1
00:00:00,040 --> 00:00:03,240
That was probably one of the 
first big magical moments for me

2
00:00:03,720 --> 00:00:07,040
with using agent harnesses. 
This is Adios Mani. 

3
00:00:07,200 --> 00:00:10,720
He's a director at Google Cloud 
AI, famous for his work on 

4
00:00:10,720 --> 00:00:14,640
Google Chrome and more recently 
focused on AI agents in software

5
00:00:14,640 --> 00:00:17,160
engineering. 
I'll delegate any technical task

6
00:00:17,440 --> 00:00:19,800
to my harness and and see what 
it does. 

7
00:00:20,000 --> 00:00:24,160
I can give it a goal and it will
be able to figure out a path to 

8
00:00:24,160 --> 00:00:27,520
accomplishing that goal. 
Record it live at Google IO. 

9
00:00:27,560 --> 00:00:30,840
We talk about agents, harnesses 
and the back on software 

10
00:00:30,840 --> 00:00:31,880
engineers. 
Good. 

11
00:00:32,119 --> 00:00:35,360
And bad. 
We can also start to lose our 

12
00:00:35,360 --> 00:00:38,360
muscle memory around how things 
get done. 

13
00:00:38,680 --> 00:00:41,480
Cognitive surrender. 
And I think that what you need 

14
00:00:41,480 --> 00:00:46,720
is. 
Have you experimented with like 

15
00:00:46,720 --> 00:00:48,600
Open Claw and Hermes? 
Oh yeah, yeah, yeah. 

16
00:00:48,800 --> 00:00:52,480
What's your thoughts so far? 
I think that in the early days, 

17
00:00:52,480 --> 00:00:56,120
I had a lot of hope and I got, I
felt I felt productive. 

18
00:00:56,120 --> 00:01:01,000
But after a while, having to 
babysit the whole setup became a

19
00:01:01,000 --> 00:01:02,840
little bit more than I was 
expecting. 

20
00:01:02,920 --> 00:01:05,400
Yeah. 
But I feel like being able to 

21
00:01:05,400 --> 00:01:07,720
have it connect to so many 
different third party services 

22
00:01:07,720 --> 00:01:11,240
is like a big on luck because 
you can begin to like delegate 

23
00:01:11,240 --> 00:01:12,800
things you'd usually do 
yourself. 

24
00:01:13,040 --> 00:01:16,360
And it's easier if you're just a
single person trying to do this 

25
00:01:16,360 --> 00:01:18,360
stuff, because if you're a team 
or a business, that's very, 

26
00:01:18,800 --> 00:01:20,680
very, very different, you know? 
Yeah. 

27
00:01:21,080 --> 00:01:24,320
This term babysitting agents is 
something I hear from people 

28
00:01:24,320 --> 00:01:28,280
that are kind of taking a dab 
into working with agents in a 

29
00:01:28,280 --> 00:01:29,720
gentic. 
What would be the biggest thing 

30
00:01:29,720 --> 00:01:31,760
they need to do to no longer 
feel like a babysitter? 

31
00:01:32,440 --> 00:01:35,040
I think that there's probably 
two or three things. 

32
00:01:35,040 --> 00:01:39,200
So the first is if you are 
building a personal agent and 

33
00:01:39,200 --> 00:01:42,240
you're giving it access to, you 
know, your credentials or you're

34
00:01:42,240 --> 00:01:46,200
sharing tokens or anything like 
that, you, you probably have a 

35
00:01:46,200 --> 00:01:48,560
level of comfort about things 
going wrong. 

36
00:01:48,760 --> 00:01:50,440
Maybe you don't, you know, there
are people who've done this 

37
00:01:50,440 --> 00:01:51,680
where they don't really know 
what they're doing. 

38
00:01:52,960 --> 00:01:55,640
If you were trying to graduate 
from babysitting to something 

39
00:01:55,640 --> 00:01:59,920
more serious, you kind of need 
to have identity around your 

40
00:01:59,920 --> 00:02:02,120
agents. 
So if you've got lots of sub 

41
00:02:02,120 --> 00:02:04,040
agents going off and doing 
different things, you kind of 

42
00:02:04,040 --> 00:02:08,240
know well what is doing what. 
You kind of want governance of 

43
00:02:08,240 --> 00:02:11,920
some sort around them as well so
that, you know, if agents can do

44
00:02:11,920 --> 00:02:14,840
anything, you want to limit the 
blast radius because we have 

