TRANSCRIPT 🔗
WEBVTT
00:00:00.020 --> 00:00:04.580
Picture a dam built for a calm river. Now the
00:00:04.580 --> 00:00:07.599
river's running way faster than anyone planned
00:00:07.599 --> 00:00:10.789
for. And the water is pouring right over the
00:00:10.789 --> 00:00:14.679
top. That's what AI-generated code is doing to
00:00:14.679 --> 00:00:17.820
everything after the pull request. The reviews
00:00:17.820 --> 00:00:20.920
and approvals downstream were built for people
00:00:20.920 --> 00:00:24.620
working eight hour days. My guest today says
00:00:24.620 --> 00:00:28.100
AI feels like alien technology that dropped from
00:00:28.100 --> 00:00:30.969
the sky and we don't know how to control it
00:00:30.969 --> 00:00:34.289
yet. But his answer isn't less automation, it's
00:00:34.289 --> 00:00:37.229
more. And he thinks that we should onboard an
00:00:37.229 --> 00:00:41.429
AI agent the same way we'd onboard a new engineer.
00:00:41.729 --> 00:00:44.250
I'm Brian Teller from Teller's Tech, and this
00:00:44.250 --> 00:00:45.602
is Ship It Weekly.
00:01:02.572 --> 00:01:04.180
Welcome back to Ship It Weekly,
00:01:04.359 --> 00:01:07.400
where I filter the noise and focus on what matters
00:01:07.400 --> 00:01:09.939
when you are the one running infrastructure and
00:01:09.939 --> 00:01:13.459
owning reliability. Most weeks, it's a quick
00:01:13.459 --> 00:01:17.750
DevOps, SRE, platform, cloud, and security news
00:01:17.750 --> 00:01:21.310
recap. In between those, I do conversation episodes
00:01:21.310 --> 00:01:24.930
with people building and operating these systems.
00:01:25.870 --> 00:01:29.790
Ship It Weekly turns one this November. To celebrate,
00:01:29.969 --> 00:01:32.870
through mid-December, you're getting two conversations
00:01:32.870 --> 00:01:36.450
a week, Mondays and Wednesdays, with the news
00:01:36.450 --> 00:01:40.170
every Friday. Then, we're back to our usual rhythm.
00:01:40.390 --> 00:01:45.269
Today, I'm joined by Darko Fabijan, CEO of SuperPlane
00:01:45.269 --> 00:01:49.790
and co-founder of Semaphore, the CI/CD platform.
00:01:50.209 --> 00:01:53.790
We talk about what starts breaking after the
00:01:53.790 --> 00:01:56.709
pull request when AI is writing more of the code,
00:01:56.870 --> 00:02:00.030
what should sit between an AI agent and production,
00:02:00.290 --> 00:02:03.609
why he thinks you should onboard an agent the
00:02:03.609 --> 00:02:07.450
way you'd onboard a new engineer, and the first
00:02:07.450 --> 00:02:11.590
workflow he'd actually trust AI with on an ops
00:02:11.590 --> 00:02:13.938
team. All right. Let's jump in.
00:02:18.586 --> 00:02:19.000
Darko, thank
00:02:19.000 --> 00:02:20.759
you for joining me. Thank you for having me.
00:02:21.180 --> 00:02:24.000
Developers are using AI to write more code faster.
00:02:24.180 --> 00:02:27.520
What starts breaking once that code hits the
00:02:27.520 --> 00:02:30.080
real software delivery and operations workflow?
00:02:30.539 --> 00:02:33.699
Well, yeah, I think that we are very fast with
00:02:33.699 --> 00:02:37.340
generating code. I mean, some are also very fast
00:02:37.340 --> 00:02:39.930
at putting it into production, but there are
00:02:39.930 --> 00:02:41.990
just some numbers that are not as they used to
00:02:41.990 --> 00:02:44.189
be. So I think definitely it's a time of, you
00:02:44.189 --> 00:02:46.349
know, stability and, you know, various other
00:02:46.349 --> 00:02:48.789
metrics when it comes to the quality of, you
00:02:48.789 --> 00:02:51.310
know, what is more hitting the developers back.
00:02:52.129 --> 00:02:54.469
But, you know, for the operations people, it's
00:02:54.469 --> 00:02:56.270
just that code that we are shipping, you know,
00:02:56.270 --> 00:02:59.990
it just has more bugs and it's just by nature
00:02:59.990 --> 00:03:02.479
of things. Pull requests are bigger. They are
00:03:02.479 --> 00:03:05.479
more frequent. The pressure is all over the place
00:03:05.479 --> 00:03:07.979
just within the industry of like making software.
00:03:08.120 --> 00:03:11.020
So there are like a couple of forces that are,
00:03:11.039 --> 00:03:14.340
you know, just very new and things are breaking
00:03:14.340 --> 00:03:17.460
in various shapes and forms. Yeah, I would not
00:03:17.460 --> 00:03:19.520
say only in the production and operation, but
00:03:19.520 --> 00:03:22.060
kind of also in the society a little bit. It's
00:03:22.060 --> 00:03:24.020
just the velocity and the amount of things that
00:03:24.020 --> 00:03:26.699
we're putting out there. Is the post-PR world
00:03:26.699 --> 00:03:29.479
becoming the harder part of engineering? Or you
00:03:29.479 --> 00:03:31.259
had mentioned the social part as well, but what
00:03:31.259 --> 00:03:34.159
is becoming harder specifically from the changes
00:03:34.159 --> 00:03:37.520
with AI? I would say, I mean, operation side
00:03:37.520 --> 00:03:40.860
of things is becoming harder. So if I'm just
00:03:40.860 --> 00:03:44.319
thinking from these numbers perspective, so the
00:03:44.319 --> 00:03:47.280
complexity in the software is increasing just
00:03:47.280 --> 00:03:49.680
by the numbers of lines of code that are being
00:03:49.680 --> 00:03:52.379
put out there. The frequency of those deliveries
00:03:52.379 --> 00:03:56.069
is increasing. I mean, in the good old, you know,
00:03:56.210 --> 00:03:59.490
CI, CD practice, continuous integration as a
00:03:59.490 --> 00:04:02.110
practice is you want to ship, you know, small
00:04:02.110 --> 00:04:05.169
chunks of changes because then you minimize,
00:04:05.210 --> 00:04:07.310
you know, the blast radius, you minimize the
00:04:07.310 --> 00:04:09.729
surface area. Now the pull requests are just,
00:04:09.750 --> 00:04:12.810
you know, bigger. It's getting kind of out of
00:04:12.810 --> 00:04:14.830
hand, you know, what's the size of the thing
00:04:14.830 --> 00:04:19.290
that we are shipping. I mean, I'm seeing it in,
00:04:19.310 --> 00:04:21.370
you
Load full transcript
know, in our team and I'm hearing it all
00:04:21.370 --> 00:04:23.699
over the place. You know, those pull requests
00:04:23.699 --> 00:04:25.939
of a couple of thousands of lines of code, you
00:04:25.939 --> 00:04:28.240
know, who can manage that. And I think that,
00:04:28.300 --> 00:04:31.180
I mean, it's not, I think it is getting out of,
00:04:31.180 --> 00:04:33.300
you know, cognitive capacity of people to just
00:04:33.300 --> 00:04:35.500
want to review all those pull requests. And it's
00:04:35.500 --> 00:04:37.439
just that, you know, things are getting into
00:04:37.439 --> 00:04:41.879
production that are not really that well reviewed
00:04:41.879 --> 00:04:44.819
and are causing the issues on the other side.
