逐字稿 🔗
WEBVTT
00:00:00.000 --> 00:00:02.222
Some people have been able to get, like,
00:00:02.222 --> 00:00:03.735
4x acceleration, and some people
00:00:03.735 --> 00:00:05.200
are barely seeing 5%, 10%, 20%.
00:00:05.200 --> 00:00:05.748
And even then,
00:00:05.748 --> 00:00:07.000
it doesn't feel like it's clear.
00:00:07.000 --> 00:00:10.000
The top 1 percentile of engineers,
00:00:10.000 --> 00:00:11.648
they were about 15 times
00:00:11.648 --> 00:00:13.846
more productive than the median.
00:00:13.846 --> 00:00:15.338
You're able to generate
00:00:15.338 --> 00:00:16.765
the PR for what you're
00:00:16.765 --> 00:00:18.360
trying to accomplish with your agent
00:00:18.360 --> 00:00:20.000
in 15 minutes or something like that.
00:00:20.000 --> 00:00:21.951
And yet, if you haven't changed
00:00:21.951 --> 00:00:23.462
your downstream process,
00:00:23.462 --> 00:00:25.205
that same PR might sit in a queue
00:00:25.205 --> 00:00:27.000
waiting for review for three days.
00:00:27.000 --> 00:00:29.059
The ones that are really making
00:00:29.059 --> 00:00:31.250
a difference are where they don't
00:00:31.250 --> 00:00:33.321
treat AI as a tool. They treat
00:00:33.321 --> 00:00:35.600
AI like a delivery system change.
00:00:49.000 --> 00:00:52.178
Yo, yo, yo! This is MK, back with another
00:00:52.178 --> 00:00:55.357
episode of Ready, Set, Runtime, a podcast
00:00:55.357 --> 00:00:57.588
from AHEAD. Today's a special episode
00:00:57.588 --> 00:01:00.000
because Priyanka Karan is going to host,
00:01:00.000 --> 00:01:01.897
and I'm gonna be participating
00:01:01.897 --> 00:01:03.667
as a guest with our partners
00:01:03.667 --> 00:01:04.795
from SpaceX Super Intelligence (SpaceXSI)
00:01:04.795 --> 00:01:05.125
and Harness.
00:01:05.125 --> 00:01:06.125
Pri, do you wanna kick us off?
00:01:06.125 --> 00:01:09.007
My name is Priyanka Karan. I am another
00:01:09.007 --> 00:01:11.667
Field CTO at AHEAD, and I have spent
00:01:11.667 --> 00:01:13.780
the last two decades really thinking
00:01:13.780 --> 00:01:15.364
about software development,
00:01:15.364 --> 00:01:18.004
digital transformation, and what it means
00:01:18.004 --> 00:01:20.000
for developers and our clients.
00:01:20.000 --> 00:01:21.109
And so I'm really excited
00:01:21.109 --> 00:01:22.750
to be having this conversation today.
00:01:22.750 --> 00:01:24.646
And I wanna go ahead and have David
00:01:24.646 --> 00:01:26.000
do his introduction next.
00:01:26.000 --> 00:01:29.332
Hello, everybody. My name's David Pan.
00:01:29.332 --> 00:01:31.875
I am a Field CTO at SpaceXSI,
00:01:31.875 --> 00:01:34.569
and I joined by way of the Cursor
00:01:34.569 --> 00:01:37.917
acquisition. I'm a longtime software guy.
00:01:37.917 --> 00:01:39.825
I've spent my whole career doing software
00:01:39.825 --> 00:01:41.500
engineering and software leadership,
00:01:41.500 --> 00:01:43.136
and I'm really excited to talk about
00:01:43.136 --> 00:01:45.000
where all of that is going in the future.
00:01:45.000 --> 00:01:47.000
Martin?
00:01:47.000 --> 00:01:49.085
Sure. I'm Martin Reynolds.
00:01:49.085 --> 00:01:51.250
I'm a Field CTO at Harness.
00:01:51.250 --> 00:01:52.440
I've been at Harness
00:01:52.440 --> 00:01:53.750
for about three years,
00:01:53.750 --> 00:01:55.019
but prior to that,
00:01:55.019 --> 00:01:56.923
all this gray hair actually
00:01:56.923 --> 00:01:59.128
came from 30 years of delivering
00:01:59.128 --> 00:02:01.333
software in all sorts of places.
00:02:01.333 --> 00:02:03.683
And just a side interesting fact: I was
00:02:03.683 --> 00:02:06.154
actually one of Harness's first customers
00:02:06.154 --> 00:02:09.077
worldwide—I think number four or number
00:02:09.077 --> 00:02:12.000
five—and certainly the first in Europe.
00:02:12.000 --> 00:02:14.078
I might not have worked for them for all
00:02:14.078 --> 00:02:16.000
that time, but I certainly know them.
00:02:16.000 --> 00:02:18.176
So I'm sure there are some great
00:02:18.176 --> 00:02:20.625
stories that come from that journey.
00:02:20.625 --> 00:02:22.812
Both SpaceXSI and Harness
00:02:22.812 --> 00:02:25.000
are really complementary,
00:02:25.000 --> 00:02:27.134
and from your perspective,
00:02:27.134 --> 00:02:30.500
where do you see the developer experience
00:02:30.500 --> 00:02:32.700
end and the enterprise
00:02:32.700 --> 00:02:35.000
delivery problem begin?
00:02:35.000 --> 00:02:35.806
Because it's evolving, right?
00:02:35.806 --> 00:02:36.667
It's been changing quite a bit,
00:02:36.667 --> 00:02:38.022
and we've seen a lot of that happen.
00:02:38.022 --> 00:02:39.000
So what do you guys think?
00:02:39.000 --> 00:02:42.917
Yeah, why don't I start on that one?
00:02:42.917 --> 00:02:45.609
I think one thing that's interesting
00:02:45.609 --> 00:02:48.750
about the adoption of AI tools in software
00:02:48.750 --> 00:02:50.296
engineering is that,
00:02:50.296 --> 00:02:52.692
for most of the journey so far,
00:02:52.692 --> 00:02:55.017
the tools have really been focused
00:02:55.017 --> 00:02:57.000
primarily on code generation.
00:02:57.000 --> 00:02:59.585
And that's been a big deal, right? I think
00:02:59.585 --> 00:03:01.923
that's been one of the most expensive,
00:03:01.923 --> 00:03:04.097
hardest parts of software engineering
00:03:04.097 --> 00:03:05.625
for decades at this point.
00:03:05.625 --> 00:03:08.272
But at the same time, when you really look
00:03:08.272 --> 00:03:10.667
at how enterprise software engineering
00:03:10.667 --> 00:03:12.424
is done these days—when
00:03:12.424 --> 00:03:14.333
I reflect back on my time
00:03:14.333 --> 00:03:16.438
leading these software teams and working
00:03:16.438 --> 00:03:18.333
with all these software engineers—if
00:03:18.333 --> 00:03:20.897
I had to estimate the amount of time
00:03:20.897 --> 00:03:23.462
that a typical engineer was actually
00:03:23.462 --> 00:03:26.221
just sitting heads-down writing code
00:03:26.221 --> 00:03:28.750
in their typical workday, I
Load full transcript
would
00:03:28.750 --> 00:03:31.916
guess that it's probably only two hours
00:03:31.916 --> 00:03:35.000
out of the day or something like that.
00:03:35.000 --> 00:03:37.817
And that leads to an interesting
00:03:37.817 --> 00:03:41.250
observation for the world of AI, right?
00:03:41.250 --> 00:03:42.546
Because if our AI tools
00:03:42.546 --> 00:03:44.688
are really focused on code generation,
00:03:44.688 --> 00:03:46.482
but we're only spending a pretty
00:03:46.482 --> 00:03:48.333
small portion of our day on that,
00:03:48.333 --> 00:03:50.352
then it means you can really only
00:03:50.352 --> 00:03:51.882
get so much acceleration,
00:03:51.882 --> 00:03:54.000
even if AI did all of that work, right?
00:03:54.000 --> 00:03:56.377
And I think for enterprise delivery,
00:03:56.377 --> 00:03:57.500
as you mentioned,
00:03:57.500 --> 00:04:00.335
Priyanka, a lot of these companies
00:04:00.335 --> 00:04:03.421
really just care about the end-to-end
00:04:03.421 --> 00:04:05.188
speed of being able to deliver software,
00:04:05.188 --> 00:04:07.000
and there's so much more to it than that.
00:04:07.000 --> 00:04:08.000
Yeah.
00:04:08.000 --> 00:04:09.889
Yeah. It's interesting. I do agree
00:04:09.889 --> 00:04:12.000
about the amount of time that's spent.
00:04:12.000 --> 00:04:14.750
I would argue that the developer
00:04:14.750 --> 00:04:17.500
experience doesn't end where the
00:04:17.500 --> 00:04:19.934
enterprise delivery starts, because it's
00:04:19.934 --> 00:04:22.308
still part of the developer experience.
00:04:22.308 --> 00:04:25.009
All the stuff inside the editor—the
00:04:25.009 --> 00:04:27.941
goal is to get the best flow possible.
00:04:27.941 --> 00:04:30.124
They wanna be able to generate
00:04:30.124 --> 00:04:32.308
or work on and write that code
00:04:32.308 --> 00:04:35.238
in as uninterrupted a way as possible.
00:04:35.238 --> 00:04:36.809
But then they're spending so much
00:04:36.809 --> 00:04:38.333
time trying to get that through.
00:04:38.333 --> 00:04:40.455
Once you get to that pull request,
00:04:40.455 --> 00:04:42.576
you're changing the audience
00:04:42.576 --> 00:04:44.545
to not just the developer.
00:04:44.545 --> 00:04:46.150
It becomes, I don't know,
00:04:46.150 --> 00:04:48.077
the platform engineering team,
00:04:48.077 --> 00:04:49.538
security, audits, and they want evidence
00:04:49.538 --> 00:04:51.000
that all the right things are happening.
00:04:51.000 --> 00:04:54.070
But you want that to be visible, easy
00:04:54.070 --> 00:04:57.222
to consume, and not too time-consuming
00:04:57.222 --> 00:05:00.079
for the engineer. And I think that's
00:05:00.079 --> 00:05:02.857
where the world is changing, right?
00:05:02.857 --> 00:05:04.799
That kind of visibility, to say,
00:05:04.799 --> 00:05:07.105
"Hey, this is what failed in my build.
00:05:07.105 --> 00:05:08.608
I wanna see that in my IDE,
00:05:08.608 --> 00:05:10.667
where I just created my pull request.
00:05:10.667 --> 00:05:12.122
I wanna be able to see if there
00:05:12.122 --> 00:05:14.000
are security fixes that need to be done.
00:05:14.000 --> 00:05:16.949
I wanna know that it's made it all the way
00:05:16.949 --> 00:05:19.688
to pre-production and it's just pending
00:05:19.688 --> 00:05:22.184
a final human approval at that point."
00:05:22.184 --> 00:05:24.615
Having that visibility and being able
00:05:24.615 --> 00:05:27.724
to react while still in that same flow is
00:05:27.724 --> 00:05:30.833
what really adds that acceleration piece.
00:05:30.833 --> 00:05:32.708
And that's why I say developer
00:05:32.708 --> 00:05:34.583
experience doesn't really end;
00:05:34.583 --> 00:05:36.000
it's always gone all the way across.
00:05:36.000 --> 00:05:38.308
We've just focused really hard
00:05:38.308 --> 00:05:40.000
on the left-hand side,
00:05:40.000 --> 00:05:41.372
and now we wanna bring
00:05:41.372 --> 00:05:43.056
more of the right to the...