45
00:02:14,840 --> 00:02:17,240
seen lots of cases of them going
off and doing things that you 

46
00:02:17,240 --> 00:02:20,000
didn't really expect. 
And you know, I've, I've heard 

47
00:02:20,000 --> 00:02:22,520
people telling me examples of 
like, well, I don't, I, I give 

48
00:02:22,520 --> 00:02:25,800
it access to, you know, Google 
CL is, is it going to delete my 

49
00:02:25,800 --> 00:02:28,160
Gmail or is it going to delete 
my customer data or anything 

50
00:02:28,160 --> 00:02:30,520
like that? 
So you really just want to think

51
00:02:30,520 --> 00:02:33,680
about how they can help you be 
effective while also mitigating 

52
00:02:33,680 --> 00:02:35,560
any of the crazy things that can
happen. 

53
00:02:36,040 --> 00:02:39,760
Whenever I hear kind of people 
working with agents and they're 

54
00:02:39,760 --> 00:02:42,040
against it, they give the 
biggest example or someone give 

55
00:02:42,040 --> 00:02:45,160
it access to production and it 
deleted XY and Z catastrophic 

56
00:02:45,160 --> 00:02:46,360
example. 
Absolutely. 

57
00:02:46,640 --> 00:02:49,800
And I think the challenge is 
that, you know, there's people 

58
00:02:49,800 --> 00:02:51,840
feel a lot of FOMO right now, 
right? 

59
00:02:51,840 --> 00:02:54,040
You, you see so many people 
experimenting with their own 

60
00:02:54,040 --> 00:02:57,240
personal agent set up and 
they'll say like, Oh yeah, I've 

61
00:02:57,240 --> 00:03:00,040
never been more productive. 
I'm having it do like half of 

62
00:03:00,040 --> 00:03:03,040
the stuff I'd usually do myself.
And that sounds very impressive 

63
00:03:03,040 --> 00:03:05,280
on the surface. 
But then somebody who doesn't 

64
00:03:05,280 --> 00:03:07,560
really know what they're doing, 
you know, goes and does that and

65
00:03:07,760 --> 00:03:10,920
they could easily leak API keys.
They could end up with a very 

66
00:03:10,920 --> 00:03:13,840
large, you know, bill at the end
of it, those types of things. 

67
00:03:14,200 --> 00:03:17,480
And so I just think people need 
to exercise care while also 

68
00:03:17,480 --> 00:03:20,320
experimenting, you know? 
From a tooling perspective, do 

69
00:03:20,320 --> 00:03:24,040
you think it's the tool's job to
kind of lower the floor a little

70
00:03:24,080 --> 00:03:25,840
bit? 
Easier barrier of entry? 

71
00:03:26,080 --> 00:03:28,480
Better agent experience. 
I guess absolutely. 

72
00:03:29,680 --> 00:03:34,160
There's a great quote by Grady 
Buch, who is the creator of UML,

73
00:03:34,240 --> 00:03:38,120
and he says that the history of 
software engineering is of 

74
00:03:38,120 --> 00:03:42,040
rising abstractions. 
And we have that moment every 

75
00:03:42,040 --> 00:03:46,240
time new technology comes out. 
So, you know, we we have 

76
00:03:46,240 --> 00:03:48,480
JavaScript as a language. 
Well, then we have frameworks 

77
00:03:48,480 --> 00:03:51,240
and libraries that sit on top, 
make it easier for people to to 

78
00:03:51,240 --> 00:03:55,560
build on top of it. 
Now we have agents and harnesses

79
00:03:55,560 --> 00:03:57,840
and models and tool calling and 
all this stuff. 

80
00:03:58,360 --> 00:04:01,040
And so I think we're going to 
see increasing like more 

81
00:04:01,040 --> 00:04:03,440
abstractions that sit on top 
that just make it easier for 

82
00:04:03,440 --> 00:04:06,480
people to not shoot themselves 
in the foot and ideally follow 

83
00:04:06,480 --> 00:04:08,520
those best practices as well as 
they can. 

84
00:04:08,720 --> 00:04:11,000
I I know people are 
experimenting with different 

85
00:04:11,120 --> 00:04:13,000
harnesses. 
Last year for me was the 

86
00:04:13,000 --> 00:04:15,640
experimentation year for agents 
and this year it's more 

87
00:04:15,640 --> 00:04:18,519
harnesses specifically. 
What have you seen that in 

88
00:04:18,519 --> 00:04:21,519
harnesses work really 
effectively if people master 

89
00:04:21,519 --> 00:04:26,160
that from a tool perspective? 
So I, I, I'm, I'm always happy 

90
00:04:26,160 --> 00:04:28,200
to tell people what I use. 
I'm, I'm a very big fan. 