00:04:45.310 --> 00:04:48.509
I don't know the exact URL, obviously, by heart,
00:04:48.610 --> 00:04:51.589
but there is a Faros.ai report that recently
00:04:51.589 --> 00:04:55.050
came out. And I think we can link it probably
00:04:55.050 --> 00:04:58.449
somewhere. There is an increase in the number
00:04:58.449 --> 00:05:00.889
of defects that are being shipped, in the number
00:05:00.889 --> 00:05:04.430
of incidents. So it's all between 50% and 200%.
00:05:04.430 --> 00:05:08.290
That's my kind of rough understanding and memory
00:05:08.290 --> 00:05:10.430
of how those charts look. And it's all in red.
00:05:10.589 --> 00:05:14.339
So yeah, that's one side of it. I've seen this
00:05:14.339 --> 00:05:16.300
in my day job as well. Like we have AI agents
00:05:16.300 --> 00:05:19.199
creating code using Claude Code or Cursor or whatever.
00:05:19.360 --> 00:05:21.800
And then you have AI agents reviewing code. Like
00:05:21.800 --> 00:05:24.459
you have like a Codex plugin on maybe GitHub
00:05:24.459 --> 00:05:27.620
that reviews the PR. And then you have maybe
00:05:27.620 --> 00:05:30.959
even AI in your CI/CD pipeline, like in your actual
00:05:30.959 --> 00:05:33.699
delivery pipelines as well, where it's interfacing
00:05:33.699 --> 00:05:36.740
with the production. Where should AI agents not
00:05:36.740 --> 00:05:39.339
exist? Like where should we not have AI or still
00:05:39.339 --> 00:05:42.649
have like very specific? Because again, it's
00:05:42.649 --> 00:05:45.449
non-deterministic, right? So at a certain point,
00:05:45.550 --> 00:05:49.389
how can we reduce that risk? Well, one way we
00:05:49.389 --> 00:05:53.930
can reduce it is by having more AI. But it takes
00:05:53.930 --> 00:05:57.350
time to figure out how exactly and where to put
00:05:57.350 --> 00:06:01.629
that AI. So in terms of unsolved problems, it's
00:06:01.629 --> 00:06:04.410
definitely the right access to production systems.
00:06:04.850 --> 00:06:08.709
So people are not really into that. There are
00:06:08.709 --> 00:06:11.250
some people that are. I think it's, you know,
00:06:11.290 --> 00:06:14.949
sub 1% of the industry at best, but people are
00:06:14.949 --> 00:06:18.509
trying to solve that. And we are also trying
00:06:18.509 --> 00:06:21.610
to solve that also in our own ways and industry
00:06:21.610 --> 00:06:24.649
as a whole. In the broadest possible sense, I
00:06:24.649 --> 00:06:28.519
would say I understand the situation. Yes, AI
00:06:28.519 --> 00:06:30.620
exists for the last couple of years in this form
00:06:30.620 --> 00:06:33.860
that is, you know, usable, but in just like last,
00:06:34.100 --> 00:06:36.620
you know, six to nine or something like that,
00:06:36.779 --> 00:06:38.800
it has really become, you know, super powerful,
00:06:38.899 --> 00:06:42.180
you know. And I perceive it as that, you know,
00:06:42.240 --> 00:06:46.500
alien technology has dropped from the sky and
00:06:46.750 --> 00:06:50.269
we don't know how to control it. Part of it is
00:06:50.269 --> 00:06:54.610
that in our process of producing software, there
00:06:54.610 --> 00:06:57.449
are still so many things that are manual in various
00:06:57.449 --> 00:06:59.949
ways. I'm not saying that we are creating EC2
00:06:59.949 --> 00:07:02.449
instances by hand from console and all of that,
00:07:02.529 --> 00:07:05.850
but the processes are just not made for the AI.
00:07:06.370 --> 00:07:08.889
Processes are made for humans. So there is, you
00:07:08.889 --> 00:07:11.199
know... The time that the human being, you know,
00:07:11.220 --> 00:07:15.120
can, you know, review something, understand something,
00:07:15.339 --> 00:07:17.899
you know, work those eight hours a day or, you
00:07:17.899 --> 00:07:20.420
know, a little bit less, more on the day and
00:07:20.420 --> 00:07:23.920
so on. So that is a very recent phenomenon of
00:07:23.920 --> 00:07:27.699
us being to create so much code at such a velocity.
00:07:28.019 --> 00:07:30.560
But other parts of the system are just not made
00:07:30.560 --> 00:07:34.879
for the AI yet. And that has to change, you know,
00:07:35.000 --> 00:07:39.379
iteratively over time. In the same way that we
00:07:39.379 --> 00:07:41.339
are introducing change in the production systems,
00:07:41.620 --> 00:07:46.420
we want to make one change at a time and control
00:07:46.420 --> 00:07:49.639
the risk. And that whole pipeline has to change
00:07:49.639 --> 00:07:52.339
fundamentally, but it cannot change overnight.
00:07:52.459 --> 00:07:55.300
We have to figure out and understand how to change
00:07:55.300 --> 00:07:58.639
it so that velocity that we got to that AI creation
00:07:58.639 --> 00:08:02.199
side. So yes, I can write like 10,000 lines
00:08:02.199 --> 00:08:05.139
of code in three hours of live coding, but how
00:08:05.139 --> 00:08:07.300
the rest of the system... can really deliver
00:08:07.300 --> 00:08:09.639
it and work and how we can handle that amount
00:08:09.639 --> 00:08:12.759
of change. So that has not been figured out.
00:08:13.160 --> 00:08:17.980
So kind of our main hypothesis is that there
00:08:17.980 --> 00:08:20.759
needs to be more automation and we have to figure
00:08:20.759 --> 00:08:23.699
out what
that automation is and what's the nature
00:08:23.699 --> 00:08:26.620
of it to be able to really fully harness and
00:08:26.620 --> 00:08:29.180
control, I would say, mainly control the risk
00:08:29.180 --> 00:08:32.539
that this power that we just recently got has
00:08:32.539 --> 00:08:36.070
introduced. So I would say that's... part of
00:08:36.070 --> 00:08:38.610
the, that's a part of the, of the answer. I mean,
00:08:38.610 --> 00:08:40.809
definitely, other, the other piece that I mentioned
00:08:40.809 --> 00:08:43.710
is how do we give agents the right access to
00:08:43.710 --> 00:08:46.149
the production systems. So policy enforcement,
00:08:46.149 --> 00:08:49.210
that sort of thing. Yeah, or least privilege.