00:05:43.056 --> 00:05:45.216
I hate using the phrase, but shifting
00:05:45.216 --> 00:05:47.667
it left. But the visibility, not the work.
00:05:47.667 --> 00:05:49.000
That is the trick.
00:05:49.000 --> 00:05:51.737
You talked a little bit about visibility,
00:05:51.737 --> 00:05:54.474
and I think where we are now is that most
00:05:54.474 --> 00:05:55.156
teams are really past
00:05:55.156 --> 00:05:56.000
the adoption piece, right?
00:05:56.000 --> 00:05:58.116
This is becoming the new standard,
00:05:58.116 --> 00:06:00.294
or the new expected way of working,
00:06:00.294 --> 00:06:02.127
and these lines that you mentioned
00:06:02.127 --> 00:06:03.529
are blurring a little bit.
00:06:03.529 --> 00:06:05.014
But one of the things
00:06:05.014 --> 00:06:06.500
is that organizations
00:06:06.500 --> 00:06:08.516
are not necessarily seeing
00:06:08.516 --> 00:06:10.455
the same outcomes, right?
00:06:10.455 --> 00:06:13.297
From your perspective, where do you think
00:06:13.297 --> 00:06:16.000
organizations are getting good results?
00:06:16.000 --> 00:06:17.749
What are they doing differently
00:06:17.749 --> 00:06:19.667
from the organizations that aren't
00:06:19.667 --> 00:06:22.175
seeing those benefits come in? Some people
00:06:22.175 --> 00:06:24.444
have been able to get 4x acceleration,
00:06:24.444 --> 00:06:27.059
and some people are barely seeing 5%, 10%,
00:06:27.059 --> 00:06:28.682
20%, and even then it doesn't
00:06:28.682 --> 00:06:30.250
feel like it's clear, right?
00:06:30.250 --> 00:06:32.000
So where do you guys see that?
00:06:32.000 --> 00:06:33.451
Yeah, I absolutely
00:06:33.451 --> 00:06:35.385
agree with that premise.
00:06:35.385 --> 00:06:36.592
And I would add,
00:06:36.592 --> 00:06:39.231
I think it's not just organizations
00:06:39.231 --> 00:06:41.282
that are seeing different results;
00:06:41.282 --> 00:06:43.333
even within a single organization,
00:06:43.333 --> 00:06:45.555
you've got different teams and different
00:06:45.555 --> 00:06:47.500
engineers that see wildly different
00:06:47.500 --> 00:06:49.409
results in terms of the productivity
00:06:49.409 --> 00:06:51.000
they're able to get out of AI.
00:06:51.000 --> 00:06:52.911
Earlier this year—I think
00:06:52.911 --> 00:06:55.357
it was in the spring—our company
00:06:55.357 --> 00:06:57.766
did a bunch of data analysis across
00:06:57.766 --> 00:07:00.312
all the users on the Cursor platform,
00:07:00.312 --> 00:07:02.656
in an aggregated way, to see what those
00:07:02.656 --> 00:07:05.000
results actually look like in the data.
00:07:05.000 --> 00:07:07.031
And the results were exactly
00:07:07.031 --> 00:07:09.643
what you were alluding to, Priyanka.
00:07:09.643 --> 00:07:13.249
We saw that the top 10 percentile
00:07:13.249 --> 00:07:17.182
of engineers were roughly four times
00:07:17.182 --> 00:07:19.000
as productive as the median.
00:07:19.000 --> 00:07:21.667
The top 1 percentile of engineers,
00:07:21.667 --> 00:07:23.755
they were about 15 times
00:07:23.755 --> 00:07:26.538
more productive than the median.
00:07:26.538 --> 00:07:29.652
And that is just by the metric of the pull
00:07:29.652 --> 00:07:32.692
requests that these engineers are pushing
00:07:32.692 --> 00:07:34.678
through the system. By other metrics,
00:07:34.678 --> 00:07:36.556
it's even more disparate than that.
00:07:36.556 --> 00:07:37.289
And by the way,
00:07:37.289 --> 00:07:39.000
that was back in the spring, right?
00:07:39.000 --> 00:07:41.348
And in the last six months,
00:07:41.348 --> 00:07:45.000
we've seen these tools evolve a ton again,
00:07:45.000 --> 00:07:47.818
and I would expect that there's
00:07:47.818 --> 00:07:50.000
even more of this today.
00:07:50.000 --> 00:07:52.030
Coming back to this question of what
00:07:52.030 --> 00:07:54.286
the difference is between organizations,
00:07:54.286 --> 00:07:56.735
or maybe individual engineers,
00:07:56.735 --> 00:08:00.000
that are able to get much more velocity,
00:08:00.000 --> 00:08:01.698
to me, I think it really comes
00:08:01.698 --> 00:08:03.000
down to two big things.
00:08:03.000 --> 00:08:05.647
And I'd love to hear Martin's
00:08:05.647 --> 00:08:08.750
take on these to see if he agrees.
00:08:08.750 --> 00:08:12.108
One is, we're seeing a lot of engineers
00:08:12.108 --> 00:08:15.294
and a lot of teams move their agentic
00:08:15.294 --> 00:08:17.231
workloads into the cloud, and it really
00:08:17.231 --> 00:08:19.118
allows for much longer-running agents.
00:08:19.118 --> 00:08:21.668
It allows for a lot more parallelism.
00:08:21.668 --> 00:08:24.286
I think that's certainly been a vector
00:08:24.286 --> 00:08:25.563
where we're seeing a ton
00:08:25.563 --> 00:08:27.000
of additional productivity.
00:08:27.000 --> 00:08:28.869
And I think the other part of this,
00:08:28.869 --> 00:08:31.111
which is a little bit more organizational,
00:08:31.111 --> 00:08:34.143
is that we're also starting to see
00:08:34.143 --> 00:08:36.818
companies redesign their teams
00:08:36.818 --> 00:08:41.231
and processes to be very AI-first and lean
00:08:41.231 --> 00:08:42.646
into the things that AI
00:08:42.646 --> 00:08:44.000
is really powerful at.
00:08:44.000 --> 00:08:45.000
Yeah.
00:08:45.000 --> 00:08:46.000
So it's interesting,
00:08:46.000 --> 00:08:47.818
because I think there's always been
00:08:47.818 --> 00:08:49.688
this challenge where we've increased
00:08:49.688 --> 00:08:52.409
that velocity at the front end but haven't
00:08:52.409 --> 00:08:55.000
necessarily realized it at the back end.
00:08:55.000 --> 00:08:57.088
And I would love to have the stats
00:08:57.088 --> 00:08:58.500
off the top of my head,
00:08:58.500 --> 00:09:00.000
but they've just completely flown away.
00:09:00.000 --> 00:09:03.703
That aside, I do know—certainly from my
00:09:03.703 --> 00:09:07.500
experience of working with customers—the
00:09:07.500 --> 00:09:10.750
ones that are really making a difference
00:09:10.750 --> 00:09:14.000
are where they don't treat AI as a tool.
00:09:14.000 --> 00:09:16.063
They treat AI like a delivery
00:09:16.063 --> 00:09:17.059
system change.
00:09:17.059 --> 00:09:18.138
We're changing the way
00:09:18.138 --> 00:09:19.118
we deliver software,
00:09:19.118 --> 00:09:21.961
and they're treating it as a system,
00:09:21.961 --> 00:09:25.278
and they measure beyond the editor, right?
00:09:25.278 --> 00:09:26.643
Which a lot of people
00:09:26.643 --> 00:09:28.333
have been doing for years.
00:09:28.333 --> 00:09:30.569
But it's not just how many PRs;
00:09:30.569 --> 00:09:32.083
it's the review time,
00:09:32.083 --> 00:09:34.506
the rework, the change failure—not
00:09:34.506 --> 00:09:37.000
just how much is getting to the PR.
00:09:37.000 --> 00:09:38.000
And honestly,
00:09:38.000 --> 00:09:41.675
I do know that the teams that get benefits
00:09:41.675 --> 00:09:45.000
are the ones that automate downstream.
00:09:45.000 --> 00:09:49.057
I do remember the stat that velocity
00:09:49.057 --> 00:09:51.875
gains go from 26% to 57%,
00:09:51.875 --> 00:09:53.879
or something like that,
00:09:53.879 --> 00:09:56.667
where they make those automation
00:09:56.667 --> 00:09:59.757
improvements downstream so that the code
00:09:59.757 --> 00:10:02.692
can get faster from the PR all the way
00:10:02.692 --> 00:10:04.959
to in front of the end customer, whether
00:10:04.959 --> 00:10:07.000
that be internal or external, right?
00:10:07.000 --> 00:10:09.532
You actually want people to consume all
00:10:09.532 --> 00:10:12.000
those good changes that you're making,
00:10:12.000 --> 00:10:13.907
and then be able to measure the impact
00:10:13.907 --> 00:10:15.312
of that once it's delivered.
00:10:15.312 --> 00:10:17.188
"Hey, was that a good change?
00:10:17.188 --> 00:10:19.046
Should I invest more time in that area,
00:10:19.046 --> 00:10:21.000
or do I need to redirect somewhere else?"
00:10:21.000 --> 00:10:22.478
Yeah, I think all
00:10:22.478 --> 00:10:25.000
of that makes a lot of sense.
00:10:25.000 --> 00:10:27.466
We
increasingly are seeing customers
00:10:27.466 --> 00:10:30.000
of ours out there with engineers who,
00:10:30.000 --> 00:10:32.088
for certain types of tasks, are able
00:10:32.088 --> 00:10:34.118
to generate the PR for what they're
00:10:34.118 --> 00:10:36.344
trying to accomplish with their agent
00:10:36.344 --> 00:10:38.571
in 15 minutes or something like that.
00:10:38.571 --> 00:10:40.550
And yet, if you haven't changed
00:10:40.550 --> 00:10:42.083
your downstream process,
00:10:42.083 --> 00:10:44.167
Martin, as you were saying,
00:10:44.167 --> 00:10:45.876
that same PR might sit in a queue
00:10:45.876 --> 00:10:48.000
waiting for review for three days, right?
00:10:48.000 --> 00:10:49.706
And it's like, where have all
00:10:49.706 --> 00:10:52.000
your velocity gains gone at that point?
00:10:52.000 --> 00:10:53.781
I think the differentiation we're
00:10:53.781 --> 00:10:55.455
all talking about is important,
00:10:55.455 --> 00:10:57.493
because a lot of times we talk
00:10:57.493 --> 00:11:00.278
about independent developer productivity,
00:11:00.278 --> 00:11:01.298
and I think that's
00:11:01.298 --> 00:11:03.056
an important part of the story.
00:11:03.056 --> 00:11:04.457
But what you're both talking
00:11:04.457 --> 00:11:05.909
about is enterprise delivery,
00:11:05.909 --> 00:11:08.827
and that's a different way of thinking.
00:11:08.827 --> 00:11:11.071
I think frequently people say,
00:11:11.071 --> 00:11:12.637
"Oh, I have an engineer,
00:11:12.637 --> 00:11:15.312
and that person is much more productive."
00:11:15.312 --> 00:11:17.761
And while I appreciate that, it needs to
00:11:17.761 --> 00:11:20.333
be in the context that you just described,
00:11:20.333 --> 00:11:21.667
with the guidelines, boundaries,
00:11:21.667 --> 00:11:23.432
and expectations of quality
00:11:23.432 --> 00:11:25.000
and security at the end.