91
00:04:28,200 --> 00:04:30,160
I'm a very big fan of lots of 
different harnesses, including 

92
00:04:30,160 --> 00:04:32,800
the work that Google is doing 
with anti gravity and a lot of 

93
00:04:32,800 --> 00:04:35,280
personal projects. 
I also love using Claude code. 

94
00:04:35,480 --> 00:04:37,400
Shout out to Anthropic and the 
team there. 

95
00:04:38,400 --> 00:04:42,240
I think that one of the things I
love about a good harness is 

96
00:04:42,240 --> 00:04:49,000
that I can give it a goal, and 
regardless of how specific my 

97
00:04:49,000 --> 00:04:51,960
instructions may or may not have
been, it will be able to figure 

98
00:04:51,960 --> 00:04:54,240
out a path to accomplishing that
goal. 

99
00:04:54,720 --> 00:04:57,960
That was probably one of the 
first big magical moments for me

100
00:04:58,360 --> 00:05:02,440
with using agent harnesses. 
So for example, a lot of people 

101
00:05:02,440 --> 00:05:04,760
when they think about agent 
harnesses, they think about the 

102
00:05:04,760 --> 00:05:07,880
eugenic coding use cases, right?
Like we're programmers. 

103
00:05:07,880 --> 00:05:10,920
That of course, is one of the 
first things we think about, but

104
00:05:11,360 --> 00:05:14,560
increasingly for me over the 
last couple of months, I'll 

105
00:05:14,560 --> 00:05:18,400
delegate any technical task to 
my harness and, and see what it 

106
00:05:18,400 --> 00:05:21,440
does. 
So for example, if I have a 

107
00:05:21,440 --> 00:05:26,000
directory of, you know, I, I, I 
do EDM music on the side, right.

108
00:05:26,200 --> 00:05:32,200
So if I have a directory full of
WAV files and I know that well, 

109
00:05:32,200 --> 00:05:35,280
I didn't actually add a nice 
fade out at the end, but I want 

110
00:05:35,280 --> 00:05:37,920
it to be very natural at the 
point when you decide to do the 

111
00:05:37,920 --> 00:05:41,480
fade out. 
I could spend a long time doing 

112
00:05:41,480 --> 00:05:44,160
that manually. 
I decided to give that task to 

113
00:05:44,160 --> 00:05:47,440
Claude code. 
It figured out what tools to 

114
00:05:47,440 --> 00:05:50,440
download. 
It figured out how to determine 

115
00:05:50,960 --> 00:05:54,520
what is the correct point in 
each track to begin the fade 

116
00:05:54,520 --> 00:05:57,680
out, and then it installed the 
tools necessary to actually 

117
00:05:57,680 --> 00:06:00,880
process and make those changes. 
That's a small thing. 

118
00:06:00,880 --> 00:06:03,960
It's a very small thing, but the
fact that you can now delegate 

119
00:06:04,360 --> 00:06:07,720
any of these tasks to an agent 
and have them actually complete 

120
00:06:07,720 --> 00:06:11,000
them, kind of amazing, right? 
I'm not telling it what tools to

121
00:06:11,000 --> 00:06:12,640
use. 
I'm not telling it where to get 

122
00:06:12,640 --> 00:06:15,040
these tools. 
I'm not installing any extra 

123
00:06:15,040 --> 00:06:17,640
skills for these kinds of tasks,
but they still figure them out. 

124
00:06:17,760 --> 00:06:19,960
Something amazing about that. 
Yeah, fantastic. 

125
00:06:20,440 --> 00:06:24,720
From a harness perspective, the 
thing we build around agents or 

126
00:06:24,720 --> 00:06:27,760
kind of the capabilities we have
in place for agents to thrive, 

127
00:06:28,480 --> 00:06:31,560
it's very comparable how I see 
it in practice with people 

128
00:06:31,560 --> 00:06:33,080
falling in love with their first
IDE. 

129
00:06:33,960 --> 00:06:35,560
This harness is my favorite, 
right? 

130
00:06:35,560 --> 00:06:37,880
I've seen this is the most 
effective marketing wise or 

131
00:06:37,880 --> 00:06:39,720
community wise. 
This is what people are using 

132
00:06:39,960 --> 00:06:41,800
either in their work or in their
personal projects. 