00:08:49.210 --> 00:08:53.789
Yeah. We need to figure out like what that kind
00:08:53.789 --> 00:08:57.159
of automation is and how that layer works. Is
00:08:57.159 --> 00:09:00.059
that like a fully deterministic layer that exists
00:09:00.059 --> 00:09:02.139
between, you know, AI agent and the production
00:09:02.139 --> 00:09:04.759
systems? I currently believe that that's a part
00:09:04.759 --> 00:09:06.919
of the equation and then it's going to be part
00:09:06.919 --> 00:09:09.360
of the equation for, you know, significant period
00:09:09.360 --> 00:09:12.860
of time. There are some people that we are operating
00:09:12.860 --> 00:09:15.419
against, like, that hypothesis. There has to be
00:09:15.419 --> 00:09:17.740
a deterministic layer in between and there are
00:09:17.740 --> 00:09:20.100
automations and right access that we give to
00:09:20.100 --> 00:09:22.480
the to production system to agents, but it's very
00:09:22.480 --> 00:09:25.789
scoped, very limited, goes through the approval
00:09:25.789 --> 00:09:28.610
process. But as I'm speaking with, you know,
00:09:28.629 --> 00:09:30.409
various people from industries and some, you
00:09:30.409 --> 00:09:33.289
know, very, you know, advanced companies recently
00:09:33.289 --> 00:09:36.289
spoke with a guy from Nvidia. So they are trying
00:09:36.289 --> 00:09:38.769
to figure out how to give more, you know, right
00:09:38.769 --> 00:09:41.429
production to their systems, you know, in their
00:09:41.429 --> 00:09:43.870
own way. And that's also our hypothesis that
00:09:43.870 --> 00:09:48.470
there has to be that guardrail, you know, layer
00:09:48.470 --> 00:09:52.320
in between. To be able to do that. Yeah. I'm
00:09:52.320 --> 00:09:55.460
curious, what are your thoughts on that right
00:09:55.460 --> 00:09:57.460
access to the production system? Because it is
00:09:57.460 --> 00:10:00.940
a very uncomfortable thought to a lot of people.
00:10:01.559 --> 00:10:05.720
Yeah. So, well, OPA policies, Sentinel policies,
00:10:05.779 --> 00:10:08.980
as far as like allowing changes and then on the
00:10:08.980 --> 00:10:12.539
actual production side change, having approval
00:10:12.539 --> 00:10:15.299
gates, like having least privilege and maybe
00:10:15.299 -->
00:10:18.019
gating those AI agents, I guess would be my answer.
00:10:18.139 --> 00:10:20.570
I don't know that it's... That's a full answer,
00:10:20.649 --> 00:10:24.250
though. A lot of the AI safety advice says keep
00:10:24.250 --> 00:10:27.070
a human in the loop. But what I find is in a
00:10:27.070 --> 00:10:29.590
lot of real ops, humans end up just rubber stamping
00:10:29.590 --> 00:10:33.250
stuff all the time anyway. So that part, I don't
00:10:33.250 --> 00:10:35.289
know how to fix. How do you avoid those fake
00:10:35.289 --> 00:10:37.350
approvals where it's just rubber stamping the
00:10:37.350 --> 00:10:40.389
AI changes? That's a harder challenge that I'm
00:10:40.389 --> 00:10:41.909
not sure. And I think if you can solve that,
00:10:42.070 --> 00:10:45.590
it makes way to the least privileged conversation,
00:10:45.789 --> 00:10:49.429
the OPA policy conversation is an easier gating
00:10:49.429 --> 00:10:52.250
strategy. The harder part, I think, is the human
00:10:52.250 --> 00:10:55.070
side. Like, how do you fix that human side? I
00:10:55.070 --> 00:10:59.169
mean, I recently heard that, yes, AI agents hallucinate
00:10:59.169 --> 00:11:02.789
and are unreliable, but also humans are hallucinating
00:11:02.789 --> 00:11:05.269
and are unreliable. So that side of the thing.
00:11:05.429 --> 00:11:08.789
I mean, whenever there is a system that has a
00:11:08.789 --> 00:11:11.509
lot of something in it, you have to... quantify
00:11:11.509 --> 00:11:13.909
various kinds of things. So there is like a code
00:11:13.909 --> 00:11:16.350
coverage, let's say, if you're developing something,
00:11:16.549 --> 00:11:19.559
is everything covered by tests? What's the quality
00:11:19.559 --> 00:11:21.539
of that code? There are like various metrics
00:11:21.539 --> 00:11:23.580
to assess that. And there's, you know, change,
00:11:23.759 --> 00:11:25.580
any change, you know, pull request, either your
00:11:25.580 --> 00:11:28.820
Terraform code or on your, you know, Java code,
00:11:28.940 --> 00:11:32.379
it doesn't matter. There are various, you know,
00:11:32.580 --> 00:11:36.480
heuristics, numbers that can quantify how dangerous
00:11:36.480 --> 00:11:40.159
is that change. So as we are shipping, you know,
00:11:40.649 --> 00:11:43.190
this code, I mean, last week, I mean, we, we shipped
00:11:43.190 --> 00:11:45.850
kind of 50, 50, you know, thousand lines of code,
00:11:45.850 --> 00:11:49.110
you know, and it's, you know, five, six people. Okay,
00:11:49.110 --> 00:11:51.529
we are rushing and, you know, it is, you know, kind
00:11:51.529 --> 00:11:54.210
of pre-production software and, you know, all
00:11:54.210 --> 00:11:57.769
of that. But it's, it's such a, such a big, you know,
00:11:57.769 --> 00:11:59.690
such a big change that we are introducing the
00:11:59.690 --> 00:12:02.470
system that I was thinking in terms of just pull
00:12:02.470 --> 00:12:05.110
requests, so can we have some kind of like metric
00:12:05.110 --> 00:12:08.909
in place? So if this change is just changing maybe
00:12:08.909 --> 00:12:11.860
the, you know, but name of the bucket, you know,
00:12:11.860 --> 00:12:14.100
if it's like, you know, Terraform or, you know, changing
00:12:14.100 --> 00:12:16.840
some, some smaller thing, you know, it should get
00:12:16.840 --> 00:12:20.139
the score. So it's not very dangerous. If I'm deleting
00:12:20.139 --> 00:12:23.539
the resource which is a data storage thing, either
00:12:23.539 --> 00:12:26.779
S3 bucket, RDS instance, I mean, that's 10 out of
00:12:26.779 --> 00:12:30.259
10. That, that's a red. I have not, maybe I don't
00:12:30.259 --> 00:12:33.100
know, but I haven't seen that kind of like metric
00:12:33.100 --> 00:12:36.360
that exists yet in the industry either on the
00:12:36.360 --> 00:12:39.179
development side or the ops side. I think that we
00:12:39.179 --> 00:12:42.080
need that kind of thing because of the rubber
00:12:42.080 --> 00:12:44.940
stamping that you're mentioning. Sooner or later,
00:12:45.019 --> 00:12:46.840
someone is going to be tired and there is going
00:12:46.840 --> 00:12:49.720
to be a line which is going to be like drop all
00:12:49.720 --> 00:12:52.259
tables because, I mean, we have seen that recently,
00:12:52.299 --> 00:12:54.919
last couple of months. There have been some drop
00:12:54.919 --> 00:12:58.840
all tables scenarios around. So, yeah. Yeah,
00:12:58.899 --> 00:13:00.919
I was trying to think of that because, yeah,
00:13:00.960 --> 00:13:03.620
dropping an instance or deleting something from
00:13:03.620 --> 00:13:07.080
a bucket or removing a database, that's an easy...
