S4E3: When AI Becomes a Teammate

1
00:00:09,160 --> 00:00:11,480
Welcome back to Sustainability
Forward.

2
00:00:11,560 --> 00:00:15,020
I'm your host, Rishi.
With me, as always, is my Co

3
00:00:15,020 --> 00:00:16,880
host Carmine.
How are you, Carmine?

4
00:00:17,320 --> 00:00:21,720
Good Rishi, welcome to be back.
That's right, Carmine.

5
00:00:21,720 --> 00:00:27,400
Usually on our podcast, we talk
about sustainability in terms

6
00:00:27,400 --> 00:00:32,400
of, you know, energy systems.
We talk about infrastructure,

7
00:00:32,759 --> 00:00:36,240
capital, what's the long term
impact of all of this?

8
00:00:37,040 --> 00:00:42,440
But one of the constraints that
we usually do not talk about is

9
00:00:42,440 --> 00:00:45,720
organizational capacity.
We may have touched up on this

10
00:00:45,720 --> 00:00:50,280
topic in the past, but we don't
discuss it at length.

11
00:00:50,880 --> 00:00:54,000
Of course, we've, you know,
talked about the fact that

12
00:00:54,000 --> 00:00:59,240
energy demand, demand is rising
quite a bit and systems are

13
00:00:59,240 --> 00:01:03,600
getting more and more complex
with the focus on energy

14
00:01:03,600 --> 00:01:06,640
transition, focus on safety,
reliability.

15
00:01:07,280 --> 00:01:11,400
And yet it feels like most
organizations are being asked to

16
00:01:11,400 --> 00:01:15,800
deliver all of this with the
same teams, more or less the

17
00:01:15,800 --> 00:01:19,120
same kind of structure and the
same ways of working.

18
00:01:21,080 --> 00:01:25,760
Having sort of read some of the
things in and around this topic,

19
00:01:26,520 --> 00:01:30,760
I thought today we'll ask one
simple question and that is how

20
00:01:30,760 --> 00:01:34,920
do you scale output?
How do you improve decision

21
00:01:34,920 --> 00:01:39,200
quality and reliability without
simply adding more people,

22
00:01:39,760 --> 00:01:41,440
right?
That's the central thing that we

23
00:01:41,440 --> 00:01:45,040
want to explore in the in this
episode.

24
00:01:45,600 --> 00:01:49,320
And to do that and help us
understand the nuances of this,

25
00:01:49,600 --> 00:01:52,600
we have a special guest.
You please introduce him a

26
00:01:52,600 --> 00:01:54,480
minute.
Yes.

27
00:01:55,200 --> 00:01:58,920
So we're talking today about the
topic and the section of

28
00:01:58,960 --> 00:02:05,840
technology and people, and we
had the pleasure to have Subhot

29
00:02:05,840 --> 00:02:09,479
Kumar.
And Subhot is the founder and

30
00:02:09,479 --> 00:02:13,800
CEO of Brisk AI and alumnus of
Howard Business School.

31
00:02:14,360 --> 00:02:17,760
He brings over 18 years of
global experience across the

32
00:02:17,760 --> 00:02:22,400
energy and technology sectors,
with roles spanning operations,

33
00:02:22,640 --> 00:02:26,840
strategy and product leadership.
Other organizations such as

34
00:02:26,840 --> 00:02:30,920
Shell, BP, the Company, and Dell
Technologies.

35
00:02:31,600 --> 00:02:35,840
This work focuses on helping
enterprises translate data and

36
00:02:35,880 --> 00:02:38,800
AI into measurable business
impact.

37
00:02:39,800 --> 00:02:43,640
Subot is the author of Agentic
AI for Leaders, where the

38
00:02:43,640 --> 00:02:48,360
explorer leaders can build AI
fluency, redesign work, and

39
00:02:48,360 --> 00:02:52,360
scale agentic AI from
experimentation to enterprise

40
00:02:52,360 --> 00:02:55,840
adoption.
This perspective bridges deep

41
00:02:56,040 --> 00:03:01,480
industry context with modern AI
capabilities, helping leaders

42
00:03:01,480 --> 00:03:06,320
think and act transformationally
as they transition towards AI

43
00:03:06,320 --> 00:03:11,800
native operating model.
Subo has also written a book

44
00:03:12,120 --> 00:03:17,200
recently, the title release
Agentic AI for Leaders, about

45
00:03:17,280 --> 00:03:19,600
which we will hear more in this
episode.