133
00:06:42,040 --> 00:06:43,960
Would you recommend people 
experimenting with different 

134
00:06:43,960 --> 00:06:45,440
harnesses and what would the 
gain be? 

135
00:06:45,520 --> 00:06:49,960
Absolutely. 
I think that if you take a look 

136
00:06:50,000 --> 00:06:56,440
at every major model lab or 
developer tooling company, if 

137
00:06:56,440 --> 00:06:59,560
they are vending an agentic 
engineering solution, an agentic

138
00:06:59,560 --> 00:07:02,280
coding solution, they're 
probably offering you up some 

139
00:07:02,280 --> 00:07:04,800
version of a harness. 
And behind that harness is the 

140
00:07:04,800 --> 00:07:07,000
model, right? 
So an agent is the model and the

141
00:07:07,000 --> 00:07:11,400
harness together. 
And so very often, you know your

142
00:07:11,400 --> 00:07:13,800
perspective is going to be, 
well, if I'm using the harness 

143
00:07:13,800 --> 00:07:17,240
that comes from that company. 
Surely the combination of them 

144
00:07:17,520 --> 00:07:20,960
having those teams all working 
together, it being optimized, 

145
00:07:21,040 --> 00:07:23,360
the model being optimized for 
the harness, all of those things

146
00:07:23,360 --> 00:07:25,600
will lead to the very best 
possible outcome. 

147
00:07:26,560 --> 00:07:29,640
Maybe that statement holds true,
but it's also possible, 

148
00:07:29,640 --> 00:07:32,040
especially if you can use that 
model, whether it's your 

149
00:07:32,040 --> 00:07:35,400
subscription or your API keys 
with a different harness, then 

150
00:07:35,400 --> 00:07:37,040
you might get even better 
outcomes. 

151
00:07:37,560 --> 00:07:41,080
And and so for example, open 
code or there there are a list 

152
00:07:41,080 --> 00:07:43,560
of other open source harnesses 
people can try out. 

153
00:07:44,040 --> 00:07:46,560
But I think that the value there
is really just understanding 

154
00:07:46,800 --> 00:07:50,080
what is possible. 
Even if you decide ultimately to

155
00:07:50,080 --> 00:07:52,600
stick with a harness that you 
know the main company is giving 

156
00:07:52,600 --> 00:07:54,680
you because that will give you 
some idea. 

157
00:07:54,680 --> 00:07:57,240
Well, maybe they will go in that
direction in the future. 

158
00:07:57,520 --> 00:08:00,440
And if they don't, I can always 
go try that harness out myself 

159
00:08:00,640 --> 00:08:02,400
to get that extra power. 
Yeah. 

160
00:08:03,000 --> 00:08:07,120
For me, the the industry moves 
so fast that for people that 

161
00:08:07,120 --> 00:08:10,360
say, well, I can't really keep 
up with what is in in this from 

162
00:08:10,360 --> 00:08:13,080
an industry perspective and I 
already have my hands full with 

163
00:08:13,080 --> 00:08:15,080
kind of day-to-day job and I 
don't have bandwidth to 

164
00:08:15,080 --> 00:08:17,560
experiment. 
I feel like it's OK to just 

165
00:08:17,560 --> 00:08:20,360
focus on one harness because I 
also see harnesses learning from

166
00:08:20,360 --> 00:08:21,640
what the other capabilities are 
doing. 

167
00:08:21,640 --> 00:08:24,720
Absolutely. 
We had a panel discussion 

168
00:08:24,720 --> 00:08:29,800
earlier where we were talking 
about how a lot of us feel FOMO 

169
00:08:29,800 --> 00:08:32,880
about our tools today, right? 
Like every other week you'll see

170
00:08:32,880 --> 00:08:36,360
some update where you're like, 
oh man, you know, I, I just 

171
00:08:36,360 --> 00:08:38,559
switched to cloud code, or I 
just switched to codecs or I 

172
00:08:38,559 --> 00:08:40,520
just switched to anti grad and I
have to go and switch over to 

173
00:08:40,520 --> 00:08:43,720
another tool. 
And what I was telling people I 

174
00:08:43,720 --> 00:08:45,480
think applies to harnesses as 
well. 