00:13:07.419 --> 00:13:10.840
gate to have, and that's an easy score to score
00:13:10.840 --> 00:13:14.080
on a metric. The problem I would think is with
00:13:14.080 --> 00:13:16.340
ancillary processes, like a Lambda that maybe
00:13:16.340 --> 00:13:19.820
does some data manipulation that could potentially
00:13:19.820 --> 00:13:23.059
also, maybe there's an unintended change and
00:13:23.059 --> 00:13:25.279
it's down the road and you're deleting objects,
00:13:25.460 --> 00:13:28.799
you know, and that's not, it's harder to understand,
00:13:28.860 --> 00:13:31.500
I think, to actually quality gate that because
00:13:31.500 --> 00:13:33.879
there could be related resources with unintended
00:13:33.879 --> 00:13:36.720
consequences further down the chain. Absolutely.
00:13:37.559 --> 00:13:39.639
Absolutely. But I mean, that's an area where
00:13:39.639 --> 00:13:42.320
theoretically, I mean, we spoke about AI writing
00:13:42.320 --> 00:13:45.080
code and reviewing it, then, you know, approving
00:13:45.080 --> 00:13:48.320
it. Yeah, no, I know. Yeah, that's a hard problem.
00:13:48.419 --> 00:13:50.779
Yeah, it's not solved either. I would say just
00:13:50.779 --> 00:13:55.320
the area of like qualifying the risk. That's
00:13:55.320 --> 00:13:58.399
a billion dollar startup idea over there. And
00:13:58.399 --> 00:14:01.809
I mean, it's just, there's a lot of work to be
00:14:01.809 --> 00:14:04.590
done in the area. I guess there are some, you
00:14:04.590 --> 00:14:06.960
know, research going around some universities
00:14:06.960 --> 00:14:11.440
around qualifying that. So you have all of these
00:14:11.440 --> 00:14:14.179
different services that are part of your control
00:14:14.179 --> 00:14:16.879
plane, and you have scripts that are all glued
00:14:16.879 --> 00:14:20.299
together. Most ops teams already have some version
00:14:20.299 --> 00:14:22.419
of this held together with GitHub Actions and
00:14:22.419 --> 00:14:25.379
a Slack app and maybe Argo CD and Argo Workflows
00:14:25.379 --> 00:14:28.879
and scripts and runbooks, maybe Ansible, whatever,
00:14:29.039 --> 00:14:32.360
dashboards, and probably in someone's memory
00:14:32.360 --> 00:14:35.259
somewhere. What's wrong with that model and how
00:14:35.259 --> 00:14:37.620
do you solve it? I don't think there is like
00:14:37.620 --> 00:14:40.059
anything particularly wrong with that model.
00:14:40.299 --> 00:14:43.779
It's just the question of like how AI can interface
00:14:43.779 --> 00:14:46.820
with that in the best way possible. So part of
00:14:46.820 --> 00:14:50.809
our hypothesis that, you know, there needs to
00:14:50.809 --> 00:14:53.269
be like a bigger context for AI to be able to
00:14:53.269 --> 00:14:55.769
understand the system. And a lot of what you
00:14:55.769 --> 00:14:59.049
mentioned ends up being, you know, what I like
00:14:59.049 --> 00:15:03.210
to call those point solutions. So there are things
00:15:03.210 --> 00:15:05.789
which are in PagerDuty. There are things within
00:15:05.789 --> 00:15:07.950
your GitHub Actions. There are some things in
00:15:07.950 --> 00:15:11.710
your repo. Some things are in CloudWatch or Datadog.
00:15:11.809 --> 00:15:14.970
So things are all over the place. And I think
00:15:14.970 --> 00:15:17.889
it's a situation for the vast majority of the
00:15:17.889 --> 00:15:19.669
companies. Yes, there are companies that are
00:15:19.669 --> 00:15:22.850
fully AWS. Nothing goes out. I think in that
00:15:22.850 --> 00:15:25.600
situation, it's a little bit easier. It's about
00:15:25.600 --> 00:15:28.620
giving that context and figuring out, you know,
00:15:28.620 --> 00:15:31.200
what should go in that context so decision can
00:15:31.200 --> 00:15:33.840
be made. So I would say that's one piece of the
00:15:33.840 --> 00:15:37.840
puzzle. There is also a question of coordinating
00:15:37.840 --> 00:15:43.019
and how AI can coordinate those things and, you
00:15:43.019 --> 00:15:45.419
know, stitch them together. And when we get to
00:15:45.419 --> 00:15:47.779
that question, we have to get into that question
00:15:47.779 --> 00:15:50.779
mark of which API actions can AI call and which
00:15:50.779 --> 00:15:54.690
API actions AI should not call. That can be solved
00:15:54.690 --> 00:15:57.590
through, you know, policies, you know, various
00:15:57.590 --> 00:16:00.750
guardrails, approvals and all of that. But as
00:16:00.750 --> 00:16:03.889
our industry, it's not science. It's not art.
00:16:03.950 --> 00:16:06.870
It's between art and science. And we have been
00:16:06.870 --> 00:16:08.789
running in loops in our industry for, you know,
00:16:08.809 --> 00:16:11.669
many times and, you know, creating things over
00:16:11.669 --> 00:16:13.710
and over, like every decade where we create things.
00:16:14.269 --> 00:16:17.169
So I would say... We are now at a stage where
00:16:17.169 --> 00:16:19.570
everything is getting recreated. Each service
00:16:19.570 --> 00:16:22.009
is going to be recreated from scratch to maybe
00:16:22.009 --> 00:16:25.490
suit the agents a little bit better. And it's
00:16:25.490 --> 00:16:28.129
about, I would say, scale and velocity. So historically,
00:16:28.129 --> 00:16:30.429
everything was made for humans. I mean, each
00:16:30.429 --> 00:16:32.909
service that you can imagine, it's made for humans,
00:16:33.110 --> 00:16:35.570
including programming languages and so on. So
00:16:35.570 --> 00:16:38.049
now the more the conversation is, okay, all of
00:16:38.049 --> 00:16:40.809
those services need to speak with AI. Some need
00:16:40.809 --> 00:16:43.090
to be rebuilt from the ground up. It's great
00:16:43.090 --> 00:16:45.009
for the new generation of builders. Everything
00:16:45.009 --> 00:16:47.090
that my father and grandfather built, let me
00:16:47.090 --> 00:16:49.830
rebuild it from scratch. Some for very good reasons,
00:16:49.889 --> 00:16:53.350
some for not. And then in that rebuilding, the
00:16:53.350 --> 00:16:55.409
question is, you know, how do you treat actually
00:16:55.409 --> 00:16:58.690
human? How do you treat agent? Are they on the
00:16:58.690 --> 00:17:01.610
same level? Are they somehow different? So it's
00:17:01.610 --> 00:17:04.859
about, I would say, becoming the reality in every
00:17:04.859 --> 00:17:08.859
part of the system. How is SuperPlane different
00:17:08.859 --> 00:17:13.700
from Argo Workflows or n8n? I mean, our main
00:17:13.700 --> 00:17:17.160
hypothesis, let's say, the space that we operate
00:17:17.160 --> 00:17:22.339
in is essentially software engineering operations,
00:17:22.859 --> 00:17:26.839
both of those domains. And we specifically want
00:17:26.839 --> 00:17:29.400
to treat, we want to start with that, that both
00:17:29.400 --> 00:17:32.700
agents and humans should be kind of on the same
00:17:33.019 --> 00:17:35.920
on the same plane. They're not today. It will
00:17:35.920 --> 00:17:39.180
take time until that becomes the reality. But
00:17:39.180 --> 00:17:42.160
where we are building service, we are building
00:17:42.160 --> 00:17:44.259
it from the ground up from that perspective.