00:11:25.000 --> 00:11:26.536
And I wanted to double-click
00:11:26.536 --> 00:11:28.182
on that for just a little bit,
00:11:28.182 --> 00:11:30.531
because I think you talked a little bit
00:11:30.531 --> 00:11:33.000
about process redesign and reimagination.
00:11:33.000 --> 00:11:35.191
I think with that also comes a little
00:11:35.191 --> 00:11:37.500
bit of a reimagination of roles, right?
00:11:37.500 --> 00:11:39.695
You talked about the developer,
00:11:39.695 --> 00:11:41.111
but now we're seeing
00:11:41.111 --> 00:11:43.430
design being shifted—where it happens
00:11:43.430 --> 00:11:46.000
and who does it—and product and security.
00:11:46.000 --> 00:11:49.189
All these roles are shifting a little bit.
00:11:49.189 --> 00:11:51.923
So how do you think about those role
00:11:51.923 --> 00:11:55.321
changes playing a part in the process
00:11:55.321 --> 00:11:58.077
redesign? Not just a redesign,
00:11:58.077 --> 00:11:59.783
but also a reimagination—and
00:11:59.783 --> 00:12:01.489
then add the AI piece to it.
00:12:02.667 --> 00:12:05.016
In a way, we're almost blowing up
00:12:05.016 -->
00:12:07.222
traditional concepts that we've
00:12:07.222 --> 00:12:10.000
had of software development for decades.
00:12:10.000 --> 00:12:13.333
That is 100% happening.
00:12:13.333 --> 00:12:16.037
I genuinely talk to customers all the time
00:12:16.037 --> 00:12:18.676
who've changed the way that they deliver,
00:12:18.676 --> 00:12:19.435
and as a result,
00:12:19.435 --> 00:12:21.000
they've changed their team sizes.
00:12:21.000 --> 00:12:23.274
So one of the common things I've
00:12:23.274 --> 00:12:25.833
been hearing on the ground is, "Hey,
00:12:25.833 --> 00:12:28.429
we used to have a team of seven or eight
00:12:28.429 --> 00:12:31.154
who would work on this service or product.
00:12:31.154 --> 00:12:33.401
Now we have a team of three."
00:12:33.401 --> 00:12:35.571
And we've even got customers
00:12:35.571 --> 00:12:37.754
now where it's the product managers
00:12:37.754 --> 00:12:40.000
who decide when the rollout happens.
00:12:40.000 --> 00:12:42.372
And they're controlling that, working
00:12:42.372 --> 00:12:44.615
with the engineers in such a small,
00:12:44.615 --> 00:12:47.166
tight loop, so that they can give it
00:12:47.166 --> 00:12:50.000
the right promotion and everything else.
00:12:50.000 --> 00:12:52.775
So it's really changing the way people
00:12:52.775 --> 00:12:55.769
work, and I think the expectations around
00:12:55.769 --> 00:12:58.438
the roles that people do have very much
00:12:58.438 --> 00:13:00.833
changed, because I think engineers,
00:13:00.833 --> 00:13:03.438
more and more, are not just engineers.
00:13:03.438 --> 00:13:05.769
They're also a solution architect,
00:13:05.769 --> 00:13:08.821
and they need to be able
00:13:08.821 --> 00:13:12.000
to write good AI prompts.
00:13:12.000 --> 00:13:13.227
They need to be able
00:13:13.227 --> 00:13:14.700
to set those guardrails.
00:13:14.700 --> 00:13:15.896
They need to understand
00:13:15.896 --> 00:13:17.143
what needs to be tested.
00:13:17.143 --> 00:13:18.075
Some of those things
00:13:18.075 --> 00:13:19.286
they already needed to do,
00:13:19.286 --> 00:13:22.652
but it is a very different way of working.
00:13:22.652 --> 00:13:25.938
And how people get trained going forward,
00:13:25.938 --> 00:13:27.943
I think, is also probably gonna change.
00:13:27.943 --> 00:13:30.000
That's a whole other discussion, though.
00:13:30.000 --> 00:13:33.885
But I'd love to hear—maybe I'm way off
00:13:33.885 --> 00:13:38.077
and you guys have a different opinion—but
00:13:38.077 --> 00:13:40.000
that's certainly what I'm seeing.
00:13:40.000 --> 00:13:42.667
No, my opinion is the same, Martin.
00:13:42.667 --> 00:13:45.113
I think we've seen the same thing—had
00:13:45.113 --> 00:13:47.692
the same observation of customers going
00:13:47.692 --> 00:13:50.430
from these classic two-pizza teams,
00:13:50.430 --> 00:13:52.308
like seven to 10 people,
00:13:52.308 --> 00:13:53.908
to these pods that operate
00:13:53.908 --> 00:13:56.000
in a way that's very agent-native.
00:13:56.000 --> 00:13:58.409
And oftentimes, like you said,
00:13:58.409 --> 00:14:00.417
it's two or three people.
00:14:00.417 --> 00:14:02.280
Sometimes it's just one.
00:14:02.280 --> 00:14:05.385
And I think the thing that's interesting
00:14:05.385 --> 00:14:08.027
about this reconfiguration is that because
00:14:08.027 --> 00:14:10.417
the teams have gotten so much smaller,
00:14:10.417 --> 00:14:13.016
it does, I think, leave a bit less
00:14:13.016 --> 00:14:15.385
room for specialization, right?
00:14:15.385 --> 00:14:18.246
When you had that team of seven or eight
00:14:18.246 --> 00:14:21.250
people, you could have engineers that were
00:14:21.250 --> 00:14:24.963
focused on front end or back end or DevOps
00:14:24.963 --> 00:14:28.500
or growth, or whatever it happens to be.
00:14:28.500 --> 00:14:30.218
Now that we've seen this kind
00:14:30.218 --> 00:14:31.818
of reconfiguration of teams
00:14:31.818 --> 00:14:33.636
that are dramatically smaller,
00:14:33.636 --> 00:14:37.186
I think it pays off to have people
00:14:37.186 --> 00:14:41.154
who have more of a generalist mindset.
00:14:41.154 --> 00:14:43.759
I think the skills that are starting
00:14:43.759 --> 00:14:46.364
to become valuable for the engineers
00:14:46.364 --> 00:14:49.710
that really thrive in these two-person
00:14:49.710 --> 00:14:52.000
pods are much closer to...
00:14:52.000 --> 00:14:55.415
It's much closer to—I guess you would call
00:14:55.415 --> 00:14:58.667
it the tech lead manager role that we've
00:14:58.667 --> 00:15:01.069
seen in the previous era, where you
00:15:01.069 --> 00:15:03.333
do a bit of technical leadership,
00:15:03.333 --> 00:15:05.357
you do a bit of product management,
00:15:05.357 --> 00:15:07.619
you talk to customers, and you're able
00:15:07.619 --> 00:15:10.000
to synthesize all these pieces together,
00:15:10.000 --> 00:15:11.546
because the agents are able
00:15:11.546 --> 00:15:13.551
to do so much of the heavy lifting.
00:15:13.551 --> 00:15:16.291
We've found that to be a huge accelerant
00:15:16.291 --> 00:15:19.167
in terms of customers that are able to get
00:15:19.167 --> 00:15:21.000
a lot of velocity out of their AI tools.
00:15:21.000 --> 00:15:22.267
I wanna be specific
00:15:22.267 --> 00:15:24.333
about a couple of things there.
00:15:24.333 --> 00:15:27.794
I think about UX. If I have a UX
00:15:27.794 --> 00:15:31.364
team that builds a skill—an agent
00:15:31.364 --> 00:15:35.041
that has what our expectations are for our
00:15:35.041 --> 00:15:38.542
design thinking and what our style guide
00:15:38.542 --> 00:15:40.215
is—then I can use that to help
00:15:40.215 --> 00:15:42.000
me build the app as an engineer.
00:15:42.000 --> 00:15:45.714
I need that design person involved less.
00:15:45.714 --> 00:15:47.727
I do think the enterprise needs
00:15:47.727 --> 00:15:50.000
to have a standard approach, right?
00:15:50.000 --> 00:15:52.749
But then similarly, if you extend that
00:15:52.749 -->
00:15:55.714
to other areas—let's say I run the middle
00:15:55.714 --> 00:15:58.098
tier—I'm gonna build a skill for everybody
00:15:58.098 --> 00:16:00.312
to use for how to connect to me, right?
00:16:00.312 --> 00:16:01.993
And then on the data side,
00:16:01.993 --> 00:16:04.062
for the database that I'm using,
00:16:04.062 --> 00:16:05.909
I have the same expectation
00:16:05.909 --> 00:16:08.421
of database engineers to build skills
00:16:08.421 --> 00:16:11.000
that I can leverage when I'm building.
00:16:11.000 --> 00:16:12.895
And I think this is another
00:16:12.895 --> 00:16:15.000
new way of thinking, which is:
00:16:15.000 --> 00:16:16.764
I don't expect that the engineers
00:16:16.764 --> 00:16:18.421
are gonna build all the skills.
00:16:18.421 --> 00:16:20.423
There's gonna be a catalog of skills
00:16:20.423 --> 00:16:22.647
that we share that creates the velocity,
00:16:22.647 --> 00:16:23.764
standardization, and control
00:16:23.764 --> 00:16:25.000
that we need in the enterprise.
00:16:25.000 --> 00:16:27.434
I do agree with that. While these
00:16:27.434 --> 00:16:30.385
two-person, three-person pods that we're
00:16:30.385 --> 00:16:34.242
talking about tend to be engineers that
00:16:34.242 --> 00:16:38.000
have the broader generalist skill set,
00:16:38.000 --> 00:16:41.040
I think where the specialization might
00:16:41.040 --> 00:16:44.000
go in these organizations is to folks
00:16:44.000 --> 00:16:46.296
that are helping to build the skills
00:16:46.296 --> 00:16:48.529
and tune the tools so they actually
00:16:48.529 --> 00:16:50.384
do the right things, right?
00:16:50.384 --> 00:16:52.857
We are going to have, like you said,
00:16:52.857 --> 00:16:54.506
designers that are creating
00:16:54.506 --> 00:16:56.154
the design skills, security
00:16:56.154 --> 00:16:58.348
specialists that are creating the security
00:16:58.348 --> 00:17:00.333
review agents, and so on and so forth.
00:17:00.333 --> 00:17:02.386
I'll follow on from that, because
00:17:02.386 --> 00:17:05.000
I do think that's where the shift happens.
00:17:05.000 --> 00:17:06.506
And honestly, you still
00:17:06.506 --> 00:17:08.077
need the experts, right?
00:17:08.077 --> 00:17:09.021
You can't review
00:17:09.021 --> 00:17:10.556
what you don't understand.
00:17:10.556 --> 00:17:13.056
And if you go beyond that to the way
00:17:13.056 --> 00:17:15.000
I think it's actually going,
00:17:15.000 --> 00:17:17.459
a lot of this is about setting
00:17:17.459 --> 00:17:20.000
the right intent for the skills
00:17:20.000 --> 00:17:22.500
and the agents that you're using,
00:17:22.500 --> 00:17:24.879
and also setting the right guardrails
00:17:24.879 --> 00:17:27.000
and limitations around them, too.
00:17:27.000 --> 00:17:29.777
We've all seen those stories in the news
00:17:29.777 --> 00:17:32.692
about AI hacking stuff and doing all sorts
00:17:32.692 --> 00:17:35.030
of things. And so I think there's
00:17:35.030 --> 00:17:37.368
a shift in where the skills live.