175
00:08:46,920 --> 00:08:51,840
Think about the time that you 
take to invest in trying in 

176
00:08:51,840 --> 00:08:55,160
another tool as having an 
innovation token or an 

177
00:08:55,160 --> 00:08:58,200
innovation budget, right? 
You don't have all the time in 

178
00:08:58,200 --> 00:09:00,960
the world. 
So be very explicit about where 

179
00:09:00,960 --> 00:09:02,560
you spend that experimentation 
time. 

180
00:09:03,040 --> 00:09:06,320
If there's a harness you're 
using or tool you're using that 

181
00:09:06,320 --> 00:09:10,320
is actually delivering you good 
value right now, stick with it. 

182
00:09:11,280 --> 00:09:16,440
Similar to how orchestration U 
XS are all kind of converging on

183
00:09:16,440 --> 00:09:19,880
the same look and feel. 
I think we see that harnesses 

184
00:09:19,880 --> 00:09:24,840
are also in places converging. 
And so even if you know, you 

185
00:09:24,840 --> 00:09:27,520
feel FOMO about, oh hey, I heard
good things about that other 

186
00:09:27,520 --> 00:09:31,000
project or that other harness 
the patterns that you're 

187
00:09:31,000 --> 00:09:32,920
building up the muscle memory 
you're building up with your 

188
00:09:32,920 --> 00:09:35,720
current one will probably still 
apply if you decide to switch at

189
00:09:35,720 --> 00:09:38,360
a later point. 
So I'm always looking for not 

190
00:09:38,360 --> 00:09:42,560
what is converging, but what is 
like the thing that people are 

191
00:09:42,560 --> 00:09:45,720
trying out that is different to 
everybody else, right? 

192
00:09:45,720 --> 00:09:47,800
Like is somebody trying out 
something in a completely 

193
00:09:47,800 --> 00:09:51,320
different way, even if it looks 
crazy, that'll at least let me 

194
00:09:51,320 --> 00:09:52,920
understand. 
Well, is there value to that 

195
00:09:52,920 --> 00:09:56,920
other idea that's like left 
field, or I'm actually right to 

196
00:09:56,920 --> 00:09:58,080
just stick with my current 
tools? 

197
00:09:58,240 --> 00:10:00,280
Yeah. 
This feeling of FOMO really 

198
00:10:00,280 --> 00:10:03,600
resonates with me because a year
back I had a role as product 

199
00:10:03,600 --> 00:10:05,040
person. 
I always wanted to do that. 

200
00:10:05,040 --> 00:10:07,840
It's very close to engineering. 
I wanted to be responsible, make

201
00:10:07,840 --> 00:10:09,720
decisions, drive a vision, 
etcetera. 

202
00:10:10,320 --> 00:10:13,640
And then specifically kind of 
Q3, a lot of agentic 

203
00:10:13,760 --> 00:10:17,720
capabilities happened last year 
and I felt exactly that FOMO. 

204
00:10:17,720 --> 00:10:20,280
I went back to a hands on role 
because I felt like things are 

205
00:10:20,280 --> 00:10:23,120
evolving so quickly that if I am
not hands on like on a 

206
00:10:23,120 --> 00:10:25,880
day-to-day, I feel like I'm 
really going to be lacking and 

207
00:10:25,880 --> 00:10:28,080
I'm not going to be able to 
resonate and have conversations 

208
00:10:28,080 --> 00:10:29,560
about what is kind of cutting 
edge. 

209
00:10:29,960 --> 00:10:33,080
I really felt that. 
Yeah, I think that's, that's a 

210
00:10:33,080 --> 00:10:37,400
feeling that's difficult to 
shake because very often the 

211
00:10:37,400 --> 00:10:40,520
industry will paint things in 
these broad strokes. 

212
00:10:40,600 --> 00:10:44,640
So if you go on X for example, 
right, you can feel FOMO very 

213
00:10:44,640 --> 00:10:46,120
heavily. 
You scroll through your feet and

214
00:10:46,120 --> 00:10:50,080
it's like, Oh my gosh, there's 
like 2 new harnesses this week. 

215
00:10:50,120 --> 00:10:53,000
There's new models, new open 
source models, new patterns, 

216
00:10:53,000 --> 00:10:55,200
like new skills. 
There's so many things that you 

217
00:10:55,200 --> 00:10:59,280
could potentially try out. 
And I think it's important for 

218
00:10:59,280 --> 00:11:03,240
people to just like calm down it
calm down just a little bit. 