00:17:44.559 --> 00:17:48.960
The other element is how do we give AI right
00:17:48.960 --> 00:17:51.720
access to the production, which is a big unsolved
00:17:51.720 -->
00:17:54.839
question within the industry. And our solution
00:17:54.839 --> 00:17:58.819
here is that we need more automation. Automation
00:17:58.819 --> 00:18:01.680
in that very, I would say, traditional sense.
00:18:02.319 --> 00:18:04.579
But we need to expose that in a very controlled
00:18:04.579 --> 00:18:09.039
way. So I like to think about that story as we
00:18:09.039 --> 00:18:11.000
are onboarding, let's say, the DevOps engineer within
00:18:11.000 --> 00:18:13.640
the company, or we are onboarding the agent.
00:18:14.279 --> 00:18:17.619
So what are the things that are available to
00:18:17.619 --> 00:18:21.319
me? Where can I apply my Terraform change? What
00:18:21.319 --> 00:18:24.640
can I do? What can I introspect? So I like to
00:18:24.640 --> 00:18:27.279
think about the agent as something that has the
00:18:27.279 --> 00:18:30.130
identity. It's entering the system. Okay, let
00:18:30.130 --> 00:18:32.829
me do the introspection. Let me see what are
00:18:32.829 --> 00:18:34.990
all the things that are available to me so I
00:18:34.990 --> 00:18:38.950
can be productive. And that onboarding piece
00:18:38.950 --> 00:18:43.529
is an angle how we are telling our story. In
00:18:43.529 --> 00:18:45.710
the same way that we want to make super easy
00:18:45.710 --> 00:18:49.490
for DevOps human to onboard the production system
00:18:49.490 --> 00:18:51.569
as we understand everything, and that should
00:18:51.569 --> 00:18:55.529
be there is no half-rotten documentation, tribal
00:18:55.529 --> 00:18:59.829
knowledge, all of that. It's all codified. It
00:18:59.829 --> 00:19:02.450
works, flows. I can click on everything. I can
00:19:02.450 --> 00:19:04.829
understand how things are flowing, how things
00:19:04.829 --> 00:19:08.329
are being done. And that hit by that bus factor
00:19:08.329 --> 00:19:12.170
is removed. So it's the source of truth is, you
00:19:12.170 --> 00:19:14.569
know, there. Once we reach that level, and that's
00:19:14.569 --> 00:19:17.029
going to take time for the industry to, you know,
00:19:17.069 --> 00:19:19.829
just introduce more automation. I would say at
00:19:19.829 --> 00:19:23.029
that point in time, it would be easier to onboard
00:19:23.029 --> 00:19:26.609
also like a dev, you know, DevOps AI engineer.
00:19:27.390 --> 00:19:30.589
like in a way to be able to help with things.
00:19:30.789 --> 00:19:33.869
And I generally don't go for that, you know,
00:19:33.890 --> 00:19:36.710
doomer scenario where, you know, us engineers
00:19:36.710 --> 00:19:38.869
and, you know, crafters are going to be removed
00:19:38.869 --> 00:19:40.930
from the equation. I mean, it's obvious that
00:19:40.930 --> 00:19:43.710
things are going to change, but, you know, taste,
00:19:43.849 --> 00:19:48.130
direction, you know, all of that is going to
00:19:48.130 --> 00:19:51.049
be present. So I'm more in that scenario that
00:19:51.049 --> 00:19:54.710
things that we have not historically, we didn't
00:19:54.710 --> 00:19:57.069
have time to solve. Because it was too expensive.
00:19:57.170 --> 00:19:59.789
It was too expensive to write software. So I'm
00:19:59.789 --> 00:20:02.309
more in that category that, you know, we will
00:20:02.309 --> 00:20:05.470
need more engineers on all sides because there
00:20:05.470 --> 00:20:08.490
will be just more software. Yes, the roles will
00:20:08.490 --> 00:20:11.349
change and some will change drastically. And
00:20:11.349 --> 00:20:14.849
we'll have to figure out how that works. So when
00:20:14.849 --> 00:20:18.089
you say more AI and onboarding, are you talking
00:20:18.089 --> 00:20:20.670
about like a multi-agentic workflow? Like where
00:20:20.670 --> 00:20:23.309
maybe I have like an agent for just Terraform
00:20:23.309 --> 00:20:26.410
or just IaC in general? And like that agent only
00:20:26.410 --> 00:20:29.150
handles that? Or is it even more fractured than
00:20:29.150 --> 00:20:31.470
that? Like an agent that only handles like creation
00:20:31.470 --> 00:20:34.509
of objects or resources? And how do you think
00:20:34.509 --> 00:20:36.230
about that model? Like, how do you frame that?
00:20:36.569 --> 00:20:40.349
My hypothesis is that it's not going to be like
00:20:40.349 --> 00:20:44.910
that. It's going to be within this team that
00:20:44.910 --> 00:20:49.259
is handling maybe... encoding and stream in Riverside.
00:20:49.259 --> 00:20:50.900
I don't know how they're organized, but let's
00:20:50.900 --> 00:20:52.859
say they have a department which is more for
00:20:52.859 --> 00:20:55.180
this, you know, video transformation, production,
00:20:55.180 --> 00:20:58.720
whatever team, there they will have an agent who
00:20:58.720 --> 00:21:01.599
will have a specialized set of skills and understanding.
00:21:01.599 --> 00:21:04.180
So it will know that there is like a throughput
00:21:04.180 --> 00:21:07.750
and latency are very important. So maybe I'm not,
00:21:07.750 --> 00:21:09.990
I don't even write, you know, Terraform. I don't
00:21:09.990 --> 00:21:12.710
do this. I have some specialized functions for
00:21:12.710 --> 00:21:16.190
my team and the context is very based on that.
00:21:16.309 --> 00:21:18.910
And I'll have that maybe one monitoring agent,
00:21:19.029 --> 00:21:21.390
which is really in charge of, you know, maybe
00:21:21.390 --> 00:21:25.029
scaling and latency, understanding what's happening
00:21:25.029 --> 00:21:28.049
in the world, you know, within the AWS network.
00:21:28.170 --> 00:21:30.829
And there is a glitch and then what I can do.
00:21:32.049 --> 00:21:34.930
My sense is we're going to be very specialized
00:21:34.930 --> 00:21:38.490
and small and scoped. And that agent is going
00:21:38.490 --> 00:21:42.190
to take care of that. Of course, it does make
00:21:42.190 --> 00:21:44.089
sense that there are agents and there will be
00:21:44.089 --> 00:21:46.990
agents that will be very, very specialized in
00:21:46.990 --> 00:21:50.309
Terraform, OpenTofu, whatever, resource creation.