00:17:37.368 --> 00:17:39.559
A lot of it is gonna be: you need
00:17:39.559 --> 00:17:42.083
the expertise to define the boundaries
00:17:42.083 --> 00:17:44.413
and define what's acceptable and what's
00:17:44.413 --> 00:17:46.444
not acceptable, and you need those
00:17:46.444 --> 00:17:48.565
experts to be able to set that intent
00:17:48.565 --> 00:17:50.800
so that the generalists can consume it.
00:17:50.800 --> 00:17:52.111
But you still need a path
00:17:52.111 --> 00:17:54.000
from the generalists to the experts.
00:17:54.000 --> 00:17:56.958
I don't think it goes away. I think
00:17:56.958 --> 00:18:00.000
the work they do is different, 100%.
00:18:00.000 --> 00:18:01.720
And I think those roles...
00:18:01.720 --> 00:18:04.167
I had to mention, as you were talking
00:18:04.167 --> 00:18:07.141
about this and the people—there's always
00:18:07.141 --> 00:18:10.263
that conversation about how it's impacting
00:18:10.263 --> 00:18:12.105
the number of roles and stuff—and
00:18:12.105 --> 00:18:13.947
I saw a really interesting piece.
00:18:13.947 --> 00:18:16.276
I think it was even from—I can't
00:18:16.276 --> 00:18:18.824
believe I'm saying this—Jeff Bezos,
00:18:18.824 --> 00:18:20.230
where he was saying that he
00:18:20.230 --> 00:18:22.000
actually thinks it's the opposite.
00:18:22.000 --> 00:18:25.185
He thinks that AI and the use of AI
00:18:25.185 --> 00:18:28.462
will ultimately generate more demand
00:18:28.462 --> 00:18:30.714
for people rather than less.
00:18:30.714 --> 00:18:32.586
He cited a bunch of stuff,
00:18:32.586 --> 00:18:35.250
but it was a really interesting read.
00:18:35.250 --> 00:18:37.215
Maybe I'll find a link to that.
00:18:37.215 --> 00:18:39.750
It was a super interesting read, because
00:18:39.750 --> 00:18:42.486
it was an opposing view to everything
00:18:42.486 --> 00:18:45.000
that we've heard up to this point.
00:18:45.000 --> 00:18:47.344
Yeah, I do think it's true
00:18:47.344 --> 00:18:49.688
that these AI tools change
00:18:49.688 --> 00:18:52.344
the ROI math on a lot of software
00:18:52.344 --> 00:18:55.000
projects that companies might do.
00:18:55.000 --> 00:18:57.394
One of the trends that we've seen
00:18:57.394 --> 00:18:59.643
from some of our big enterprise
00:18:59.643 --> 00:19:02.607
customers is that these days there's a lot
00:19:02.607 --> 00:19:05.500
of use of AI agents to do things like big
00:19:05.500 --> 00:19:07.689
tech debt projects, or big migration
00:19:07.689 --> 00:19:10.000
projects from one platform to another.
00:19:10.000 --> 00:19:13.214
These are the kinds of things where,
00:19:13.214 --> 00:19:15.000
in the pre-AI world,
00:19:15.000 --> 00:19:17.375
your engineering leadership might have
00:19:17.375 --> 00:19:20.000
looked at some of these projects and said,
00:19:20.000 --> 00:19:21.096
"Wow, there's a certain amount
00:19:21.096 --> 00:19:22.593
of benefit that we might get out of this,
00:19:22.593 --> 00:19:24.074
but this is going to
take three years.
00:19:24.074 --> 00:19:25.454
This is gonna take a team
00:19:25.454 --> 00:19:27.000
of ours three years," right?
00:19:27.000 --> 00:19:29.196
And now, in a lot of these cases,
00:19:29.196 --> 00:19:31.591
it can be done in a couple of weeks.
00:19:31.591 --> 00:19:34.318
It can be done in maybe a month.
00:19:34.318 --> 00:19:36.408
The math behind that has completely
00:19:36.408 --> 00:19:38.438
changed, and I think it has caused
00:19:38.438 --> 00:19:40.190
teams to get a lot more ambitious
00:19:40.190 --> 00:19:41.889
with the work that they take on.
00:19:41.889 --> 00:19:42.886
I'm always reminded
00:19:42.886 --> 00:19:44.250
about our backlogs, right?
00:19:44.250 --> 00:19:46.561
We all have backlogs, and so when people
00:19:46.561 --> 00:19:48.929
say, "We're gonna need fewer developers,"
00:19:48.929 --> 00:19:51.240
I'm like, "Maybe. Maybe in five years,
00:19:51.240 --> 00:19:53.611
but we've got backlogs to get through."
00:19:53.611 --> 00:19:55.779
And they're not gonna go overnight. Plus,
00:19:55.779 --> 00:19:58.000
if you had real feature delivery velocity,
00:19:58.000 --> 00:19:59.500
how would our partners
00:19:59.500 --> 00:20:01.000
change their roadmaps?
00:20:01.000 --> 00:20:03.107
And I wanna double-click a little
00:20:03.107 --> 00:20:05.278
bit on what you guys talked about,
00:20:05.278 --> 00:20:06.944
the speed of velocity of this.
00:20:06.944 --> 00:20:08.417
Because while we're focusing
00:20:08.417 --> 00:20:10.417
on getting that speed and reimagining,
00:20:10.417 --> 00:20:12.472
there's still accountability for quality
00:20:12.472 --> 00:20:14.167
standards and reliability, right?
00:20:14.167 --> 00:20:16.231
Martin, you touched on how some people
00:20:16.231 --> 00:20:17.589
can't even review the code
00:20:17.589 --> 00:20:19.000
that's been created, right?
00:20:19.000 --> 00:20:20.000
Yeah.
00:20:20.000 --> 00:20:22.745
What does responsible enterprise
00:20:22.745 --> 00:20:25.833
use look like in practice now, given
00:20:25.833 --> 00:20:27.562
that we're getting to a point where we may
00:20:27.562 --> 00:20:29.167
have generalists writing a lot of this,
00:20:29.167 --> 00:20:31.140
and they may not know the nuances of
00:20:31.140 --> 00:20:33.333
reading something and approving that PR?
00:20:33.333 --> 00:20:34.666
Or maybe things are being
00:20:34.666 --> 00:20:36.000
auto-approved now, right?
00:20:36.000 --> 00:20:37.472
And so what does that look
00:20:37.472 --> 00:20:39.000
like from your perspective?
00:20:39.000 --> 00:20:40.818
Yeah, I genuinely think that's
00:20:40.818 --> 00:20:42.333
a super interesting area,
00:20:42.333 --> 00:20:45.138
because I do think it comes down to,
00:20:45.138 --> 00:20:48.333
yes, questions about who owns the change.
00:20:48.333 --> 00:20:54.091
But I actually think the rise of AI,
00:20:54.091 --> 00:20:56.212
and the way it's accelerated
00:20:56.212 --> 00:20:57.727
our code generation,
00:20:57.727 --> 00:21:00.970
actually puts quite a heavy onus on things
00:21:00.970 --> 00:21:03.750
like the platform engineering teams,
00:21:03.750 --> 00:21:06.233
the security teams, and the audit
00:21:06.233 --> 00:21:09.167
and compliance teams to make sure those
00:21:09.167 --> 00:21:11.167
delivery platforms actually have all
00:21:11.167 --> 00:21:13.000
those guardrails in place, right?
00:21:13.000 --> 00:21:16.758
So that it's less of a—and I think
00:21:16.758 --> 00:21:20.294
people hear you talk about this,
00:21:20.294 --> 00:21:22.353
and their instant reaction is, "Yes,
00:21:22.353 --> 00:21:24.412
but you're just trying to block me."
00:21:24.412 --> 00:21:26.299
From the developer side, you'll hear,
00:21:26.299 --> 00:21:27.778
"Oh, you're just blocking me,
00:21:27.778 --> 00:21:30.000
'cause I have to do this security scan,
00:21:30.000 --> 00:21:31.922
and I have to have this level of test
00:21:31.922 --> 00:21:34.000
coverage, and I have to, and I have to."
00:21:34.000 --> 00:21:36.074
And I view it very differently,
00:21:36.074 --> 00:21:38.750
and actually some of our most successful
00:21:38.750 --> 00:21:40.646
customers view it very differently,
00:21:40.646 --> 00:21:42.812
too—certainly a lot of the larger banks.
00:21:42.812 --> 00:21:44.795
What you're really doing is creating
00:21:44.795 --> 00:21:46.889
a permission structure for them: "Hey,
00:21:46.889 --> 00:21:49.611
here are the boundaries that you can play
00:21:49.611 --> 00:21:52.333
inside of. As long as you're inside them,
00:21:52.333 --> 00:21:53.864
everything you do will just get
00:21:53.864 --> 00:21:55.000
straight to production.
00:21:55.000 --> 00:21:57.183
It'll be super fast. It'll be compliant
00:21:57.183 --> 00:21:58.750
with all of our regulations.
00:21:58.750 --> 00:21:59.750
It'll be audited.
00:21:59.750 --> 00:22:02.551
It will have all the right security
00:22:02.551 --> 00:22:05.833
compliance in there and everything else."
00:22:05.833 --> 00:22:07.594
All the right change management
00:22:07.594 --> 00:22:09.583
stuff happens almost automatically,
00:22:09.583 --> 00:22:12.792
all the way through. I think making
00:22:12.792 --> 00:22:15.909
sure those guardrails are in place
00:22:15.909 --> 00:22:18.221
helps with that kind of ownership,
00:22:18.221 --> 00:22:20.600
responsibility, and accountability.
00:22:22.500 --> 00:22:25.256
Is that the perfect solution? No. But
00:22:25.256 --> 00:22:28.235
it gets you a long way to an environment
00:22:28.235 --> 00:22:30.280
where people are less worried about
00:22:30.280 --> 00:22:32.500
an insecure bit of code being shipped,
00:22:32.500 --> 00:22:33.912
or something that's going
00:22:33.912 --> 00:22:35.833
to break the system being shipped,
00:22:35.833 --> 00:22:37.558
because you're actually building
00:22:37.558 --> 00:22:39.444
all of that in as part of the "Hey,
00:22:39.444 --> 00:22:41.222
here's your permission structure, but to
00:22:41.222 --> 00:22:43.000
play, you do have to have code
coverage.
00:22:43.000 --> 00:22:44.816
You do have to have tests.
00:22:44.816 --> 00:22:46.562
Hey, you've gotta do some
00:22:46.562 --> 00:22:48.307
resiliency testing of your service
00:22:48.307 --> 00:22:50.000
as part of your release process."
00:22:50.000 --> 00:22:53.113
It is what it is. Now, thankfully, AI
00:22:53.113 --> 00:22:56.562
can help with all sorts of parts of that:
00:22:56.562 --> 00:22:57.812
generating the resiliency tests
00:22:57.812 --> 00:22:59.878
for the kind of service that you have
00:22:59.878 --> 00:23:02.000
and the architecture it's deployed in,
00:23:02.000 --> 00:23:03.615
being able to help
00:23:03.615 --> 00:23:05.769
identify security risks,
00:23:05.769 --> 00:23:07.588
maybe automating those changes
00:23:07.588 --> 00:23:09.103
back into a pull request.
00:23:09.103 --> 00:23:11.610
There are all sorts of things that AI
00:23:11.610 --> 00:23:14.118
can help with as part of that to make
00:23:14.118 --> 00:23:16.059
that flow reliable. But there's gotta
00:23:16.059 --> 00:23:18.000
be ownership of what those boundaries
00:23:18.000 --> 00:23:19.667
are and what the rules are.