46
00:03:19,600 --> 00:03:22,160
So Subot, welcome to
Sustainability Forward.

47
00:03:23,400 --> 00:03:26,160
Thank you.
Thanks Carmine and Rishi for

48
00:03:26,160 --> 00:03:29,280
inviting me to this podcast.
Looking forward to a discussion

49
00:03:29,280 --> 00:03:32,160
on this topic.
Yeah, no, absolutely.

50
00:03:32,800 --> 00:03:37,160
And for a Full disclosure, I
must say that Subot and I go a

51
00:03:37,160 --> 00:03:41,760
long back when we started our
careers in in an energy company.

52
00:03:42,560 --> 00:03:46,400
And so today's discussion is
kind of also reflection of what

53
00:03:46,400 --> 00:03:50,480
has happened since the time we
started working in the energy

54
00:03:50,480 --> 00:03:53,520
industry.
Obviously, a lot has changed and

55
00:03:53,520 --> 00:03:58,440
in the context of AI and things
that have evolved in the energy

56
00:03:58,440 --> 00:04:02,840
industry, today's discussion is
going to be, I think a very

57
00:04:02,840 --> 00:04:06,920
fruitful 1.
So Swarton, the first thing that

58
00:04:06,920 --> 00:04:13,640
I that I was thinking about is,
you know, as I mentioned in the

59
00:04:13,680 --> 00:04:18,279
introduction, the industry seems
to have become more complex

60
00:04:19,040 --> 00:04:21,200
because of things that have
happened within the industry as

61
00:04:21,200 --> 00:04:26,440
well as as market forces.
Is the complexity and the

62
00:04:26,440 --> 00:04:30,920
cognitive load that the industry
faces today, is it more

63
00:04:30,920 --> 00:04:36,280
substantial or bigger than the
workforce size that companies

64
00:04:36,440 --> 00:04:38,600
that companies have?
Can can you help us unpack that

65
00:04:38,600 --> 00:04:41,600
a little bit?
Absolutely, Rashid.

66
00:04:42,000 --> 00:04:45,840
So the way I'm seeing the
industry and as I work with the

67
00:04:46,120 --> 00:04:50,760
leaders in these industries both
on upstream and downstream, one

68
00:04:50,760 --> 00:04:54,520
of the things is becoming
clearer is it's not only the

69
00:04:54,520 --> 00:04:57,480
headcount issue that the
industry is facing.

70
00:04:57,960 --> 00:05:01,280
Like you rightly said, it's all
about the complexity and the

71
00:05:01,280 --> 00:05:05,720
cognitive load that the existing
people, whether they are in

72
00:05:05,720 --> 00:05:09,320
engineering operating roles that
they have to deal with.

73
00:05:09,800 --> 00:05:15,400
So look around globally, right,
Our asset bases are aging and

74
00:05:15,400 --> 00:05:19,160
they have been retrofitted with
additional layers of digital

75
00:05:19,160 --> 00:05:22,920
sensors systems and the
regulations have gone more

76
00:05:22,920 --> 00:05:25,000
stricter.
So there are more compliance

77
00:05:25,120 --> 00:05:29,120
requirements and then you add to
it another layer of energy

78
00:05:29,120 --> 00:05:32,000
transition, right.
So there are these new type of

79
00:05:32,000 --> 00:05:36,200
assets and technologies getting
added in the whole value chain.

80
00:05:37,240 --> 00:05:41,360
So that's what I see as
increasing that cognitive load

81
00:05:42,600 --> 00:05:44,680
and the need to preserve the
context.

82
00:05:45,520 --> 00:05:49,360
So, so that's what I see.
And indeed, like one of the

83
00:05:49,800 --> 00:05:52,640
operations leaders that I was
talking to, he made a very

84
00:05:52,640 --> 00:05:56,520
interesting point.
He said in 90% of the cases when

85
00:05:56,520 --> 00:06:00,760
something goes down in the
plant, they said we are very

86
00:06:00,760 --> 00:06:04,480
sure that this issue has come in
the past, but they just don't

87
00:06:04,480 --> 00:06:08,360
know when it happened and how it
was fixed, right.