219
00:11:03,560 --> 00:11:07,840
And also realize that there's a 
big difference between, you 

220
00:11:07,840 --> 00:11:12,880
know, the risk that a person 
who's, for example, a solo 

221
00:11:12,880 --> 00:11:15,800
founder or a three person 
company can take trying out like

222
00:11:15,800 --> 00:11:19,640
all these new things, especially
if it's in production and what a

223
00:11:19,640 --> 00:11:22,120
slightly more established 
business would maybe be able to 

224
00:11:22,120 --> 00:11:25,440
do because they need more 
guarantees in place and security

225
00:11:25,440 --> 00:11:27,480
and privacy and all those other 
things need to be thought about.

226
00:11:28,040 --> 00:11:31,760
But at the same time, I think 
the way I look at it is I always

227
00:11:31,760 --> 00:11:34,000
try to stay on top. 
I always try to read, even if 

228
00:11:34,000 --> 00:11:35,680
I'm not trying out every single 
thing. 

229
00:11:35,680 --> 00:11:38,520
I try to read other people's 
experiences about it because it 

230
00:11:38,560 --> 00:11:40,600
might be a leading indicator of 
what's going to come. 

231
00:11:40,960 --> 00:11:44,440
So you can still feel your way 
through the FOMO a little bit 

232
00:11:44,440 --> 00:11:46,080
even if you're not trying out 
everything, I think. 

233
00:11:46,600 --> 00:11:50,000
One of the biggest kind of 
accelerators in my career was 

234
00:11:50,000 --> 00:11:52,680
working in collaboration and a 
big component of that was 

235
00:11:52,680 --> 00:11:55,840
literally just pair programming.
Sitting next together next to 

236
00:11:55,840 --> 00:11:59,640
someone physically or when COVID
happened way more remotely, 

237
00:11:59,800 --> 00:12:02,920
watching someone do their thing,
hands on keyboard or just from 

238
00:12:02,920 --> 00:12:06,280
an ID perspective, navigated 
things, how they solve problems,

239
00:12:06,560 --> 00:12:08,400
thinking before acting, 
etcetera, etcetera. 

240
00:12:08,880 --> 00:12:12,120
I feel like we are a little bit 
losing that because people are 

241
00:12:12,160 --> 00:12:14,280
way more pairing with agents, 
right? 

242
00:12:14,280 --> 00:12:17,440
They put their ideas into a 
terminal, answers come out, they

243
00:12:17,440 --> 00:12:21,360
vibe like that way more. 
I also do that, but I feel like 

244
00:12:21,360 --> 00:12:23,880
then that's maybe a miss because
especially if things are 

245
00:12:23,880 --> 00:12:26,200
accelerating and learnings are 
to be had everywhere. 

246
00:12:26,520 --> 00:12:29,120
I feel like as soon as I see 
someone do that I can 

247
00:12:29,120 --> 00:12:31,640
immediately ask questions and 
get a feeling for what someone 

248
00:12:31,640 --> 00:12:34,280
is doing. 
There there are a few topics 

249
00:12:34,280 --> 00:12:37,080
around this that I've I've 
talked about recently. 

250
00:12:37,160 --> 00:12:42,600
One of them is cognitive debt. 
And so especially as we get used

251
00:12:42,600 --> 00:12:45,480
to the fact that code is getting
generated so quickly right now, 

252
00:12:46,360 --> 00:12:50,880
we can also start to lose our 
muscle memory around how things 

253
00:12:50,880 --> 00:12:53,680
get done right. 
Like, it can erode our memory, 

254
00:12:53,680 --> 00:12:57,480
it can erode our skills because 
we now feel comfortable having 

255
00:12:57,480 --> 00:12:59,960
an AI like always available 
there to help us. 

256
00:13:00,600 --> 00:13:04,400
And for a lot of things that can
be OK, but we're also not being 

257
00:13:04,400 --> 00:13:07,640
very proactive about, well, how 
am I learning everyday? 

258
00:13:07,880 --> 00:13:10,840
How am I staying a little bit 
closer to what's getting 

259
00:13:10,840 --> 00:13:15,080
generated so that if I have to, 
I can roll up my sleeves and go 

260
00:13:15,080 --> 00:13:17,600
in there and make the changes I 
need to just in case the agent 

261
00:13:17,600 --> 00:13:19,400
isn't able to, to help out with 
that. 

262
00:13:20,320 --> 00:13:22,960
And the other end of that 
spectrum is what we're seeing 

263
00:13:22,960 --> 00:13:25,920
more and more of these days, 
which is cognitive surrender. 