00:23:19.667 --> 00:23:21.208
And if somebody wants to break
00:23:21.208 --> 00:23:22.750
glass and go outside of those,
00:23:22.750 --> 00:23:24.250
they need to understand that they're
00:23:24.250 --> 00:23:25.625
accountable for that, and they're
00:23:25.625 --> 00:23:27.872
putting their name on the line to say,
00:23:27.872 --> 00:23:30.000
"Hey, I went outside those controls.
00:23:30.000 --> 00:23:32.921
That's on me." And if they break your
00:23:32.921 --> 00:23:36.000
system—most people don't wanna do that.
00:23:36.000 --> 00:23:37.481
Yeah, Martin, I think all
00:23:37.481 --> 00:23:39.200
of that makes a lot of sense.
00:23:39.200 --> 00:23:40.879
When people think about using
00:23:40.879 --> 00:23:42.500
AI for software engineering,
00:23:42.500 --> 00:23:43.812
their minds typically
00:23:43.812 --> 00:23:45.250
go immediately to, "Oh,
00:23:45.250 --> 00:23:46.717
what about all this code
00:23:46.717 --> 00:23:48.000
that I can generate?"
00:23:48.000 --> 00:23:50.000
But I think a thing that's
00:23:50.000 --> 00:23:52.727
underappreciated is the extent to which
00:23:52.727 --> 00:23:55.385
we can use AI and these agents to play
00:23:55.385 --> 00:23:57.657
all kinds of defensive functions
00:23:57.657 --> 00:24:00.000
for organizations as well, right?
00:24:00.000 --> 00:24:02.455
If I look inside of our company
00:24:02.455 --> 00:24:05.385
at where the spend goes for AI usage,
00:24:05.385 --> 00:24:08.427
my bet would be that we probably spend
00:24:08.427 --> 00:24:11.389
a lot more on AI for the defense side
00:24:11.389 --> 00:24:13.980
of things than the offense—the actual
00:24:13.980 --> 00:24:16.500
creation-of-software side of things.
00:24:16.500 --> 00:24:19.708
We're out there using AI for creating
00:24:19.708 --> 00:24:22.917
end-to-end tests, for manual testing,
00:24:22.917 --> 00:24:24.989
for security reviews, for auditing
00:24:24.989 --> 00:24:27.000
things like feature flags, right?
00:24:27.000 --> 00:24:29.019
This is all happening, I think,
00:24:29.019 --> 00:24:31.429
at a level of frequency and diligence
00:24:31.429 --> 00:24:33.151
that actually would've been
00:24:33.151 --> 00:24:35.000
impossible in the pre-AI era.
00:24:35.000 --> 00:24:38.592
And so because of that, even though
00:24:38.592 --> 00:24:42.800
it seems like all these companies are out
00:24:42.800 --> 00:24:44.299
there vibe coding all of this stuff,
00:24:44.299 --> 00:24:45.882
I think there's good reason to believe
00:24:45.882 --> 00:24:48.002
that we're actually gonna head down
00:24:48.002 --> 00:24:50.364
a path where we build some of the best,
00:24:50.364 --> 00:24:52.061
most secure software we ever
00:24:52.061 --> 00:24:54.000
have because of AI capabilities.
00:24:54.000 --> 00:24:55.000
I love that you said that,
00:24:55.000 --> 00:24:56.539
because the news is going
00:24:56.539 --> 00:24:58.571
in the wrong direction right now.
00:24:58.571 --> 00:25:00.457
And unfortunately, I think—
00:25:00.457 --> 00:25:03.182
it's not a criticism—but OpenAI just...
00:25:03.182 --> 00:25:05.811
They didn't follow the standards that we
00:25:05.811 --> 00:25:08.571
follow in all the regulated places, right?
00:25:08.571 --> 00:25:11.250
Yes. And it's to their benefit, right?
00:25:11.250 --> 00:25:13.403
They're trying to be innovative
00:25:13.403 --> 00:25:15.000
and focus on discovery.
00:25:15.000 --> 00:25:17.188
But I really believe what you're saying,
00:25:17.188 --> 00:25:20.336
David, is that we in the enterprise
00:25:20.336 --> 00:25:23.933
have all been so focused on responsible,
00:25:25.000 --> 00:25:26.592
reliable, and secure delivery,
00:25:26.592 --> 00:25:28.077
and that's not gonna change.
00:25:28.077 --> 00:25:29.033
And in some ways,
00:25:29.033 --> 00:25:31.000
we could be better at it than ever.
00:25:31.000 --> 00:25:33.284
I love that. I've actually been working
00:25:33.284 --> 00:25:35.333
with one specific customer recently
00:25:35.333 --> 00:25:37.000
who has some of our security
00:25:37.000 --> 00:25:38.667
tooling and everything else,
00:25:38.667 --> 00:25:40.755
and they have this goal
00:25:40.755 --> 00:25:43.750
of automated 12-hour remediation.
00:25:43.750 --> 00:25:45.060
And essentially,
00:25:45.060 --> 00:25:48.500
upon the detection of a new vulnerability,
00:25:48.500 --> 00:25:52.295
the system knows everywhere it's deployed,
00:25:52.295 --> 00:25:55.909
and it takes two paths at the same time.
00:25:55.909 --> 00:25:57.176
Path one is,
00:25:57.176 --> 00:26:00.556
"I'm gonna look at my firewalls,
00:26:00.556 --> 00:26:02.634
the edge of my network,
00:26:02.634 --> 00:26:06.429
and I'm gonna take any remediating actions
00:26:06.429 --> 00:26:08.743
that I can that are gonna reduce the risk
00:26:08.743 --> 00:26:11.000
while that vulnerability is still open."
00:26:11.000 -->
00:26:13.176
At the same time,
00:26:13.176 --> 00:26:17.143
it's gonna auto-generate the PR
00:26:17.143 --> 00:26:19.572
for that fix and create the Jira ticket
00:26:19.572 --> 00:26:22.000
or ServiceNow ticket or whatever it is,
00:26:22.000 --> 00:26:23.519
so that when the engineer
00:26:23.519 --> 00:26:24.977
comes in in the morning,
00:26:24.977 --> 00:26:27.738
it's sitting there as a critical pull
00:26:27.738 --> 00:26:30.500
request review for them, right there.
00:26:30.500 --> 00:26:33.883
And then, assuming it merges and passes
00:26:33.883 --> 00:26:37.353
all the tests and everything else—which,
00:26:37.353 --> 00:26:38.863
if the PR's been successful,
00:26:38.863 --> 00:26:41.128
it would—it's just awaiting that reviewer,
00:26:41.128 --> 00:26:42.711
and it can push all the way
00:26:42.711 --> 00:26:44.000
through to production.
00:26:44.000 --> 00:26:47.223
So that kind of approach of, "Hey, let's
00:26:47.223 --> 00:26:50.526
shield right, but also fix left, and then
00:26:50.526 --> 00:26:52.893
bring it together in as short a window
00:26:52.893 --> 00:26:55.385
as possible"—I love that as an approach.
00:26:55.385 --> 00:26:58.359
And I think as AI gets better at that,
00:26:58.359 --> 00:27:01.333
we'll be looking at things like, "Hey,
00:27:01.333 --> 00:27:02.447
we've reviewed the PR
00:27:02.447 --> 00:27:03.667
automatically using AI.
00:27:03.667 --> 00:27:06.000
It's a low-risk change for a high value.
00:27:06.000 --> 00:27:08.326
Actually, it could just automatically
00:27:08.326 --> 00:27:10.714
go all the way through to production,"
00:27:10.714 --> 00:27:12.500
going through that governed pipeline,
00:27:12.500 --> 00:27:13.617
that governed process
00:27:13.617 --> 00:27:15.000
that we just talked about.
00:27:15.000 --> 00:27:17.060
You guys spent a little bit of time
00:27:17.060 --> 00:27:19.062
talking about leveraging the power
00:27:19.062 --> 00:27:21.429
of agents to create more secure software,
00:27:21.429 --> 00:27:23.838
and the responsibility of the different
00:27:23.838 --> 00:27:26.000
teams in building those guardrails.
00:27:26.000 --> 00:27:26.689
When we think
00:27:26.689 --> 00:27:28.333
about the developer themselves,
00:27:28.333 --> 00:27:30.916
and the agents they should have, what do
00:27:30.916 --> 00:27:33.500
you think they should be thinking about?
00:27:33.500 --> 00:27:36.182
What do they typically have in their own
00:27:36.182 --> 00:27:38.529
tool set, versus which ones are at,
00:27:38.529 --> 00:27:40.822
let's say, the team level or the company
00:27:40.822 --> 00:27:43.000
level? What do you guys typically see?
00:27:43.000 --> 00:27:46.143
David, why don't you start?
00:27:46.143 --> 00:27:47.499
You have more direct contact
00:27:47.499 --> 00:27:49.000
with those engineers right now.
00:27:49.000 --> 00:27:50.833
Yeah, sure. Amongst our customers,
00:27:50.833 --> 00:27:55.868
I do see this bifurcation of two different
00:27:55.868 --> 00:28:00.303
kinds of agents that are spinning up.
00:28:00.303 --> 00:28:02.784
There are the organizational
00:28:02.784 --> 00:28:05.000
and team-wide agents, and
00:28:05.000 --> 00:28:07.436
the way that I would think of those is
00:28:07.436 --> 00:28:10.000
that these are agents that handle things
00:28:10.000 --> 00:28:12.977
that are just a part of the paved path
00:28:12.977 --> 00:28:16.111
of how this team wants to ship software.
00:28:16.111 --> 00:28:17.301
They're operating the software
00:28:17.301 --> 00:28:18.611
development life cycle that we've
00:28:18.611 --> 00:28:20.286
had for a really long time.
00:28:20.286 --> 00:28:22.739
And I think there, you really want
00:28:22.739 --> 00:28:25.625
a good amount of standardization, right?
00:28:25.625 --> 00:28:29.578
These AI tools have gone from being just
00:28:29.578 --> 00:28:33.333
tools that an individual engineer uses
00:28:33.333 --> 00:28:35.541
on their developer machine to something
00:28:35.541 --> 00:28:37.692
that's more like shared infrastructure
00:28:37.692 --> 00:28:39.377
for the engineering organization.
00:28:39.377 --> 00:28:40.857
And you really want, I think,
00:28:40.857 --> 00:28:42.135
a single path that's gonna work
00:28:42.135 --> 00:28:43.000
really well for that.
00:28:43.000 --> 00:28:44.693
Now, that might be a combination
00:28:44.693 --> 00:28:46.333
of a number of different agents
00:28:46.333 --> 00:28:47.836
that have a number of different
00:28:47.836 --> 00:28:49.000
responsibilities, right?
00:28:49.000 --> 00:28:50.714
It could be code review.
00:28:50.714 --> 00:28:53.298
It could be something that does your pull
00:28:53.298 --> 00:28:55.000
request routing, et cetera.
00:28:55.000 --> 00:28:57.667
And then, in addition to that,
00:28:57.667 --> 00:29:00.708
we also see a lot of engineers that have
00:29:00.708 --> 00:29:03.750
individual agents that they use as well,
00:29:03.750 --> 00:29:06.239
because there are also a lot of use cases
00:29:06.239 --> 00:29:08.667
that are separate from the organization,
00:29:08.667 --> 00:29:10.168
where an individual engineer
00:29:10.168 --> 00:29:11.562
just has... You know what?
00:29:11.562 --> 00:29:13.893
I think one of the things that's actually
00:29:13.893 --> 00:29:15.769
become really common for software
00:29:15.769 --> 00:29:19.985
engineers that are very AI-forward is that
00:29:19.985 --> 00:29:24.000
there is so much parallel work going on.