00:21:50.970 --> 00:21:53.990
But I'm more on this side. How do we fit? How
00:21:53.990 --> 00:21:56.730
do we find help? How do we find that small agent?
00:21:56.829 --> 00:21:59.470
Not replacing anyone, helping someone with a
00:21:59.470 --> 00:22:02.369
very specific task, which is maybe more suited
00:22:02.369 --> 00:22:04.349
for AI because there are a lot of numbers and
00:22:04.349 --> 00:22:07.329
metrics and less for human. Yeah. Do you think
00:22:07.329 --> 00:22:09.349
that compliance and audit requirements change
00:22:09.349 --> 00:22:12.630
when part of the workflow came from an AI agent?
00:22:12.890 --> 00:22:15.089
I think that they are going to change. I think
00:22:15.089 --> 00:22:17.230
the change is going to come by this, you know,
00:22:17.269 --> 00:22:19.509
velocity, which is coming in. So whenever I think
00:22:19.509 --> 00:22:22.150
about this, you know, code creation in AI, I
00:22:22.150 --> 00:22:25.490
imagine like a dam and a river. And like river
00:22:25.490 --> 00:22:27.930
is so strong and it's, you know, overflowing
00:22:27.930 --> 00:22:30.369
the dam and, you know, you have to do something
00:22:30.369 --> 00:22:34.049
different. So I think that, you know, compliance
00:22:34.049 --> 00:22:37.869
piece will have to change and we have to figure
00:22:37.869 --> 00:22:41.730
out how that will work. At some point, agent
00:22:41.730 --> 00:22:44.269
is probably going to have to carry, you know,
00:22:44.309 --> 00:22:47.529
some element of responsibility in some shape
00:22:47.529 --> 00:22:50.130
and form. No clue how is that going to work.
00:22:50.230 --> 00:22:53.490
I would say... Maybe this, you know, self-driving
00:22:53.490 --> 00:22:58.809
cars industry will maybe, you know, teach us
00:22:58.809 --> 00:23:02.450
something. So if I get killed on the streets
00:23:02.450 --> 00:23:05.210
of San Francisco by Waymo, who is responsible?
00:23:06.069 --> 00:23:08.170
I know that they have some framework. They figure
00:23:08.170 --> 00:23:11.529
it out, you know, one way or another. So if like
00:23:11.529 --> 00:23:13.970
a database, you know, a database table or, you
00:23:13.970 --> 00:23:15.789
know, whatever gets dropped or, you know, some
00:23:15.789 --> 00:23:18.549
data corrupted or leak or whatever, whatever
00:23:18.549 --> 00:23:21.630
other way. There is going to be some legal framework
00:23:21.630 --> 00:23:25.650
around that. I wish I have an answer, so I don't.
00:23:25.650 --> 00:23:29.690
I'm just, you know, thinking out loud. I think
00:23:29.690 --> 00:23:38.690
it touches the law and those areas also in various
00:23:38.690 --> 00:23:42.670
ways. What's the framework there? If something
00:23:42.670 --> 00:23:44.730
really bad happens in a publicly traded company,
00:23:44.930 --> 00:23:48.210
they have various policies in place. I would
00:23:48.210 --> 00:23:50.430
say that's the domain where probably the biggest
00:23:50.430 --> 00:23:52.910
guys should solve it and let us know what to
00:23:52.910 --> 00:23:55.789
do. So I would say, you know, Google, Apple,
00:23:55.890 --> 00:23:59.269
whoever has the most AI, the biggest production,
00:23:59.529 --> 00:24:02.210
all of that, and the rest of the industry would
00:24:02.210 --> 00:24:07.930
follow us. So I would say it's not my job and
00:24:07.960 --> 00:24:11.000
of many other folks. But it is an interesting
00:24:11.000 --> 00:24:13.319
topic for sure. And one thing that feels very
00:24:13.319 --> 00:24:15.940
real is how much production knowledge lives in
00:24:15.940 --> 00:24:20.119
people's heads. How should teams capture operational
00:24:20.119 --> 00:24:22.299
memory without pretending that documentation
00:24:22.299 --> 00:24:25.420
will stay updated? How do we solve that problem?
00:24:25.799 --> 00:24:30.299
Again, getting back to our hypothesis at SuperPlane
00:24:30.299 --> 00:24:33.420
that we live by, I think it's about automation.
00:24:33.839 --> 00:24:37.039
So as things get more and more done through the
00:24:37.039 --> 00:24:43.029
automation, the more records do we have about
00:24:43.029 --> 00:24:46.690
what happened and when and why and whose approval
00:24:46.690 --> 00:24:51.230
and whose idea and which prompt and so on. So
00:24:51.230 --> 00:24:55.069
all that's tied into a chain of a decision chain,
00:24:55.170 --> 00:25:00.170
how we came to that. So that's one piece of the
00:25:00.170 --> 00:25:05.250
puzzle for sure. How in practice that gets pulled
00:25:05.250 --> 00:25:08.890
off, we'll see. I mean, in terms of documentation,
00:25:09.109 --> 00:25:12.720
I think that all documentation should be, kind
00:25:12.720 --> 00:25:16.019
of AI-generated by those by those automations
00:25:16.019 --> 00:25:19.940
that we have, you know, in place, again, iteratively
00:25:19.940 --> 00:25:23.240
getting that piece by piece. But I think that's
00:25:23.240 --> 00:25:24.839
the main piece. That's the main piece of the
00:25:24.839 --> 00:25:28.500
puzzle of that. Yes, AI definitely is better at
00:25:28.500 --> 00:25:31.900
maintaining documentation over practitioners,
00:25:32.119 --> 00:25:34.940
just given that, you know, practitioners can
00:25:34.940 --> 00:25:37.019
get lazy, right with comments, or just markdown
00:25:37.019 --> 00:25:40.640
files and AI will be much more verbose, but not
00:25:40.640 --> 00:25:43.880
overly verbose, like concise in its verboseness.
00:25:44.079 --> 00:25:47.039
I agree with that. It's also about generating
00:25:47.039 --> 00:25:50.380
the documentation. So I think it will be each
00:25:50.380 --> 00:25:53.279
paragraph sentence, I think should be clickable
00:25:53.279 --> 00:25:56.099
in that documentation with like some decision
00:25:56.099 --> 00:25:58.200
tree. How do you came up with that conclusion?
00:25:58.839 --> 00:26:02.259
Because if I perform some action based on like
00:26:02.259 --> 00:26:04.700
incorrect
documentation, I do something bad.
00:26:05.119 --> 00:26:07.299
I need to have that chain. Okay, you inform.