88
00:06:08,360 --> 00:06:12,160
So in this industry, I mean,
there are some unique aspects,

89
00:06:12,160 --> 00:06:15,040
right?
Work happens in shifts and the

90
00:06:15,040 --> 00:06:17,600
problems are episodic.
So they are, they may not be

91
00:06:17,600 --> 00:06:19,640
regular.
So it's all about how do you

92
00:06:19,880 --> 00:06:23,200
build that context, reserve that
context and surface it at the

93
00:06:23,200 --> 00:06:28,400
right time so that people don't
have to deal with finding the

94
00:06:28,400 --> 00:06:31,880
information, but rather quickly
with judgement and act on it.

95
00:06:32,840 --> 00:06:37,440
Yeah, I, I remember about when
we were starting off in one of

96
00:06:37,440 --> 00:06:42,880
the earlier kind of foundational
courses about process

97
00:06:42,880 --> 00:06:46,960
engineering, someone said that
you'll not only encounter these

98
00:06:46,960 --> 00:06:50,760
problems again in your career,
you'll also encounter the same

99
00:06:50,760 --> 00:06:53,960
solutions again.
It's just that you may not

100
00:06:53,960 --> 00:06:56,640
remember that this solution was
done by somebody.

101
00:06:57,280 --> 00:07:01,440
So I think that is, that's
probably, you know, it was

102
00:07:01,440 --> 00:07:05,360
probably the same thing that was
reflected many years ago that we

103
00:07:05,360 --> 00:07:09,360
are talking about probably now
at A at a much bigger, bigger

104
00:07:09,360 --> 00:07:12,080
scale.
Yeah.

105
00:07:12,840 --> 00:07:16,920
Yeah, I would say so.
And I mean, one additional force

106
00:07:16,920 --> 00:07:20,200
that I am seeing, right, that
leaders, especially in Northern

107
00:07:20,200 --> 00:07:23,760
America like US, Canada, and I
would love to hear your context

108
00:07:23,760 --> 00:07:30,480
from Europe, is the expertise
retiring faster than the new

109
00:07:30,760 --> 00:07:32,320
talent joining the industry,
right?

110
00:07:32,320 --> 00:07:36,160
So even if it's not that there
are less people in the industry,

111
00:07:36,480 --> 00:07:39,960
it's about leaving, losing that
institutional knowledge.

112
00:07:40,120 --> 00:07:41,880
So I think that's becoming a
bigger issue.

113
00:07:43,840 --> 00:07:46,000
And this is where technology
could help.

114
00:07:47,520 --> 00:07:53,800
So there is lot of talk here and
there about what AI can bring.

115
00:07:54,160 --> 00:07:58,520
But before entering more into
details, probably is helpful to

116
00:07:58,520 --> 00:08:09,640
baseline with you AI tools from
Argentic AI or let's say other

117
00:08:09,640 --> 00:08:14,600
AI types, how they differ in
terms of automation or a

118
00:08:14,600 --> 00:08:18,280
copilot, how they can be a tool
or a teammate.

119
00:08:19,480 --> 00:08:21,600
Can you help us to understand a
little bit how they can play?

120
00:08:21,600 --> 00:08:25,320
I can help us here.
Absolutely common.

121
00:08:25,320 --> 00:08:28,760
And like that's a great point.
And I think the time at which we

122
00:08:28,760 --> 00:08:31,440
are talking about is, is really
transformational.

123
00:08:31,880 --> 00:08:35,360
Like if if you look into some of
the recent news like the SAS

124
00:08:35,360 --> 00:08:40,799
companies stocks have been under
a huge downfall with one news

125
00:08:40,799 --> 00:08:43,360
coming from Entropic that they
launched their core work

126
00:08:43,360 --> 00:08:46,360
platform.
So like you rightly said, so

127
00:08:46,560 --> 00:08:49,840
there are tools that are coming
up, but the bigger thing is the

128
00:08:49,840 --> 00:08:54,240
way in which Agenti Ki has
evolved is that it's no longer

129
00:08:54,240 --> 00:08:58,960
about adopting another tool.
It's really thinking of it as

130
00:08:58,960 --> 00:09:03,080
your teammate, right?
So it's a teammate that can

131
00:09:03,080 --> 00:09:06,680
actually own parts of the work
that can augment the existing

132
00:09:06,680 --> 00:09:11,120
human talent so that you can do
a lot more with the same

133
00:09:11,320 --> 00:09:16,160
workforce, right?
So in the past, our industry in

134
00:09:16,160 --> 00:09:22,520
the past 1520 years, we adopted
lots of machine learning based

135
00:09:22,520 --> 00:09:26,120
solutions, be it on the
operations, on the maintenance

136
00:09:26,120 --> 00:09:29,280
side of things.
But any of those solutions would

137
00:09:29,440 --> 00:09:31,960
eventually give you some
recommendations and then the

138
00:09:31,960 --> 00:09:35,720
humans will make further
judgments, act on those.

