.744
compass that can actually drive AI to
00:34:47.744 --> 00:34:50.874
think, "Oh, this is a good thing for me
00:34:50.874 --> 00:34:53.552
to be doing or for me to be thinking about." And
00:34:53.552 --> 00:34:56.664
even, they found… I forget what the study was,
00:34:56.664 --> 00:34:59.196
and maybe I, need to go back and look at the
00:34:59.196 --> 00:35:02.183
source, but it was like, if you're nice to
00:35:02.183 --> 00:35:05.259
someone in a chat on, Stack Overflow, you're
00:35:05.259 --> 00:35:08.079
more likely to get a good response than if
00:35:08.079 --> 00:35:11.096
you're mean to them. And you take those little
00:35:11.096 --> 00:35:13.828
subtle things that n- were meaningless maybe at
00:35:13.828 --> 00:35:16.836
the time to the poster, and that could be,
00:35:16.836 --> 00:35:20.017
like, the foundation for how AI interacts
00:35:20.017 --> 00:35:22.878
with us. It's just really fascinating. I'm such a
00:35:22.878 --> 00:35:25.616
big, optimist, I think, for where we'll be. It's
00:35:25.616 --> 00:35:27.697
gonna be hard. There are gonna be big transitions-
00:35:27.697 --> 00:35:30.780
Yeah and big disruptions, but, yeah, I'm a huge,
00:35:30.780 --> 00:35:33.885
tech optimist, I think. I wanna pivot just
00:35:33.885 --> 00:35:35.995
for a second because I'm thinking about some of
00:35:35.995 --> 00:35:38.114
the language that we use around what we're
00:35:38.114 --> 00:35:40.692
delivering, and I know you're both close to this.
00:35:41.473 --> 00:35:44.190
I saw something recently about organizations
00:35:44.190 --> 00:35:47.286
building out MCP teams, and I think it's a
00:35:47.286 --> 00:35:50.003
funny idea. like- Maybe
00:35:50.003 --> 00:35:52.854
'cause I've been through this a
00:35:52.854 --> 00:35:55.883
couple of times, I lived through,
00:35:55.883 --> 00:35:59.007
as I mentioned, service-oriented architectures
00:35:59.007 --> 00:36:02.042
or APIs, and it seems like we develop new terms
00:36:02.042 --> 00:36:05.008
for almost very similar things that we've done in
00:36:05.008 --> 00:36:07.904
the past. But when you think about MCP and
00:36:10.165 --> 00:36:12.839
the attention it's getting, do you think that is
00:36:12.839 --> 00:36:15.686
gonna persist, that we're always gonna be thinking
00:36:15.686 --> 00:36:17.215
about the MCP as a traffic
00:36:17.747 --> 00:36:20.935
orchestrator, inside the enterprise?
00:36:20.956 --> 00:36:23.336
Or do you think that this is still the beginning
00:36:23.336 --> 00:36:24.793
of something that's evolving?
00:36:26.336 --> 00:36:28.855
I think it's still evolving, for sure. I think
00:36:28.855 --> 00:36:31.287
right now MCPs are really important and a point of
00:36:31.287 --> 00:36:33.957
discussion with a lot of customers because of the
00:36:33.957 --> 00:36:36.617
security risk that they pose. A lot of
00:36:36.617 --> 00:36:39.807
customers that really value being super in
00:36:39.807 --> 00:36:42.743
control of everything that's going in and out
00:36:42.743 --> 00:36:45.382
will have MCPs disabled for that reason, because
00:36:45.382 --> 00:36:47.260
they can't really control what's going in and out
00:36:47.260 --> 00:36:50.367
with those, and it requires a lot of governance
00:36:50.367 --> 00:36:53.501
from their end. I do think, however,
00:36:53.501 --> 00:36:56.509
that'll evolve as, we've seen these things evolve
00:36:56.509 --> 00:36:59.441
in the past few years. I think that
00:36:59.441 --> 00:37:02.198
there will be more novel solutions that maybe come
00:37:02.198 --> 00:37:04.791
about with integrations that are
00:37:04.791 --> 00:37:07.940
natively more secure maybe than MCP. And
00:37:07.940 --> 00:37:10.827
at the end of the day, like MCPs are just, like
00:37:10.827 --> 00:37:13.695
another way to do an API. So again, I-- to the
00:37:13.695 --> 00:37:16.722
point that it's a lot of the same concepts being
00:37:16.725 --> 00:37:19.846
done differently, I think we'll see
00:37:19.846 --> 00:37:22.707
it evolve or c- like clients and
00:37:22.709 --> 00:37:24.976
customers will figure out a way to make it more
00:37:24.976 --> 00:37:27.751
secure, 'cause I think as it stands, it poses a
00:37:27.751
--> 00:37:30.478
big security risk for like financial
00:37:30.478 --> 00:37:33.087
institutions or healthcare institutions, and
00:37:33.087 --> 00:37:36.063
they're a lot more hesitant with those. Yeah.