00:26:08.180 --> 00:26:10.900
You gave me something, you know, some misinformation
00:26:10.900 --> 00:26:13.740
based on some, you know, wrong reasoning. And
00:26:13.740 --> 00:26:15.599
then we should be able to have like root cause
00:26:15.599 --> 00:26:19.460
analysis of what caused that. So in my head,
00:26:19.660 --> 00:26:22.019
I mean, we are not there yet, but we have some
00:26:22.019 --> 00:26:24.920
elements in our documentation generation where,
00:26:25.029 --> 00:26:27.329
you know, you have a sentence or you have a certain
00:26:27.329 --> 00:26:30.410
action and you click on it and it links back
00:26:30.410 --> 00:26:34.000
to some part of the workflow and you see, okay,
00:26:34.519 --> 00:26:37.019
how that, you know, documentation came
00:26:37.019 --> 00:26:40.559
came about. So, and I keep going and kind of maybe
00:26:40.559 --> 00:26:43.119
defending not knowing the answer and saying about
00:26:43.119 --> 00:26:45.960
that has to be done, you know, iteratively. So the
00:26:45.960 --> 00:26:49.720
whole documentation generation problem. I think
00:26:49.720 --> 00:26:52.140
we'll have to, we'll have to get into that check
00:26:52.140 --> 00:26:54.680
that we should know how each line of that documentation
00:26:54.680 --> 00:26:57.839
was generated and, you know, it traced back to some
00:26:57.839 --> 00:27:00.880
to some code or some fact some configuration
00:27:00.880 --> 00:27:04.519
somewhere And then how do we keep that in sync?
00:27:04.519 --> 00:27:08.259
Is that eventual consistency or whenever I change
00:27:08.259 --> 00:27:10.440
some configuration, my load balancer, is that
00:27:10.440 --> 00:27:13.519
immediately reflected on the documentation? What's
00:27:13.519 --> 00:27:17.119
the gap, you know, between that? Again, one of
00:27:17.119 --> 00:27:20.259
the areas that deserves a special kind of attention,
00:27:20.539 --> 00:27:23.460
especially because of that, as you said, no one
00:27:23.460 --> 00:27:26.839
likes doing that. So our industry is awful, you
00:27:26.839 --> 00:27:29.440
know, when it comes to having like up-to-date
00:27:29.440 --> 00:27:32.289
documentation. Yep, for sure. Every company I've
00:27:32.289 --> 00:27:34.210
worked at, you go on Confluence and there is,
00:27:34.230 --> 00:27:35.750
you know, a document that hasn't been updated in
00:27:35.750 --> 00:27:38.250
three years and the process is not even the same
00:27:38.250 --> 00:27:40.150
anymore. You know, that's the only document that
00:27:40.150 --> 00:27:42.750
exists. Or you're trying to hunt through Slack
00:27:42.750 --> 00:27:44.930
to find, you know, a conversation that engineers
00:27:44.930 --> 00:27:47.789
have had. And that's a real problem, I think,
00:27:47.809 --> 00:27:50.009
yeah, that is just not solved at a lot of companies.
00:27:50.390 --> 00:27:53.890
And AI definitely can help with that 100%. Okay,
00:27:54.009 --> 00:27:57.549
so wrapping up, if a platform or SRE team wants
00:27:57.549 --> 00:28:00.089
to start using AI in infrastructure operations,
00:28:00.940 --> 00:28:03.119
What's the first workflow that you would trust
00:28:03.119 --> 00:28:05.180
it with? Like, where would you start? I mean,
00:28:05.200 --> 00:28:07.079
it's definitely this, you know, read-only mode
00:28:07.079 --> 00:28:11.559
where I would want it to explain me how something
00:28:11.559 --> 00:28:15.200
is done and point me in the right direction.
00:28:15.420 --> 00:28:18.839
So if I'm onboarding like a brownfield project,
00:28:19.079 --> 00:28:23.259
let's say on AWS, it has some complexity in terms
00:28:23.259 --> 00:28:25.900
of the, you know, software stack, infrastructure
00:28:25.900 --> 00:28:28.819
stack, all of that. I'm generally, what I really
00:28:28.819 --> 00:28:31.670
like is that, you know, I pretend to be super,
00:28:31.730 --> 00:28:34.170
super stupid, you know, and just a lot of
00:28:34.170 --> 00:28:37.029
dumb questions and like onboard me, help me understand
00:28:37.029 --> 00:28:39.849
how system works. And then I ask, okay, what's
00:28:39.849 --> 00:28:43.849
the smallest possible change that I can do on
00:28:43.849 --> 00:28:46.329
the system to just help me get, you know, some
00:28:46.329 --> 00:28:52.089
deeper understanding. I think that I could be,
00:28:52.190 --> 00:28:54.569
I could misinterpreted your question. So feel
00:28:54.569 --> 00:28:57.769
free to, you know, dig deeper. But I mean, for
00:28:57.769 --> 00:29:00.190
me personally, that's, you know, discovery phase
00:29:00.190 --> 00:29:03.390
is like what I would do as a first step. Yeah.
00:29:03.390 --> 00:29:06.809
Okay, so let's say we've done discovery phase.
00:29:06.809 --> 00:29:10.289
We've done read only. It has the context of our
00:29:10.289 --> 00:29:14.769
entire IaC. It knows our deployment stack. It has
00:29:14.769 --> 00:29:17.130
access to GitHub, so it knows all of our GitHub
00:29:17.130 --> 00:29:20.390
Actions that we use for our CI/CD platform. It
00:29:20.390 --> 00:29:22.670
maybe even has access to Grafana and Prometheus
00:29:22.670 --> 00:29:25.970
endpoints, so it's getting monitoring. All read only.
00:29:26.319 --> 00:29:28.420
What would be the first write operation? Where
00:29:28.420 --> 00:29:31.240
would you feel the safest starting giving that
00:29:31.240 --> 00:29:34.200
autonomy to an AI agent to start writing? I mean,
00:29:34.240 --> 00:29:37.059
I would say we have two big categories here.
00:29:37.180 --> 00:29:40.559
One is the creation of, you know, just the change,
00:29:40.880 --> 00:29:44.599
which is, again, writing some code. So it's either
00:29:44.599 --> 00:29:47.779
Terraform or, you know, Java code within our
00:29:47.779 --> 00:29:50.960
application. That's one domain that goes through
00:29:50.960 --> 00:29:53.599
a review process. And then that's, you know,
00:29:53.619 -->
00:29:57.730
being applied in production. I would say that's
00:29:57.730 --> 00:30:00.089
the category where you have human in the loop
00:30:00.089 --> 00:30:04.049
approval process, which is lowering the risk
00:30:04.049 --> 00:30:06.710
there to some extent. And that's what happened
00:30:06.710 --> 00:30:13.450
in the industry at scale already. So, I mean,
00:30:13.470 --> 00:30:16.349
I haven't said anything new. I just kind of described
00:30:16.349 --> 00:30:20.190
the reality that we have today. I would say maybe
00:30:20.190 --> 00:30:23.420
the first next step after that are... I would
00:30:23.420 --> 00:30:26.400
say in that category of these loops that we keep
00:30:26.400 --> 00:30:29.900
mentioning. So if we want to move some metric,
00:30:30.000 --> 00:30:34.180
you know, want to lower the latency of a certain
00:30:34.180 --> 00:30:37.420
system, then I would say I would work on those
00:30:37.420 --> 00:30:40.740
numerical things that are like easy to measure
00:30:40.740 --> 00:30:43.960
and easy to test. So yes, we have some response
00:30:43.960 --> 00:30:46.160
time somewhere. It can be due to the way the
00:30:46.160 --> 00:30:48.299
code is written or how infrastructure is configured.
00:30:48.940 --> 00:30:51.839
And maybe we want to have like a faster disk,
00:30:52.079 --> 00:30:55.559
faster database, whatever. And I would say it's
00:30:55.559 --> 00:30:59.519
in that performance optimization area. What are
00:30:59.519 --> 00:31:02.640
some of the incremental steps that we can make?