139
00:09:36,160 --> 00:09:40,200
And in many cases any solution
that you develop is customized

140
00:09:40,200 --> 00:09:42,480
to a certain plan to a
customized to a certain

141
00:09:42,480 --> 00:09:48,560
scenario.
But now with generative AI that

142
00:09:48,560 --> 00:09:53,080
scaling has been achieved, that
the same trained model can be

143
00:09:53,080 --> 00:09:56,440
taken and applied to many
different problems and many

144
00:09:56,440 --> 00:09:59,920
different contexts with a
reasonable level of certainty.

145
00:10:00,600 --> 00:10:04,920
So that's what is making this
generative AI powered solutions

146
00:10:05,160 --> 00:10:07,800
as your teammates than just
another tool.

147
00:10:09,440 --> 00:10:14,360
And so both is this volume play,
is an efficiency play, is it a

148
00:10:14,360 --> 00:10:18,200
quality play For all the above,
how we should think about it?

149
00:10:19,760 --> 00:10:23,280
I would think all of the above,
but most importantly it's about

150
00:10:23,280 --> 00:10:27,160
the expertise really bringing
more expertise that

151
00:10:27,680 --> 00:10:32,960
organizations that as humans we
have for our rescue, for our

152
00:10:33,040 --> 00:10:36,200
available to, to help us in our
daily jobs, right?

153
00:10:36,800 --> 00:10:40,600
Because efficiency play, I would
say is more about automation.

154
00:10:41,720 --> 00:10:44,440
So even in the traditional
automation, when we have a

155
00:10:44,440 --> 00:10:48,160
reasonable understanding of the
steps that it needs to be taken,

156
00:10:49,160 --> 00:10:52,160
that level of automation has
been in existence for maybe over

157
00:10:52,160 --> 00:10:55,680
a couple of decades now.
I think it's about that

158
00:10:55,680 --> 00:11:00,160
intelligence clear, which to me
is really exciting.

159
00:11:02,920 --> 00:11:10,480
So is IT support incorrect to
think of what is coming into the

160
00:11:10,480 --> 00:11:13,000
industry as the next level of
automation?

161
00:11:13,680 --> 00:11:15,200
That's probably the wrong
benchmark, right?

162
00:11:15,200 --> 00:11:19,320
It's not like we are going from
one level of automation to a

163
00:11:19,320 --> 00:11:21,760
much higher level of automation
with these new tools.

164
00:11:21,760 --> 00:11:25,760
We are actually talking about
intelligence that's going to

165
00:11:25,760 --> 00:11:28,600
recommend that's what to do and
not necessarily just automate

166
00:11:28,600 --> 00:11:31,440
something that we already know
what the steps are.

167
00:11:33,520 --> 00:11:36,960
See the way they see actually
thinking of it as an automation

168
00:11:37,560 --> 00:11:42,880
kind of sets the wrong bar to to
A to a large extent, because

169
00:11:42,960 --> 00:11:47,000
when you think of automation,
you like, we tend to think of a

170
00:11:47,000 --> 00:11:49,400
near perfection, right?
So here we automated it.

171
00:11:49,400 --> 00:11:52,520
So given any of these inputs,
this is output.

172
00:11:53,080 --> 00:11:57,040
But in an agentic AI context, I
think the more we think of it as

173
00:11:57,040 --> 00:12:00,720
a team member, like a new team
member, it may not be giving the

174
00:12:00,720 --> 00:12:05,560
right answers on day one, but we
need to remember that AI based

175
00:12:05,560 --> 00:12:11,480
systems can learn much faster
than lot of the human humans and

176
00:12:11,480 --> 00:12:14,560
they only tend to become better
and better over time, right?

177
00:12:14,560 --> 00:12:19,480
So that's where I tend to think
that it's not about automation,

178
00:12:19,480 --> 00:12:23,920
it's more about intelligence
that works alongside the humans.