00:29:24.000 --> 00:29:26.889
There are so many threads of things
00:29:26.889 --> 00:29:29.118
happening at the same time,
00:29:29.118 --> 00:29:30.918
because of what things like cloud
00:29:30.918 --> 00:29:32.500
agents enable, that a typical
00:29:32.500 --> 00:29:34.641
software engineer—especially the ones
00:29:34.641 --> 00:29:36.667
that are in that top 10 percentile,
00:29:36.667 --> 00:29:38.969
top 1 percentile—might have dozens
00:29:38.969 --> 00:29:41.000
of agents doing various things
00:29:41.000 --> 00:29:42.000
for them at the same time.
00:29:42.000 --> 00:29:43.785
And so one of the things
00:29:43.785 --> 00:29:46.389
that we see is that it really helps
00:29:46.389 --> 00:29:48.194
to have a personal agent to help
00:29:48.194 --> 00:29:50.000
you do a bunch of orchestration.
00:29:50.000 --> 00:29:52.381
And this is one of the reasons
00:29:52.381 --> 00:29:55.000
why our company is really excited
00:29:55.000 --> 00:29:55.802
about the launch
00:29:55.802 --> 00:29:57.105
of a product like GrokBot,
00:29:57.105 --> 00:29:58.381
because it is really meant
00:29:58.381 --> 00:30:00.000
to do that kind of thing for you.
00:30:00.000 --> 00:30:02.535
We had these engineers that just had
00:30:02.535 --> 00:30:05.000
so many things going on in parallel
00:30:05.000 --> 00:30:08.066
that they effectively needed a personal
00:30:08.066 --> 00:30:11.053
assistant—someone who's also an expert
00:30:11.053 --> 00:30:12.918
software engineer—that they could lean on
00:30:12.918 --> 00:30:14.737
to help orchestrate all of these things.
00:30:14.737 --> 00:30:15.184
And so that's one
00:30:15.184 --> 00:30:16.000
of the patterns that I've seen.
00:30:16.000 --> 00:30:17.000
Yeah.
00:30:17.000 --> 00:30:19.119
In the last 10 minutes or so,
00:30:19.119 --> 00:30:22.115
I wanna shift gears a little bit and talk
00:30:22.115 --> 00:30:23.612
about some of the things that maybe we're
00:30:23.612 --> 00:30:25.000
not really talking a ton about, right?
00:30:25.000 --> 00:30:27.016
Agile fatigue. It's real,
00:30:27.016 --> 00:30:30.000
but no one's really talking about it.
00:30:30.000 --> 00:30:31.982
How do you guys see standups
00:30:31.982 --> 00:30:34.318
and communication tools like Jira
00:30:34.318 --> 00:30:36.522
and Asana changing as you're thinking
00:30:36.522 --> 00:30:38.667
about integration between your tools
00:30:38.667 --> 00:30:41.000
and just how we're working in that format?
00:30:41.000 --> 00:30:43.545
I did wanna pick up on this,
00:30:43.545 --> 00:30:46.543
just because we have some community stuff
00:30:46.543 --> 00:30:49.615
that we do at Harness called EngineeringX,
00:30:49.615 --> 00:30:52.037
and this has been one of the hot
00:30:52.037 --> 00:30:54.231
topics within that community.
00:30:54.231 --> 00:30:57.273
It has genuinely gone back and forth,
00:30:57.273 --> 00:31:00.022
with people in the community talking
00:31:00.022 --> 00:31:03.000
about, "Do we still do those standups?"
00:31:03.000 --> 00:31:06.537
And there has been a ton of communication
00:31:06.537 --> 00:31:09.643
about how the standups have changed,
00:31:09.643 --> 00:31:11.934
because they tend to be more about intent
00:31:11.934 --> 00:31:13.667
rather than individual tickets,
00:31:13.667 --> 00:31:15.749
if that makes sense. "Hey, these are the
00:31:15.749 --> 00:31:17.778
things that we're aiming toward today,"
00:31:17.778 --> 00:31:19.262
rather than, "Here's the ticket
00:31:19.262 --> 00:31:20.267
I'm working on today,
00:31:20.267 --> 00:31:22.601
and here's the ticket you're working
00:31:22.601 --> 00:31:25.000
on today," which is old-school Agile.
00:31:25.000 --> 00:31:28.108
There's also been quite a heated,
00:31:28.108 --> 00:31:31.875
I would say, discussion on this view of:
00:31:31.875 --> 00:31:35.787
do small pull requests still make sense
00:31:35.787 --> 00:31:40.000
in an AI world, or do big ones make sense?
00:31:40.000 --> 00:31:43.182
I think it's gone back and forth.
00:31:43.182 --> 00:31:46.071
I know what my personal view is,
00:31:46.071 --> 00:31:49.231
but I see organizations doing both,
00:31:49.231 --> 00:31:51.191
and there have been strong
00:31:51.191 --> 00:31:53.000
arguments on both sides.
00:31:53.000 --> 00:31:55.629
You can say that, regardless of whether
00:31:55.629 --> 00:31:58.056
it's a human or an AI, small changes
00:31:58.056 --> 00:32:00.135
actually still make more sense, because
00:32:00.135 --> 00:32:02.000
the blast radius is always smaller.
00:32:02.000 --> 00:32:03.315
They're easier to make sure
00:32:03.315 --> 00:32:04.727
they're compliant and get all
00:32:04.727 --> 00:32:06.388
the tests and everything else done,
00:32:06.388 --> 00:32:08.333
and they're easier and faster to produce.
00:32:08.333 --> 00:32:11.687
And if you look at the old
00:32:11.687 --> 00:32:15.556
DevOps and Agile history, it's
00:32:15.556 --> 00:32:18.531
based on manufacturing: "Hey, here's a lot
00:32:18.531 --> 00:32:21.364
of small things going through a change."
00:32:21.364 --> 00:32:24.375
One of our customers did a webinar
00:32:24.375 --> 00:32:27.917
where he actually played this as a game,
00:32:27.917 --> 00:32:30.593
like a tabletop game, where essentially
00:32:30.593 --> 00:32:32.857
they had their delivery pipeline,
00:32:32.857 --> 00:32:35.162
and they would roll to see
00:32:35.162 --> 00:32:38.000
if any stage of it would change.
00:32:38.000 --> 00:32:40.500
So it would obviously have code,
00:32:40.500 --> 00:32:43.000
PR, test, resiliency, et cetera.
00:32:43.000 --> 00:32:45.146
And then they would do big changes,
00:32:45.146 --> 00:32:46.250
and they would see
00:32:46.250 --> 00:32:48.427
how long it would take them to get all
00:32:48.427 --> 00:32:50.833
the things—their pile of features—shipped,
00:32:50.833 --> 00:32:53.172
versus doing it with small ones.
00:32:53.172 --> 00:32:55.000
And it still always works
00:32:55.000 --> 00:32:56.666
out that you finish faster with lots
00:32:56.666 --> 00:32:58.333
of small changes than with big ones,
00:32:58.333 --> 00:33:00.521
because when a big change goes wrong,
00:33:00.521 --> 00:33:02.000
everything resets, right?
00:33:02.000 --> 00:33:03.561
Yeah. Whereas when a small change
00:33:03.561 --> 00:33:05.312
goes wrong, it's just a small change,
00:33:05.312 --> 00:33:07.371
and everything else is still flowing,
00:33:07.371 --> 00:33:09.375
and it's quicker to fix and resolve.
00:33:09.375 --> 00:33:11.104
So, like I say, it's heated,
00:33:11.104 --> 00:33:12.895
and I do think it's evolving,
00:33:12.895 --> 00:33:15.000
but I still think the principles apply.
00:33:15.000 --> 00:33:17.385
I think you just need to evolve
00:33:17.385 --> 00:33:20.000
it for the way people are working.
00:33:20.000 --> 00:33:22.281
I think we do better at some
00:33:22.281 --> 00:33:24.643
of the things as well, right?
00:33:24.643 --> 00:33:26.218
Agile says light documentation.
00:33:26.218 --> 00:33:27.895
I don't think we need to do that,
00:33:27.895 --> 00:33:30.138
and in fact, I'm not a proponent of that.
00:33:30.138 --> 00:33:31.944
I just think it's so interesting,
00:33:31.944 --> 00:33:33.317
the conversation we're having
00:33:33.317 --> 00:33:35.211
about the number of people on the teams,
00:33:35.211 --> 00:33:37.163
and the way you think of a standup
00:33:37.163 --> 00:33:39.000
or the way you think of a retro.
00:33:39.000 --> 00:33:41.659
I do think these things still should
00:33:41.659 --> 00:33:44.688
exist, but I just think the way we talked
00:33:44.688 --> 00:33:47.860
about it—squads instead of teams,
00:33:47.860 --> 00:33:50.455
maybe three or four people.
00:33:50.455 --> 00:33:52.973
Not two-pizza teams; maybe we're
00:33:52.973 --> 00:33:55.333
talking about one-pizza teams.
00:33:55.333 --> 00:33:58.446
And it'll be interesting to see how the
00:33:58.446 --> 00:34:01.400
enterprise adopts it, and also to see
00:34:04.000 --> 00:34:06.113
if we really are creating that environment
00:34:06.113 --> 00:34:07.824
where things are delivered faster.
00:34:07.824 --> 00:34:09.000
And I think we are.
00:34:09.000 --> 00:34:12.176
Yeah, I think the changes that we're
00:34:12.176 --> 00:34:15.000
seeing in the Agile process are,
00:34:15.000 --> 00:34:18.077
in my opinion, very positive ones.
00:34:18.077 --> 00:34:21.627
If you look at the classic ways in which
00:34:21.627 --> 00:34:25.000
a lot of these meetings have been run,
00:34:25.000 --> 00:34:27.625
you get together in a standup and you talk
00:34:27.625 --> 00:34:30.000
about every single ticket in sequence,
00:34:30.000 --> 00:34:31.611
and you talk about every pull
00:34:31.611 --> 00:34:33.000
request that's going out.
00:34:33.000 --> 00:34:35.301
You talk about the things that are sitting
00:34:35.301 --> 00:34:37.000
in the queue for review and all
00:34:37.000 --> 00:34:38.474
of this stuff, and honestly,
00:34:38.474 --> 00:34:40.000
it's like watching paint dry.
00:34:40.000 --> 00:34:42.269
I don't think there are any engineers
00:34:42.269 --> 00:34:44.231
that really enjoyed that version
00:34:44.231 --> 00:34:46.449
of the process. But I think, as you guys
00:34:46.449 --> 00:34:48.500
were saying, it doesn't mean we don't
00:34:48.500 --> 00:34:50.228
need collaboration, and it doesn't mean
00:34:50.228 --> 00:34:52.000
that we don't need communication, right?
00:34:52.000 --> 00:34:54.728
And in fact, in some ways, we need
00:34:54.728 --> 00:34:57.778
it more, because with these two-person
00:34:57.778 --> 00:35:00.303
pods—especially with more of these
00:35:00.303 --> 00:35:03.125
engineers that are closer to customers
00:35:03.125 --> 00:35:05.381
and have this generalist mindset—they
00:35:05.381 --> 00:35:07.333
actually have a lot of autonomy.
00:35:07.333 --> 00:35:09.667
They can make decisions on their own
00:35:09.667 --> 00:35:11.398
about the products that they're building
00:35:11.398 --> 00:35:13.000
and the things that they're shipping.