00:31:02.759 --> 00:31:05.099
And then we apply that to copy of production
00:31:05.099 --> 00:31:08.299
or staging environment, whatever. Do some testing
00:31:08.299 --> 00:31:12.519
and apply that change in production. In the same
00:31:12.519 --> 00:31:15.180
way, I would map it to how can I improve the
00:31:15.180 --> 00:31:17.859
speed of my GitHub action, which is, you know.
00:31:18.200 --> 00:31:21.099
running my code and I would, it is also how can
00:31:21.099 --> 00:31:24.500
I lower maybe the number of CVEs like in my system,
00:31:24.619 --> 00:31:26.619
there are auto security tools that are minimizing
00:31:26.619 --> 00:31:29.960
that. So I would say I would keep it in that
00:31:29.960 --> 00:31:32.720
realm where we can easily measure the impact
00:31:32.720 --> 00:31:37.309
and it's a numerical value. And we, we, we, we
00:31:37.309 --> 00:31:40.309
go, we go from there. Yeah. Yeah. It makes sense.
00:31:40.750 --> 00:31:42.569
Any closing thoughts you'd like to leave our
00:31:42.569 --> 00:31:45.829
audience with? I would say I would get back to
00:31:45.829 --> 00:31:49.750
the automation and I would say we as an industry
00:31:49.750 --> 00:31:52.490
and discipline, hardware and software together,
00:31:52.769 --> 00:31:56.009
we have been going out and changing factories
00:31:56.009 --> 00:31:59.069
and many, many other industries for many decades
00:31:59.069 --> 00:32:01.650
and automating things, automating
things, you
00:32:01.650 --> 00:32:04.390
know, factories, removing workers, like all those
00:32:04.390 --> 00:32:07.809
kinds of bad things. Now the lens has kind of,
00:32:07.809 --> 00:32:10.750
you know, shifted in the other direction and
00:32:10.750 --> 00:32:14.250
we are getting automated and, you know, also
00:32:14.250 --> 00:32:16.430
feel threatened a little bit and all those emotions
00:32:16.430 --> 00:32:19.609
that we're feeling. I think that's okay and everything
00:32:19.609 --> 00:32:22.450
is going to be okay. We just have to embrace
00:32:22.450 --> 00:32:26.829
that automation and go after it as soon as possible
00:32:26.829 --> 00:32:29.450
and just embrace the change that are happening
00:32:29.450 --> 00:32:35.369
to our roles as that is happening. So with any
00:32:35.369 --> 00:32:38.390
technology, historically, you can fight it or
00:32:38.390 --> 00:32:40.730
you can embrace it. So I would say let's just
00:32:40.730 --> 00:32:43.470
embrace it. You know, bring that, bring that
00:32:43.470 --> 00:32:45.829
automation and, you know, figure out there is
00:32:45.829 --> 00:32:47.750
going to be a lot of, you know, a lot of infrastructure,
00:32:48.130 --> 00:32:51.750
a lot of everything. And yeah, those are my thoughts.
00:32:51.829 --> 00:32:54.609
So more automation. More automation. No, I got
00:32:54.609 --> 00:32:57.750
you. Where can people follow you? SuperPlane,
00:32:57.910 --> 00:33:01.250
Semaphore? Yeah, superplane.com. That's, that's
00:33:01.250 --> 00:33:04.599
our thing that we, that we are doing and trying
00:33:04.599 --> 00:33:06.339
to solve that, you know, write access to production.
00:33:07.250 --> 00:33:09.630
Easy problem. Easy problem to solve. But yeah,
00:33:09.690 --> 00:33:11.769
we are doing it iteratively in that, you know,
00:33:11.829 --> 00:33:14.809
a lot of determinism in place. So we want to
00:33:14.809 --> 00:33:16.990
be cautious and, you know, do good things there.
00:33:17.789 --> 00:33:22.089
And semaphore.io, previous company in the CI/CD
00:33:22.089 --> 00:33:26.990
domain and DarkoFabijan at X. Yeah. So feel free
00:33:26.990 --> 00:33:29.950
to engage. A lot of, you know, hard topics to
00:33:29.950 --> 00:33:32.589
discuss. We love them. And I'll put links for
00:33:32.589 --> 00:33:34.920
all those in the show notes as well. Darko, thank
00:33:34.920 --> 00:33:36.460
you so much for coming on. Really, really appreciate
00:33:36.460 --> 00:33:40.200
it. That was my conversation with Darko Fabijan
00:33:40.200 --> 00:33:43.180
from SuperPlane. The biggest takeaway for me
00:33:43.180 --> 00:33:46.960
is that AI didn't break software delivery. It
00:33:46.960 --> 00:33:50.619
exposed how much of it still runs at human speed.
00:33:51.099 --> 00:33:55.839
Darko's dam and river image nailed it. AI-written
00:33:55.839 --> 00:34:00.019
code is the river and the dam is every review
00:34:00.019 --> 00:34:03.980
and approval we built for people. I also liked
00:34:03.980 --> 00:34:08.280
his take on rubber stamping. He pointed out humans
00:34:08.280 --> 00:34:12.039
hallucinate too. And sooner or later, a tired
00:34:12.039 --> 00:34:15.579
reviewer approves the line that drops all the
00:34:15.579 --> 00:34:20.239
tables. So he wants a risk score on every change.
00:34:20.699 --> 00:34:24.559
Renaming a bucket scores low. Deleting an S3
00:34:24.559 --> 00:34:28.559
bucket or an RDS instance is a 10 out of 10.
00:34:28.780 --> 00:34:32.119
He called that a billion dollar startup idea.
00:34:32.579 --> 00:34:36.300
and I'd buy it. His agents as new hires idea
00:34:36.300 --> 00:34:40.139
stuck with me too. Give the agent an identity,
00:34:40.380 --> 00:34:43.440
let it look around, and keep a deterministic
00:34:43.440 --> 00:34:46.980
layer between it and production. The work that
00:34:46.980 --> 00:34:50.539
makes an agent easy to onboard helps your humans
00:34:50.539 --> 00:34:54.500
too. No half-rotten docs, no tribal knowledge,
00:34:54.760 --> 00:34:59.159
no bus factor. Where to start? Darko says read
00:34:59.159 --> 00:35:02.840
only. He likes to pretend that he's super stupid,
00:35:03.000 --> 00:35:06.400
ask a lot of dumb questions, and say, onboard
00:35:06.400 --> 00:35:09.639
me. When you do let it write, pick something
00:35:09.639 --> 00:35:12.659
you can measure with a number, like latency.
00:35:13.019 --> 00:35:17.199
You can find Darko on X, plus SuperPlane and
00:35:17.199 --> 00:35:20.260
his previous company Semaphore through the links
00:35:20.260 --> 00:35:23.659
in the show notes. Follow or subscribe to Ship
00:35:23.659 --> 00:35:26.619
It Weekly wherever you listen and find previous
00:35:26.619 --> 00:35:30.619
episodes at shipitweekly.fm. Thanks for listening,
00:35:30.699 --> 00:35:32.860
and I'll see you Friday for the news.