179
00:12:26,400 --> 00:12:31,360
Makes sense.
One of the things that I was

180
00:12:31,360 --> 00:12:36,200
thinking of in this context is.
I mean some of the use cases

181
00:12:36,200 --> 00:12:42,000
come to mind very quickly and
there may be are multiple other

182
00:12:42,120 --> 00:12:46,000
use cases that we do not
typically think about in the

183
00:12:46,000 --> 00:12:50,960
context of AI in energy.
Can you help us understand some

184
00:12:50,960 --> 00:12:54,120
of the categories of these use
cases and maybe also give us

185
00:12:54,120 --> 00:12:57,400
some examples which can help us
understand this better?

186
00:12:59,400 --> 00:13:03,000
Sure.
So see there are two kind of use

187
00:13:03,000 --> 00:13:07,720
cases that I am saying, one of
one is kind of these low hanging

188
00:13:07,720 --> 00:13:10,800
fruits which are already in
practice, other are more

189
00:13:10,800 --> 00:13:13,960
aspirational.
So if I think of the low hanging

190
00:13:13,960 --> 00:13:18,040
fruits, those are the use cases
which helps you to reduce

191
00:13:18,040 --> 00:13:20,840
cognitive load and preserve
context.

192
00:13:21,320 --> 00:13:25,640
So think of A use case that we
were working with one of the

193
00:13:25,640 --> 00:13:28,760
clients into maintenance
planning.

194
00:13:28,760 --> 00:13:34,720
So having an AI powered
maintenance planning agent that

195
00:13:34,880 --> 00:13:38,920
drafts of work orders based on
your inspection data by going

196
00:13:38,920 --> 00:13:42,840
into your standard operating
procedures, equipment manuals.

197
00:13:43,240 --> 00:13:46,400
So that the human planners don't
have to spend time in the

198
00:13:46,400 --> 00:13:51,080
paperwork, but rather they spend
more time in simply prioritizing

199
00:13:51,080 --> 00:13:55,040
the tasks and not becoming a
bottleneck for your technicians

200
00:13:55,080 --> 00:13:59,040
on site, right.
So those are the kind of use

201
00:13:59,040 --> 00:14:04,920
cases and another example being
in shift handovers, AI playing a

202
00:14:04,920 --> 00:14:09,960
role as a connective tissue so
that the team members are not

203
00:14:10,000 --> 00:14:13,000
not spending time on
rediscovering the same problems.

204
00:14:13,480 --> 00:14:16,000
So these are the more kind of
practical use cases that I'm

205
00:14:16,000 --> 00:14:20,480
already seeing companies
experimenting with.

206
00:14:20,920 --> 00:14:24,120
And then there are more
aspirational use cases.

207
00:14:24,600 --> 00:14:27,920
The kind of use cases are like
one comes to my mind is related

208
00:14:27,920 --> 00:14:31,680
to subsurface, let's say
reservoir planning, thinking of

209
00:14:32,280 --> 00:14:34,640
deciding where to drill the next
wells.

210
00:14:35,120 --> 00:14:40,240
So in the current context,
engineers spend months of time

211
00:14:40,680 --> 00:14:44,440
in running these simulations and
then deciding where to drill

212
00:14:44,440 --> 00:14:47,120
that next well.
So those are the use cases where

213
00:14:47,120 --> 00:14:51,320
again there are experimentation
happening on how we use these

214
00:14:51,320 --> 00:14:55,040
large language models or large
visual models to to help

215
00:14:55,040 --> 00:14:57,800
facilitate that.
But those are fairly complex

216
00:14:57,800 --> 00:15:01,560
problems given the underlying
physics involved.

217
00:15:01,920 --> 00:15:06,120
So those are the ones I would
say will materialize probably

218
00:15:06,120 --> 00:15:10,040
with more kind of capabilities
on the modeling side.

219
00:15:10,720 --> 00:15:16,400
Yeah.
And so I've had some familiarity

220
00:15:16,840 --> 00:15:20,880
with these, with these issues.
And I think there are two

221
00:15:20,880 --> 00:15:27,360
aspects. 1 is that from
subsurface information linked to

222
00:15:27,360 --> 00:15:31,800
the reservoir?
So #1 you get a huge volume of

223
00:15:31,800 --> 00:15:33,520
data.
And that has always been true.