00:37:36.087 --> 00:37:39.075
What's old is new again, I think, with, a
00:37:39.075 --> 00:37:42.006
lot of these solutions, and it really is
00:37:42.006 --> 00:37:45.095
interesting that it all goes back to the
00:37:45.095 --> 00:37:47.769
fundamentals or the foundation and
00:37:47.769 --> 00:37:50.889
standing up strong metrics. How are you actually
00:37:50.889 --> 00:37:53.646
measuring what you're doing? Just doing something
00:37:53.646 --> 00:37:55.896
simple like that so that you have a target to
00:37:55.896 --> 00:37:58.941
shoot at and you can continue to, make progress.
00:37:59.393 --> 00:38:02.328
you look at just the simple building blocks of
00:38:02.328 --> 00:38:05.391
putting together strong, software or,
00:38:05.391 --> 00:38:08.348
like strong engineering values, those
00:38:08.348 --> 00:38:11.106
are still the things that push the
00:38:11.106 --> 00:38:13.671
projects in a good way and, the
00:38:13.671 --> 00:38:16.828
foundation… it seems like we, we do-- you're
00:38:16.828 --> 00:38:19.649
right, I think that we invent maybe, new words or
00:38:19.654 --> 00:38:22.720
terms or, you know, we're continuing to,
00:38:22.720 --> 00:38:25.560
expand or grow our knowledge, but it
00:38:25.560 --> 00:38:28.109
comes back to those core, core
00:38:28.109 --> 00:38:30.904
contracts or core, methodologies,
00:38:31.695 --> 00:38:34.408
It's interesting to me, we talk about this and I,
00:38:34.408 --> 00:38:36.788
and while I think all of this is part of the
00:38:36.788 --> 00:38:39.398
software development life cycle, we actually don't
00:38:39.398 --> 00:38:42.591
mention it that much because of the… I
00:38:42.591 --> 00:38:45.285
think we're adjusting our own
00:38:45.285 --> 00:38:48.239
expectations to the velocity that we can
00:38:48.239 --> 00:38:51.326
get, and I'm not sure the processes have caught up
00:38:51.326 --> 00:38:53.762
yet, and I think we're starting to get there.
00:38:54.828 --> 00:38:57.204
and all the things that you have talked about
00:38:57.204 --> 00:38:59.817
today are the pieces and parts that we need to
00:38:59.817 --> 00:39:02.760
understand as we move forward to create patterns
00:39:03.103 --> 00:39:06.010
for our teams and our colleagues. the
00:39:06.010 --> 00:39:08.495
question I wanna end with for both of you is, and
00:39:08.495 --> 00:39:10.510
Kurt, I love that you said you're an optimist
00:39:10.510 --> 00:39:12.790
about this. I tend to be also.
00:39:13.676 --> 00:39:16.678
my family and friends sometimes, tsk-tsk
00:39:16.678 --> 00:39:19.670
me because they just think I'm not being… I
00:39:19.670 --> 00:39:22.219
don't know. But I'm-- I don't know, I think it's
00:39:22.219 --> 00:39:24.109
cool. I like robots.
00:39:25.397 --> 00:39:28.123
so what are you most excited
about or
00:39:28.123 --> 00:39:30.954
anticipate for the next 18 months? What
00:39:30.954 --> 00:39:33.623
things do you think, "I think this is on the
00:39:33.623 --> 00:39:35.838
horizon," whether it's something tactical
00:39:35.838 --> 00:39:38.181
operationally you're doing now or something that
00:39:38.322 --> 00:39:41.205
we haven't even envisioned yet? So I'll
00:39:41.205 --> 00:39:43.258
take a start here.