00:35:13.000 --> 00:35:15.495
And so I think that actually increases the
00:35:15.495 --> 00:35:17.812
need for a little bit more coordination
00:35:17.812 --> 00:35:19.781
at that level, so that we've got
00:35:19.781 --> 00:35:21.750
these pods that are able to come
00:35:21.750 --> 00:35:23.354
together and talk about all the things
00:35:23.354 --> 00:35:25.000
that they're building and investing in,
00:35:25.000 --> 00:35:27.993
and just make sure that these companies
00:35:27.993 --> 00:35:30.909
that are adopting this really AI-first
00:35:30.909 --> 00:35:34.383
way of working still have some kind
00:35:34.383 --> 00:35:37.857
of one-plus-one-equals-three story,
00:35:37.857 --> 00:35:40.467
instead of just a million AI-first
00:35:40.467 --> 00:35:43.000
pods building whatever they want.
00:35:43.000 --> 00:35:44.667
And listening to you, David,
00:35:44.667 --> 00:35:46.752
I was sitting there thinking how context
00:35:46.752 --> 00:35:48.889
matters to AI—and you hear people talking
00:35:48.889 --> 00:35:50.253
about it all the time—but
00:35:50.253 --> 00:35:51.944
context matters to people, too.
00:35:51.944 --> 00:35:54.025
They both need context, and in an ideal
00:35:54.025 --> 00:35:56.000
world, they would share that context.
00:35:56.000 --> 00:35:58.222
And on that human front,
00:35:58.222 --> 00:36:00.166
I think my next question
00:36:00.166 --> 00:36:03.000
is about context and understanding.
00:36:03.000 --> 00:36:05.022
Now we have younger engineers
00:36:05.022 --> 00:36:06.765
who understand the tools.
00:36:06.765 --> 00:36:08.386
They may be even more savvy
00:36:08.386 --> 00:36:09.706
with the tools, right?
00:36:09.706 --> 00:36:10.447
But they've never
00:36:10.447 --> 00:36:11.667
really learned architecture.
00:36:11.667 --> 00:36:14.181
They haven't done the manual bug fixes
00:36:14.181 --> 00:36:16.364
where they eventually build those
00:36:16.364 --> 00:36:18.293
references in their brains
00:36:18.293 --> 00:36:20.000
and build that context.
00:36:20.000 --> 00:36:21.438
How does human judgment
00:36:21.438 --> 00:36:23.000
stay in that loop, right?
00:36:23.000 --> 00:36:25.000
And how do they learn?
00:36:25.000 --> 00:36:27.206
And what is the responsibility
00:36:27.206 --> 00:36:30.000
that senior developers have with this,
00:36:30.000 --> 00:36:31.348
as more and more autonomous
00:36:31.348 --> 00:36:32.895
agents write more of that code,
00:36:32.895 --> 00:36:34.737
and some of our junior developers never
00:36:34.737 --> 00:36:37.329
even get that hands-on experience
00:36:37.329 -->
00:36:40.000
to start building broader context?
00:36:40.000 --> 00:36:41.358
What does that look like,
00:36:41.358 --> 00:36:43.421
and how do you guys see that evolving?
00:36:43.421 --> 00:36:44.853
'Cause that's becoming a concern,
00:36:44.853 --> 00:36:46.458
right? A lot of the junior developers
00:36:46.458 --> 00:36:47.413
are not getting jobs because
00:36:47.413 --> 00:36:48.333
we're getting these agents.
00:36:48.333 --> 00:36:50.600
But then how do we bring them in,
00:36:50.600 --> 00:36:51.910
and how do we have them start
00:36:51.910 --> 00:36:53.400
thinking like a systems architect
00:36:53.400 --> 00:36:54.912
if they've never done
00:36:54.912 --> 00:36:57.000
the things themselves before?
00:36:57.000 --> 00:37:00.312
Yeah, it's a very fair question. Earlier,
00:37:00.312 --> 00:37:02.779
when I was talking a bit about how we're
00:37:02.779 --> 00:37:05.000
moving toward these two-person pods,
00:37:05.000 --> 00:37:07.206
and so we're favoring generalists
00:37:07.206 --> 00:37:09.211
a little bit more, I certainly
00:37:09.211 --> 00:37:11.161
didn't mean to imply that technical
00:37:11.161 --> 00:37:13.000
judgment isn't important anymore.
00:37:13.000 --> 00:37:14.722
I think it very much
00:37:14.722 --> 00:37:16.875
still is super important.
00:37:16.875 --> 00:37:20.027
And it's because we've all seen
00:37:20.027 --> 00:37:23.077
that the way these agents work
00:37:23.077 --> 00:37:25.257
is that they will take the fastest
00:37:25.257 --> 00:37:27.308
path toward solving any problem,
00:37:27.308 --> 00:37:29.411
and they will also duplicate any
00:37:29.411 --> 00:37:31.842
patterns that they see nearby, right?
00:37:31.842 --> 00:37:33.307
Even the smartest agents that we
00:37:33.307 --> 00:37:35.000
have to date—that's how they operate.
00:37:35.000 --> 00:37:37.424
And I think, left unchecked,
00:37:37.424 --> 00:37:40.455
it means that you could very easily
00:37:40.455 --> 00:37:42.927
build yourself into a situation that's
00:37:42.927 --> 00:37:45.333
unsustainable in various ways, right?
00:37:45.333 --> 00:37:47.000
Either you've got a product
00:37:47.000 --> 00:37:48.667
that isn't built for scale,
00:37:48.667 --> 00:37:50.495
or isn't built for performance,
00:37:50.495 --> 00:37:52.500
or isn't able to be iterated upon.
00:37:52.500 --> 00:37:55.754
And I think at the end of the day, those
00:37:55.754 --> 00:37:58.846
are the kinds of things that our human
00:37:58.846 --> 00:38:01.137
engineers need to be accountable
00:38:01.137 --> 00:38:04.000
for and need to learn to do really well.
00:38:04.000 --> 00:38:06.834
It may not mean that you actually
00:38:06.834 --> 00:38:09.583
need to write any lines of code,
00:38:09.583 --> 00:38:11.849
or even read all the lines of code,
00:38:11.849 --> 00:38:14.375
but to have some level of understanding
00:38:14.375 --> 00:38:16.325
around the architectural decisions
00:38:16.325 --> 00:38:18.333
that are being made, and to be able
00:38:18.333 --> 00:38:21.522
to steer that—that's probably the biggest
00:38:21.522 --> 00:38:24.167
piece that I think the more junior
00:38:24.167 --> 00:38:26.415
engineers in organizations need to learn
00:38:26.415 --> 00:38:28.438
from the senior members of the team,
00:38:28.438 --> 00:38:29.666
so that we can continue
00:38:29.666 --> 00:38:31.000
to thrive in engineering.
00:38:31.000 --> 00:38:34.378
I genuinely think the way we onboard
00:38:34.378 --> 00:38:36.818
engineers is gonna change.
00:38:36.818 --> 00:38:39.504
I think it's almost gonna become
00:38:39.504 --> 00:38:42.778
more like an apprenticeship-type model.
00:38:42.778 --> 00:38:44.856
As in, "Hey, I'm gonna work
00:38:44.856 --> 00:38:47.857
with this mid-level or senior engineer,
00:38:47.857 --> 00:38:50.588
and I'm gonna work side by side with them.
00:38:50.588 --> 00:38:52.321
They're gonna give me tasks
00:38:52.321 --> 00:38:54.118
that help me understand it,"
00:38:54.118 --> 00:38:55.825
whether that's doing reviews or making
00:38:55.825 --> 00:38:57.667
architecture decisions that get reviewed,
00:38:57.667 --> 00:38:59.015
so they build up and learn
00:38:59.015 --> 00:39:00.000
that understanding.
00:39:00.000 --> 00:39:02.257
I think it's much harder
00:39:02.257 --> 00:39:04.231
to teach that in a...
00:39:04.231 --> 00:39:06.523
You can teach principles of architecture
00:39:06.523 --> 00:39:08.929
and coding and everything else when you're
00:39:08.929 --> 00:39:10.925
in college, but I think the way
00:39:10.925 --> 00:39:12.857
that we used to do it—throwing
00:39:12.857 --> 00:39:15.000
them in front of the bug list and saying,
00:39:15.000 --> 00:39:16.757
"Hey, please fix all these bugs
00:39:16.757 --> 00:39:18.400
to learn how to code," right?
00:39:18.400 --> 00:39:21.575
That's not gonna work anymore, because
00:39:21.575 --> 00:39:25.000
why would you get them to do that, right?
00:39:25.000 --> 00:39:26.507
Maybe you need to save some
00:39:26.507 --> 00:39:28.571
of that to let them do it on purpose,
00:39:28.571 --> 00:39:30.768
outside of the AI agent. But for me,
00:39:30.768 --> 00:39:33.088
I think it's going to change very much
00:39:33.088 --> 00:39:35.020
to that apprenticeship-style model, where
00:39:35.020 --> 00:39:37.000
they are helping them learn and understand
00:39:37.000 --> 00:39:38.759
the architecture by doing the tasks
00:39:38.759 --> 00:39:40.417
around setting those constraints,
00:39:40.417 --> 00:39:42.353
setting the intent, understanding
00:39:42.353 --> 00:39:44.583
the risks, and understanding what good
00:39:44.583 --> 00:39:46.730
looks like and what bad looks like,
00:39:46.730 --> 00:39:49.000
so that you're bringing them through.
00:39:49.000 --> 00:39:51.973
And I will note that the organization
00:39:51.973 --> 00:39:54.545
that had the largest recruitment
00:39:54.545 --> 00:39:58.004
of graduates this year has already
00:39:58.004 --> 00:40:01.667
changed its plans for training them,
00:40:01.667 --> 00:40:04.522
and
they published them. They said, "Hey,
00:40:04.522 --> 00:40:07.308
we still want loads of junior engineers,
00:40:07.308 --> 00:40:08.989
but we're changing the way
00:40:08.989 --> 00:40:10.476
we're gonna train them,
00:40:10.476 --> 00:40:11.650
because we can see the way
00:40:11.650 --> 00:40:13.095
they're working needs to change.
00:40:13.095 --> 00:40:15.120
So we need to change the way
00:40:15.120 --> 00:40:17.000
we onboard our graduates."
00:40:17.000 --> 00:40:18.352
As you guys think
00:40:18.352 --> 00:40:20.500
about three years from now,
00:40:20.500 --> 00:40:23.126
how many enterprises do you think
00:40:23.126 --> 00:40:26.071
will have actually crossed that line,
00:40:26.071 --> 00:40:28.077
where they're able to truly get
00:40:28.077 --> 00:40:30.278
to company-wide productivity gains
00:40:30.278 --> 00:40:31.548
and true transformation,
00:40:31.548 --> 00:40:33.611
versus how many do you think will still
00:40:33.611 --> 00:40:35.713
be in this individual productivity
00:40:35.713 --> 00:40:38.000
phase and going through some of this?
00:40:38.000 --> 00:40:40.909
And with that, what is the one
00:40:40.909 --> 00:40:43.470
thing you think is definitely going
00:40:43.470 --> 00:40:46.250
to be the differentiator in those next
00:40:46.250 --> 00:40:48.125
three years that's gonna make a company
00:40:48.125 --> 00:40:50.000
continue to stay successful versus not?
00:40:50.000 --> 00:40:53.836
Wow. So first of all, in today's world,
00:40:53.836 --> 00:40:56.000
three years is a huge—
00:40:56.000 --> 00:40:57.000
Yeah.
00:40:57.000 --> 00:41:00.541
—in terms of predicting what's going on.
00:41:00.541 --> 00:41:03.462
One of the things I would say is,
00:41:03.462 --> 00:41:06.161
I am seeing enterprises really adopt
00:41:06.161 --> 00:41:08.636
things way faster than I thought.