224
00:15:33,520 --> 00:15:37,800
So a huge volume of data needs
to be processed and it gets

225
00:15:38,120 --> 00:15:41,480
collected, right?
And some surface locations talk

226
00:15:41,480 --> 00:15:43,200
to each other.
So if you know something about

227
00:15:43,200 --> 00:15:47,600
one part of subsurface, it'll
tell you possibly something else

228
00:15:47,600 --> 00:15:50,000
about another part of the same
structure, right?

229
00:15:50,400 --> 00:15:54,440
So with time, AI will improve,
but it can probably bring

230
00:15:54,440 --> 00:15:57,600
improvements in this, in this
understanding in the processing

231
00:15:57,600 --> 00:15:59,840
of this large quality quantity
of data.

232
00:16:00,920 --> 00:16:05,400
I also think there's a second
aspect, which is that drilling

233
00:16:05,400 --> 00:16:07,920
one well incorrectly,
particularly in an offshore

234
00:16:07,920 --> 00:16:14,760
context, can lead to significant
loss of money, right?

235
00:16:15,520 --> 00:16:20,200
And I think that's why coming up
with this pinpointed location

236
00:16:20,200 --> 00:16:24,800
has always been a big issue for
all exploration production

237
00:16:24,800 --> 00:16:27,440
companies.
And I think that's where with

238
00:16:27,440 --> 00:16:33,200
time better, better use of AI
can probably help us pinpoint,

239
00:16:34,120 --> 00:16:36,560
pinpoint accurately where the
next well needs to be drilled.

240
00:16:40,200 --> 00:16:43,880
Yeah.
Any other sort of left field

241
00:16:44,040 --> 00:16:49,560
crazy cases that you have seen
in as as examples of, you know,

242
00:16:49,560 --> 00:16:54,840
potential use cases?
I think use cases are plenty, if

243
00:16:54,840 --> 00:16:58,440
I had to think of it, like you
rightly said, Rishi, it's all

244
00:16:58,440 --> 00:17:03,640
about that tolerance for error
and what's at stake, right?

245
00:17:03,640 --> 00:17:09,720
So there are use cases which are
either economically too, too

246
00:17:09,720 --> 00:17:12,480
risky to perceive or it could be
a little safety.

247
00:17:12,880 --> 00:17:16,440
So that's what I think use
cases, you pick any function,

248
00:17:16,440 --> 00:17:21,000
any rule, and that's where like
like in the book that I have

249
00:17:21,000 --> 00:17:23,480
written, right?
So I have come with a framework

250
00:17:23,880 --> 00:17:26,880
for leaders and for team members
to literally go through

251
00:17:26,880 --> 00:17:29,800
systematically and identify what
are the use cases that they can

252
00:17:29,800 --> 00:17:34,840
have.
I think the real next step is

253
00:17:34,920 --> 00:17:37,680
about really prioritizing and
deciding what are you

254
00:17:37,680 --> 00:17:43,040
comfortable with having AI to do
versus what you continue to own.

255
00:17:43,680 --> 00:17:47,200
So I think that that's the real
decision to be made.

256
00:17:49,200 --> 00:17:52,480
It's about, let's go a little
bit deeper here, also connecting

257
00:17:52,480 --> 00:17:56,760
what you said before about
thinking AI as a teammate,

258
00:17:58,040 --> 00:18:04,640
because this will underlines
also a, a level of trust when

259
00:18:04,920 --> 00:18:10,880
thinking AI, the technology as a
colleague, a level of safety.

260
00:18:13,200 --> 00:18:19,360
So how we can still make this
happening, keeping in mind the

261
00:18:19,360 --> 00:18:22,880
human interaction on how we can
control it?

262
00:18:25,520 --> 00:18:29,720
I think that's, that's a, that's
a really, I mean, good point,

263
00:18:29,840 --> 00:18:32,520
Parvine.
I think the answers we will kind

264
00:18:32,520 --> 00:18:35,480
of find over time in the grand
scheme of things.

265
00:18:35,480 --> 00:18:40,600
But I would say the key here is
to start with use cases as we

266
00:18:40,600 --> 00:18:43,360
call it, like human in the loop
kind of use cases.