00:39:44.859 --> 00:39:48.009
I feel like the bad guy when you both and Kurt
00:39:48.009 --> 00:39:50.941
are saying you're more optimist. I'm a little
00:39:50.941 --> 00:39:53.815
more skeptical. The AI environment
00:39:53.815 --> 00:39:56.754
reminds me a lot of nuclear
00:39:56.754 --> 00:39:59.823
weapon development and how that came to
00:39:59.823 --> 00:40:02.045
be. It's the same, it's just another, like there's
00:40:02.045 --> 00:40:03.890
an AI race, and who's gonna get to the finish
00:40:03.890 --> 00:40:06.994
line. And I, don't think I'm quite at the extent
00:40:06.994 --> 00:40:10.020
of, the OpenAI an-anthropic researchers that
00:40:10.020 --> 00:40:12.914
think it's, a lot worse than it is. But I
00:40:12.914 --> 00:40:16.090
think there's concerns with it that need to be
00:40:16.090 --> 00:40:19.233
taken into account as we use AI.
00:40:19.245 --> 00:40:21.555
There's ethical boundaries that need to be
00:40:21.555 --> 00:40:24.505
respected when you're creating these foundational
00:40:24.505 --> 00:40:27.613
models that now, like I feel every week with,
00:40:27.613 --> 00:40:30.084
OpenAI's Astra, with Mythos, and I feel like
00:40:30.084 --> 00:40:32.761
Anthropic has a good take with this, like we still
00:40:32.761 --> 00:40:35.002
don't have access to Mythos, only some
00:40:35.002 --> 00:40:37.992
organizations do. Just taking that approach to
00:40:37.992 --> 00:40:41.126
seeing how things go, and I feel like with how
00:40:41.126 --> 00:40:43.675
quickly they are making these AI models
00:40:43.675 --> 00:40:46.569
significantly better, I think in the future,
00:40:46.569 --> 00:40:49.717
like in the next year or so, it will again get
00:40:49.717 --> 00:40:52.821
to a point where I think you can start seeing
00:40:52.821 --> 00:40:56.008
robotics come more into like normal day-to-day
00:40:56.008 --> 00:40:59.004
interactions. 'Cause again, like when I was in
00:40:59.004 --> 00:41:01.438
school, it was like 80 years before robots are
00:41:01.438 --> 00:41:03.569
interacting with people on a more day-to-day
00:41:03.569 --> 00:41:06.481
basis. You have your like robot vacuum,
00:41:06.481 --> 00:41:09.657
and that's an advanced robot as you're gonna get.
00:41:09.955 --> 00:41:12.203
And again, these were like professors of robotics,
00:41:12.203 --> 00:41:14.709
and they were deeply convinced in this because no
00:41:14.709 --> 00:41:17.619
one saw this happening and the boom and how good
00:41:17.619 --> 00:41:20.397
these things would get. I think that we're
00:41:20.397 --> 00:41:22.417
starting to see them with some like innovative
00:41:22.417 --> 00:41:25.598
labs that have the robot that can help you do your
00:41:25.598 --> 00:41:28.275
laundry. But I think we'll see more of, and
00:41:28.275 --> 00:41:30.832
whether people like it or not, AI doing
00:41:30.832 --> 00:41:33.927
non-traditional jobs that you wouldn't have
00:41:33.927 --> 00:41:36.482
thought. We have autonomous cars. Why not have
00:41:36.482 --> 00:41:38.972
more things become more autonomous? I think we're
00:41:38.972 --> 00:41:41.495
gonna see more of that in the next year. Thanks.
00:41:41.495 --> 00:41:41.957
And Kurt?
00:41:44.103 --> 00:41:44.428
Yeah.
00:41:46.684 --> 00:41:49.147
I have no idea where we'll be in a year actually.