264
00:13:26,400 --> 00:13:29,880
So you, you, you just stop 
having critical thinking at all 

265
00:13:29,920 --> 00:13:32,200
and you just blindly accept, 
well, whatever the, the agents 

266
00:13:32,200 --> 00:13:34,080
generated is what I'm going to 
commit. 

267
00:13:34,160 --> 00:13:37,560
It's what I'm going to ship, and
if anything goes wrong, well, I 

268
00:13:37,560 --> 00:13:39,400
don't need to worry about it 
because I'm just going to ask 

269
00:13:39,400 --> 00:13:44,320
the agent to fix it. 
And that's an OK mindset. 

270
00:13:44,320 --> 00:13:47,320
But there's a challenge there. 
How are you going to tell 

271
00:13:47,320 --> 00:13:50,880
whether the agent actually got 
those fixes right or not right? 

272
00:13:51,360 --> 00:13:55,800
So generation is now easy. 
Verification is a big thing that

273
00:13:55,800 --> 00:13:58,160
I think that we as engineers 
need to keep a close eye on. 

274
00:13:58,960 --> 00:14:01,200
Verification. 
If we solve that problem, we 

275
00:14:01,200 --> 00:14:03,560
also solve. 
I feel like a current bottleneck

276
00:14:03,640 --> 00:14:05,960
of kind of manually. 
Reviewing code. 

277
00:14:06,080 --> 00:14:08,880
Yeah, yeah, yeah. 
That I feel like is the current 

278
00:14:08,880 --> 00:14:11,840
bottleneck that we are creating.
Yeah, I would like to have the 

279
00:14:11,840 --> 00:14:15,240
confidence that I have so much 
built in place static code 

280
00:14:15,240 --> 00:14:18,400
analysis to make sure my code is
simple, easy to change. 

281
00:14:18,720 --> 00:14:21,560
Few other, I think more in depth
analysis and kind of 

282
00:14:21,560 --> 00:14:24,880
dependencies and yeah, more of 
the harness, maybe even 

283
00:14:24,880 --> 00:14:27,280
orchestration around a harness 
specifically that I have the 

284
00:14:27,280 --> 00:14:30,760
confidence that I don't even 
have to yes. 

285
00:14:30,840 --> 00:14:35,840
Absolutely. 
I was talking to a a founder 

286
00:14:35,840 --> 00:14:37,520
earlier. 
It's like a basically a one 

287
00:14:37,520 --> 00:14:40,040
person company. 
They were telling me, you know, 

288
00:14:40,040 --> 00:14:42,920
I used to do a lot of coding 
myself and then I started using 

289
00:14:42,920 --> 00:14:45,840
clawed for a lot of stuff. 
And now I have to do all of the 

290
00:14:45,840 --> 00:14:47,360
things that a business owner has
to do. 

291
00:14:47,360 --> 00:14:49,800
So I don't have enough time to 
spend actually reviewing the 

292
00:14:49,800 --> 00:14:52,880
code. 
I'm terrified of, you know, the 

293
00:14:52,920 --> 00:14:56,000
agent doing something wrong and 
then that just getting, you 

294
00:14:56,000 --> 00:14:58,840
know, having a domino effect 
over time, right? 

295
00:14:59,600 --> 00:15:02,120
And so they're asking, well, 
what, what should I do in that 

296
00:15:02,120 --> 00:15:04,640
situation? 
And I think that what you need 

297
00:15:04,640 --> 00:15:09,520
is to codify what good and what 
right means into your system. 

298
00:15:09,920 --> 00:15:15,600
So that can be everything from 
your user journeys to your 

299
00:15:15,600 --> 00:15:19,320
tests, to your visual regression
testing, to, to anything that 

300
00:15:19,720 --> 00:15:22,440
helps ground itself in what 
right and wrong is. 

301
00:15:22,760 --> 00:15:25,760
So that even if you're just 
using the agent to keep doing 

302
00:15:25,760 --> 00:15:29,080
these turns and changing the 
code, it at least has something 

303
00:15:29,080 --> 00:15:32,240
to compare against. 
Because otherwise, like, what's 

304
00:15:32,240 --> 00:15:34,120
it going to do? 
Maybe it will fire open a 

305
00:15:34,120 --> 00:15:37,360
browser and load up the app and 
then click around and see, OK, 

306
00:15:37,360 --> 00:15:39,400
well, I guess I'm able to 
register. 