267
00:18:43,720 --> 00:18:51,000
So that see AI based solutions,
you need to assume that they are

268
00:18:51,920 --> 00:18:54,720
going to make some mistakes.
They will not always be

269
00:18:54,720 --> 00:18:58,360
accurate, right?
So the key here is how do you

270
00:18:58,360 --> 00:19:03,520
design the work or the use cases
so that it's not right away

271
00:19:03,960 --> 00:19:06,960
hitting the end point, right?
Let's say if it's a customer

272
00:19:06,960 --> 00:19:10,080
facing solution, like if I pick
another example from another

273
00:19:10,080 --> 00:19:14,640
industry like Air Canada, they
had launched an AI powered chat

274
00:19:14,640 --> 00:19:18,840
bot and this chat bot ended up
in, in one of the customer

275
00:19:18,840 --> 00:19:24,080
instructions, ended up promising
a refund to the customer in a

276
00:19:24,080 --> 00:19:27,920
very famous case that went even
to the courts here, which was

277
00:19:27,920 --> 00:19:31,800
against the policy, right?
But I wonder the company,

278
00:19:31,920 --> 00:19:34,040
company had to live with that
because they could not really

279
00:19:34,040 --> 00:19:37,000
pass on the accountability to
the AI that hey, it's on the AI

280
00:19:37,000 --> 00:19:40,040
who said that?
I wonder what would have

281
00:19:40,040 --> 00:19:44,640
happened if the agent would have
upgraded the person to 1st

282
00:19:44,640 --> 00:19:51,320
class.
You know, but but it's an it's

283
00:19:51,320 --> 00:19:56,920
an interesting issue.
I think the point about trusting

284
00:19:56,920 --> 00:20:01,360
an AI teammate, I think that
probably is 1 One thing that's

285
00:20:01,360 --> 00:20:05,400
probably has a long way to go
from from now.

286
00:20:07,720 --> 00:20:09,480
At least, not really.
Out of dress.

287
00:20:09,480 --> 00:20:13,920
Maybe.
So it's about fear, it's about

288
00:20:14,120 --> 00:20:18,960
OK, probably some part of the
work is speaking taken by

289
00:20:18,960 --> 00:20:21,280
somebody else, kind of
automated.

290
00:20:21,280 --> 00:20:24,560
So there are all this kind of
the dynamic that we need to

291
00:20:24,720 --> 00:20:29,120
learn to live with, I think.
Yeah, yeah, it's the.

292
00:20:29,120 --> 00:20:31,680
Fear of even job loss to some
extent.

293
00:20:32,040 --> 00:20:37,400
And today, interestingly, one of
my ex colleagues, he, he, he

294
00:20:37,400 --> 00:20:39,920
just sent me a text right after
reading the book and he made

295
00:20:39,920 --> 00:20:41,480
some really interesting
observations.

296
00:20:41,480 --> 00:20:45,960
He's like, OK, AI as a team
member, but how do I think about

297
00:20:45,960 --> 00:20:50,320
social interactions?
Then it's he's like, OK, maybe I

298
00:20:50,320 --> 00:20:52,720
will do some work.
I will do some work, We will

299
00:20:52,720 --> 00:20:55,840
collaborate.
But I go to office for work

300
00:20:55,840 --> 00:20:58,200
because I also want to interact
with people, right?

301
00:20:58,440 --> 00:21:02,360
So I would say there, there are
some issues that are going to

302
00:21:02,360 --> 00:21:05,400
come up as we adopt the
technology more and more.

303
00:21:05,880 --> 00:21:09,480
So at least it's good to kind of
think through these potential

304
00:21:09,480 --> 00:21:13,240
issues and then slowly the
solutions will start shaping.

305
00:21:14,240 --> 00:21:16,440
Yeah.
Maybe 10 years later I'll be

306
00:21:16,440 --> 00:21:20,960
doing a podcast with with an AI
teammate probably even.

307
00:21:21,080 --> 00:21:27,640
Sooner, yeah.
Now, what I wanted to ask you

308
00:21:27,840 --> 00:21:33,440
about having read, you know,
some parts of your book, which

309
00:21:33,440 --> 00:21:39,480
is appropriately titled Agentic
AI for Leaders, is how should or

310
00:21:39,480 --> 00:21:42,480
what should leaders be doing
differently today?

311
00:21:43,240 --> 00:21:47,840
Whether it comes to redesigning
the work, you know, building AI

312
00:21:47,840 --> 00:21:51,360
fluency within their
organizations or even

313
00:21:51,360 --> 00:21:54,520
encouraging safe
experimentation, right?