00:41:49.165 --> 00:41:51.674
What I'm excited for, where I think we're going,
00:41:56.258 --> 00:41:59.226
I don't know. I think we are
00:41:59.226 --> 00:42:01.951
on the knuckle of some
00:42:01.951 --> 00:42:04.440
big, improvement, in
00:42:04.440 --> 00:42:07.545
intelligence, overall. And
00:42:07.545 --> 00:42:10.164
it's been pretty crazy reading
00:42:10.164 --> 00:42:13.290
about, the stories of, what happened
00:42:13.308 --> 00:42:16.111
with, the OpenAI research
00:42:16.111 --> 00:42:18.807
and, their agent testing
00:42:18.807 --> 00:42:21.817
gone awry, and, what these models
00:42:21.817 --> 00:42:24.570
could potentially do, going out and replicating
00:42:24.570 --> 00:42:25.431
themselves or,
00:42:26.580 --> 00:42:28.426
occupying different spaces of the web.
00:42:31.071 --> 00:42:34.264
and we definitely need the guardrails, to
00:42:34.264 --> 00:42:37.297
speak to that, but I just believe that
00:42:37.297 --> 00:42:40.413
humanity skews to the good. And so we
00:42:40.413 --> 00:42:42.770
have a bunch of big brains focused on this
00:42:42.770 --> 00:42:44.985
problem, and we have a bunch of good people that
00:42:44.985 --> 00:42:46.886
are just coming together to try to figure it out.
00:42:46.950 --> 00:42:49.898
And, for each new issue that comes up, I
00:42:49.898 --> 00:42:53.057
think we'll have just the right people find the
00:42:53.057 --> 00:42:55.635
right solutions to continue to grow and develop
00:42:55.635 --> 00:42:57.953
this. but robotics is
00:42:57.953 --> 00:43:00.942
fascinating. I think it'll be crazy to
00:43:00.942 --> 00:43:03.955
see the Optimus robots walking on the
00:43:03.955 --> 00:43:06.430
street or, going and getting groceries or
00:43:06.430 --> 00:43:09.222
something. but yeah, again, we're
00:43:09.222 --> 00:43:12.419
living in the future, so any and all things are,
00:43:12.761 --> 00:43:15.453
on the table right now, I think. I'm actually
00:43:15.453 --> 00:43:18.624
curious- That's awesome … about the, the robot
00:43:18.624 --> 00:43:21.441
delivery. I'm surprised I don't see more drones
00:43:21.441 --> 00:43:24.631
maybe flying over with, robot delivery, but
00:43:24.631 --> 00:43:27.722
who knows? Next week is a, a new week
00:43:27.722 --> 00:43:30.881
for, new companies to go out and grow and
00:43:30.964 --> 00:43:33.971
explore, I think, in this space. I did see
00:43:33.971 --> 00:43:36.820
robots on the sidewalks maybe in, I think I
00:43:36.820 --> 00:43:39.670
might have been in Miami, and they were
00:43:39.670 --> 00:43:42.334
doing, robot delivery, which, I
00:43:42.334 --> 00:43:44.916
thought was super cool. I
00:43:44.916 --> 00:43:47.881
appreciate this conversation. both of
00:43:47.881 --> 00:43:50.703
you are, day-to-day, hands-on
00:43:50.703 --> 00:43:53.336
involved in this innovation
00:43:53.336 --> 00:43:56.488
cycle, so I'm fortunate to be able to talk to you both.
00:43:57.667 --> 00:44:00.532
I look forward to when we talk next and we hear
00:44:00.532 --> 00:44:03.986
about the future innovations that you're both achieving.
00:44:05.230 --> 00:44:06.314
and thank you again.
00:44:07.317 --> 00:44:10.508
I'm MK. This has been Ready, Set,
00:44:10.508 --> 00:44:13.395
Runtime, an AHEAD podcast, and we'll see you next time.
00:44:16.796 --> 00:44:19.410
Thank you for listening to Ready, Set, Runtime.
00:44:19.886 --> 00:44:22.822
If something in today's conversation sparked an
00:44:22.822 --> 00:44:25.322
idea or a question about your own environment,
00:44:25.322 --> 00:44:28.287
AHEAD's team would love to hear from you. Reach
00:44:28.287 --> 00:44:31.401
out anytime at ahead.com, and if you haven't
00:44:31.401 --> 00:44:34.052
already, follow the show wherever you're tuning
00:44:34.052 --> 00:44:37.035
in. The views shared on this podcast are those of
00:44:37.035 --> 00:44:39.581
the speakers and don't necessarily reflect those
00:44:39.581 --> 00:44:41.708
of AHEAD. Until next time