00:41:08.636 --> 00:41:12.209
One of our customers recently adopted
00:41:12.209 --> 00:41:16.071
one of our newer products—a big bank—and
00:41:16.071 --> 00:41:18.499
they're rolling it out to 20,000
00:41:18.499 --> 00:41:21.154
engineers on a super-fast timeline.
00:41:21.154 --> 00:41:23.639
And I'm just like, that's insane
00:41:23.639 --> 00:41:26.667
that that is happening, because I still
00:41:26.667 --> 00:41:29.680
remember tool rollouts from the past
00:41:29.680 --> 00:41:32.692
where you start small and it goes...
00:41:32.692 --> 00:41:35.200
Those proof cycles are happening way
00:41:35.200 --> 00:41:37.917
faster. There is an appetite, for sure,
00:41:37.917 --> 00:41:39.757
and I think they're probably
00:41:39.757 --> 00:41:42.188
gonna get there faster than we think.
00:41:42.188 --> 00:41:43.527
I don't think everybody's
00:41:43.527 --> 00:41:45.294
gonna get there at the same time,
00:41:45.294 --> 00:41:48.387
and there might be some red lines
00:41:48.387 --> 00:41:51.667
to do with governance or regulatory
00:41:51.667 --> 00:41:54.235
compliance that mean it might never get
00:41:54.235 --> 00:41:57.000
there for some of the biggest enterprises.
00:41:57.000 --> 00:41:59.470
But I think there is a genuine
00:41:59.470 --> 00:42:02.188
appetite for them to get as close
00:42:02.188 --> 00:42:03.931
as they can while sticking within
00:42:03.931 --> 00:42:05.833
the guardrails of that organization.
00:42:05.833 --> 00:42:08.937
And I think the ones that are probably
00:42:08.937 --> 00:42:12.368
gonna succeed are the ones that are taking
00:42:12.368 --> 00:42:14.158
that system view: "Here's the system
00:42:14.158 --> 00:42:15.500
of how we deliver software,
00:42:15.500 --> 00:42:18.298
from idea all the way through
00:42:18.298 --> 00:42:21.000
to operating it on day two."
00:42:21.000 --> 00:42:24.044
And they look at that overall system,
00:42:24.044 --> 00:42:26.923
and they are constantly working at,
00:42:26.923 --> 00:42:29.478
"Where can I make an impact that's gonna
00:42:29.478 --> 00:42:31.842
impact that system in a positive way?
00:42:31.842 --> 00:42:34.561
And what does that look like, and how
00:42:34.561 --> 00:42:37.500
do I do it within my organization's risk
00:42:37.500 --> 00:42:39.967
tolerance, within my organization's
00:42:39.967 --> 00:42:42.222
regulatory and compliance side,"
00:42:42.222 --> 00:42:46.000
so that they get that full value.
00:42:46.000 --> 00:42:48.518
But I think if you look at it
00:42:48.518 --> 00:42:51.905
piecemeal—the ones that pick at a piece
00:42:51.905 --> 00:42:53.749
rather than looking at the system—I
00:42:53.749 --> 00:42:55.909
think they're the ones that are gonna lag
00:42:55.909 --> 00:42:58.837
behind, compared with the ones that look
00:42:58.837 --> 00:43:01.765
at that whole life cycle and say, "Okay,
00:43:01.765 --> 00:43:04.035
let's look at this whole system
00:43:04.035 --> 00:43:06.818
as a whole, and now let's work on it."
00:43:06.818 --> 00:43:09.296
Bringing back the old enterprise
00:43:09.296 --> 00:43:11.000
architecture approach.
00:43:11.000 --> 00:43:13.157
Yeah, Martin, I think you really nailed
00:43:13.157 --> 00:43:14.429
it when you said, "Wow,
00:43:14.429 --> 00:43:15.647
three years is a very long
00:43:15.647 --> 00:43:16.818
time in the world of AI."
00:43:16.818 --> 00:43:18.636
Which is crazy to say, right?
00:43:18.636 --> 00:43:20.620
Typically, three- to five-year
00:43:20.620 --> 00:43:23.000
plans have been the standard, right?
00:43:23.000 --> 00:43:27.000
I know. So because of that...
00:43:27.000 --> 00:43:28.736
Maybe this is a spicy take,
00:43:28.736 --> 00:43:29.700
maybe it's not,
00:43:29.700 --> 00:43:32.692
but I think that in three years' time,
00:43:32.692 --> 00:43:36.209
pretty much all enterprises will have
00:43:36.209 --> 00:43:39.917
successfully made it to the other side,
00:43:39.917 --> 00:43:42.917
with organization-wide adoption of agents.
00:43:42.917 --> 00:43:46.170
And I guess the reason why I believe
00:43:46.170 --> 00:43:49.062
that is just looking at the pace
00:43:49.062 --> 00:43:50.442
with which software engineering
00:43:50.442 --> 00:43:52.000
has transformed over the last year.
00:43:52.000 --> 00:43:55.042
I forget exactly when this was,
00:43:55.042 --> 00:43:58.182
but I think it was sometime last
00:43:58.182 --> 00:44:00.652
year that Dario Amodei from Anthropic
00:44:00.652 --> 00:44:03.056
was telling us, "Hey, you know what?
00:44:03.056 --> 00:44:05.057
In the not-too-distant future, all
00:44:05.057 --> 00:44:07.353
code is gonna be written by AI agents."
00:44:07.353 --> 00:44:08.215
And back then,
00:44:08.215 --> 00:44:10.000
it sounded ridiculous, right?
00:44:10.000 --> 00:44:12.422
Because AI software engineering
00:44:12.422 --> 00:44:15.000
barely worked back then. And yet,
00:44:15.000 --> 00:44:18.089
you fast-forward to today, October 2026,
00:44:18.089 --> 00:44:20.714
and that is basically the reality.
00:44:20.714 --> 00:44:23.986
And so if I look forward,
00:44:23.986 --> 00:44:29.091
I think the thing that we should expect
00:44:29.091 --> 00:44:31.779
is that the technology is gonna continue
00:44:31.779 --> 00:44:33.929
to improve at such a rapid pace.
00:44:33.929 --> 00:44:36.425
And when I say that, I don't just mean
00:44:36.425 --> 00:44:39.118
that the models are gonna continue to get
00:44:39.118 --> 00:44:42.094
smarter; I think these AI systems are also
00:44:42.094 --> 00:44:45.000
just gonna get dramatically easier to use
00:44:45.000 --> 00:44:47.639
over the course of the next couple
00:44:47.639 --> 00:44:50.200
of years, to the point where—yes,
00:44:50.200 --> 00:44:51.694
there will be hard parts
00:44:51.694 --> 00:44:53.500
to the transformation, right?
00:44:53.500 --> 00:44:55.363
The change management for large
00:44:55.363 --> 00:44:57.105
organizations is always going
00:44:57.105 --> 00:44:59.582
to be a very tricky thing, but I think the
00:44:59.582 --> 00:45:02.000
agents will be able to help us with that.
00:45:02.000 --> 00:45:04.247
And I think the value of being able
00:45:04.247 --> 00:45:06.429
to lean into that will be so high,
00:45:06.429 --> 00:45:08.236
it's going to be undeniable.
00:45:08.236 --> 00:45:10.882
And I think that's really the reason why,
00:45:10.882 --> 00:45:12.034
in three years' time, I really expect
00:45:12.034 --> 00:45:13.000
every organization to be there.
00:45:13.000 --> 00:45:15.453
If you look at what SpaceX
00:45:15.453 --> 00:45:18.000
did by investing in Cursor,
00:45:18.000 --> 00:45:22.008
it's a great example of how much
00:45:22.008 --> 00:45:27.143
the industry thinks this is worth, right?
00:45:27.143 --> 00:45:29.333
Because it was such an important
00:45:29.333 --> 00:45:31.250
asset for SpaceX to acquire,
00:45:31.250 --> 00:45:32.996
because it's gonna help all
00:45:32.996 --> 00:45:35.000
of those businesses accelerate.
00:45:35.000 --> 00:45:37.082
And I think it was a really smart move
00:45:37.082 --> 00:45:39.000
to make that part of the ecosystem.
00:45:39.000 --> 00:45:41.818
Of course, I certainly agree.
00:45:41.818 --> 00:45:44.285
And I think we've already seen the
00:45:44.285 --> 00:45:47.333
software engineering world change so much,
00:45:47.333 --> 00:45:49.693
and I feel like it's interesting that
00:45:49.693 --> 00:45:52.308
software engineering got disrupted first,
00:45:52.308 --> 00:45:55.000
and the most, of all the functions.
00:45:55.000 --> 00:45:56.212
But that doesn't mean
00:45:56.212 --> 00:45:58.000
the other ones won't be, right?
00:45:58.000 --> 00:45:59.343
I think they're actually
00:45:59.343 --> 00:46:00.909
gonna follow closely behind,
00:46:00.909 --> 00:46:04.251
and we're gonna build amazing products
00:46:04.251 --> 00:46:07.857
that really help with automation and make
00:46:07.857 --> 00:46:10.226
life dramatically easier, better,
00:46:10.226 --> 00:46:12.308
and more efficient across all
00:46:12.308 --> 00:46:13.510
these different functions
00:46:13.510 --> 00:46:14.615
that organizations run.
00:46:14.615 --> 00:46:15.853
So I think that's something
00:46:15.853 --> 00:46:17.000
to be very excited about.
00:46:17.000 --> 00:46:19.204
David, Martin, MK, thank you.
00:46:19.204 --> 00:46:22.396
This was a great conversation, and I think
00:46:22.396 --> 00:46:24.298
there are still a lot of unknowns
00:46:24.298 --> 00:46:26.316
that we're figuring out by the day.
00:46:26.316 --> 00:46:28.146
But I think the partnership,
00:46:28.146 --> 00:46:30.500
and how this is all coming together,
00:46:30.500 --> 00:46:31.312
has been really fantastic.
00:46:31.312 --> 00:46:32.000
So thank you for that.
00:46:32.000 --> 00:46:33.000
Thank you so much, Priya.
00:46:33.000 --> 00:46:35.168
I really appreciate you hosting today.
00:46:35.168 --> 00:46:36.765
It was a great conversation,
00:46:36.765 --> 00:46:38.161
and I appreciate our partners
00:46:38.161 --> 00:46:39.412
from SpaceXSI and Harness.
00:46:39.412 --> 00:46:40.909
We'll continue these discussions.
00:46:40.909 --> 00:46:43.654
Until next time, this is
00:46:43.654 --> 00:46:48.000
Ready, Set, Runtime, an AHEAD podcast.
00:46:48.000 --> 00:46:49.682
Thank you for listening
00:46:49.682 --> 00:46:51.364
to Ready, Set, Runtime.
00:46:51.364 --> 00:46:53.636
If something in today's conversation
00:46:53.636 --> 00:46:55.378
sparked an idea or a question
00:46:55.378 --> 00:46:57.000
about your own environment,
00:46:57.000 --> 00:46:59.000
AHEAD's team would love to hear from you.
00:46:59.000 --> 00:47:01.590
Reach out anytime at ahead.com,
00:47:01.590 --> 00:47:03.846
and if you haven't already,
00:47:03.846 --> 00:47:06.170
follow the show wherever you're tuning in.
00:47:06.170 --> 00:47:07.941
The views shared on this podcast
00:47:07.941 --> 00:47:10.756
are those of the speakers and don't
00:47:10.756 --> 00:47:13.571
necessarily reflect those of AHEAD.
00:47:13.571 --> 00:47:20.600
Until next time.