307
00:15:39,400 --> 00:15:42,040
I guess I'm able to do these 
actions, but that's not the same

308
00:15:42,040 --> 00:15:47,240
as being able to verify that the
behavior hasn't changed from one

309
00:15:47,240 --> 00:15:50,040
release to another. 
So I think that really what 

310
00:15:50,040 --> 00:15:53,200
we're talking about is just 
keeping good engineering hygiene

311
00:15:53,600 --> 00:15:56,040
in place as we're using agents. 
Yeah, absolutely. 

312
00:15:56,240 --> 00:15:58,880
This core ingredient that is 
what does good behavior look 

313
00:15:58,880 --> 00:16:00,840
like. 
I think it's pretty funny that 

314
00:16:00,840 --> 00:16:03,520
you can do that in a specific 
agent that you build. 

315
00:16:03,520 --> 00:16:05,880
You can do it in the skills. 
You can build around the harness

316
00:16:05,880 --> 00:16:08,080
specifically. 
You can assume some stuff from a

317
00:16:08,080 --> 00:16:09,440
harness perspective is already 
in there. 

318
00:16:09,440 --> 00:16:10,920
We now have a spectriven 
development. 

319
00:16:11,280 --> 00:16:14,520
You can do spectriven 
development, but tests 1st and 

320
00:16:14,520 --> 00:16:16,720
do behaviour driven tests with 
so many options. 

321
00:16:17,200 --> 00:16:19,760
And I feel like the innovation 
budget that you talked about, 

322
00:16:19,760 --> 00:16:22,360
it's incredibly important that 
you spend your personal time in 

323
00:16:22,360 --> 00:16:24,040
there. 
But I also think it's important 

324
00:16:24,040 --> 00:16:26,600
that companies need to invest in
the teams they have and also 

325
00:16:26,600 --> 00:16:30,360
allow them to innovate because 
right now we need to experiment.

326
00:16:30,480 --> 00:16:32,120
Yes, I completely agree with 
that. 

327
00:16:32,160 --> 00:16:37,040
And I think that what we often 
see is a lot of developers 

328
00:16:37,040 --> 00:16:42,120
saying, well, I feel I feel this
pressure right now around 

329
00:16:42,120 --> 00:16:45,880
velocity just need to ship 
faster, ship more speed is of 

330
00:16:45,880 --> 00:16:48,840
the essence. 
And so they worry that they're 

331
00:16:48,840 --> 00:16:51,760
not going to have as many 
opportunities to to learn on the

332
00:16:51,760 --> 00:16:54,960
job, to experiment, to get 
better at what they do. 

333
00:16:55,400 --> 00:17:00,680
Running multiple agents does not
mean that there is more of you 

334
00:17:01,200 --> 00:17:06,480
to go around because you have 
cognitive bandwidth that doesn't

335
00:17:06,480 --> 00:17:09,440
parallelize. 
And so I think what's a very 

336
00:17:09,440 --> 00:17:13,240
important is for people to 
figure out what is your comfort 

337
00:17:13,240 --> 00:17:18,000
level with using agents, with 
parallelizing your work and 

338
00:17:18,079 --> 00:17:20,480
don't push yourself too far 
beyond that. 

339
00:17:20,839 --> 00:17:23,960
For me, I'll usually split my 
work into two buckets. 

340
00:17:23,960 --> 00:17:27,119
There's usually things that are 
isolated tasks I'm very 

341
00:17:27,119 --> 00:17:29,840
comfortable delegating and maybe
I'll throw those to cloud 

342
00:17:29,840 --> 00:17:32,840
agents, background agents. 
And then there are tasks where I

343
00:17:32,840 --> 00:17:35,720
really want to use my cognitive 
bandwidth to pay closer 

344
00:17:35,720 --> 00:17:38,120
attention. 
So that can include everything 

345
00:17:38,120 --> 00:17:41,600
from just working more closely 
with the local agent to looking 

346
00:17:41,600 --> 00:17:44,240
at let the code to just paying 
more attention to the details 

347
00:17:44,440 --> 00:17:48,480
and the quality of the outcomes.
And so just be very intentional 

348
00:17:48,720 --> 00:17:51,680
I think is one piece of advice I
would say resonates maybe 

349
00:17:51,680 --> 00:17:53,240
people. 
Thank you so much for coming on 

350
00:17:53,240 --> 00:17:55,080
and sharing. 
My pleasure, thanks for the 

351
00:17:55,080 --> 00:17:55,320
chat.