314
00:21:55,000 --> 00:21:59,680
Where should they start?
Is there sort of a good sequence

315
00:21:59,680 --> 00:22:03,800
of things to do today that will
set them up for success for the

316
00:22:03,800 --> 00:22:08,160
future?
I would say Rishi, like from the

317
00:22:08,160 --> 00:22:11,720
experience that I have had after
interacting with multiple

318
00:22:11,720 --> 00:22:16,040
companies, I see one of the key
starting points has to be

319
00:22:16,760 --> 00:22:21,320
building a, a fluency for the
functional teams.

320
00:22:21,680 --> 00:22:25,640
Because a lot of the companies
that I interact with, people are

321
00:22:26,040 --> 00:22:31,960
still thinking of agent AI in
the form of automation, right?

322
00:22:31,960 --> 00:22:36,600
So, and sometimes they are even
linking a technology tool.

323
00:22:36,880 --> 00:22:40,400
The failure of the technology
tool, let's say without naming,

324
00:22:40,400 --> 00:22:44,760
say one of the copilots in a
workplace gave them wrong

325
00:22:44,760 --> 00:22:48,280
answer.
And people would very soon

326
00:22:48,360 --> 00:22:52,200
extrapolated that, hey, agent AI
is still not, hasn't still

327
00:22:52,200 --> 00:22:55,080
reached that level, right?
So that's where I think that

328
00:22:55,080 --> 00:22:59,000
fluency is the foundational
stone for people to understand

329
00:22:59,000 --> 00:23:03,080
what agent AI generated AI are
capable of, what they are not.

330
00:23:03,640 --> 00:23:06,000
Because end of the day,
functional teams have to play

331
00:23:06,000 --> 00:23:08,360
much bigger role in this
transition.

332
00:23:08,680 --> 00:23:12,680
So it's less of a technology
part, more of an operating model

333
00:23:12,680 --> 00:23:15,280
redesign that leaders need to
think about, right?

334
00:23:15,760 --> 00:23:17,720
So that's what I think the
starting point is building a

335
00:23:17,800 --> 00:23:21,960
fluency.
And as teams start to leverage

336
00:23:22,040 --> 00:23:27,400
AI, they start to experiment.
Leaders have to start thinking

337
00:23:27,400 --> 00:23:30,040
about this operating model
redesign.

338
00:23:30,040 --> 00:23:33,280
What I mean by that is what are
the new roles and

339
00:23:33,280 --> 00:23:35,840
responsibilities?
What, what are we comfortable

340
00:23:35,840 --> 00:23:39,760
with AI owning versus human
zoning, right?

341
00:23:40,040 --> 00:23:42,400
What are the right guardrails
that we put in place?

342
00:23:42,760 --> 00:23:45,280
What are the right control
mechanisms we put in place,

343
00:23:45,600 --> 00:23:47,880
right?
So let's say if there are

344
00:23:47,960 --> 00:23:53,080
autonomous cars on the roads,
are you able to switch them off

345
00:23:53,080 --> 00:23:56,960
when you need to, right?
So and if anything goes wrong,

346
00:23:56,960 --> 00:23:59,480
who takes accountability?
So designing that whole

347
00:23:59,480 --> 00:24:03,960
operating model is crucial next
step once you build that air

348
00:24:03,960 --> 00:24:09,000
fluency and start experimenting.
Yeah, so.

349
00:24:09,560 --> 00:24:12,440
Let's.
Take that on on the book on this

350
00:24:12,440 --> 00:24:15,280
new book you just wrote.
So a gentic AI for leaders.

351
00:24:17,160 --> 00:24:21,200
Why did you choose to write this
book and what do you think are

352
00:24:21,200 --> 00:24:25,240
some of the key messages and
blind spot that we should be

353
00:24:25,240 --> 00:24:27,680
aware?
Yeah.

354
00:24:27,680 --> 00:24:33,160
So my core motivation for this
Carmen has been based on the

355
00:24:33,160 --> 00:24:37,000
consistent pattern that I saw
while interacting with the

356
00:24:37,000 --> 00:24:40,720
leaders from from across
industries.

357
00:24:41,360 --> 00:24:47,440
And that's been that there's a
huge curiosity and excitement

358
00:24:47,440 --> 00:24:51,480
about the technology that
leaders have, but they don't

359
00:24:51,480 --> 00:24:54,560
know where to start, right?
So and that's what we are

360
00:24:54,560 --> 00:24:56,600
seeing, right.
The advancements in a a

361
00:24:56,680 --> 00:25:00,120
technologies have been
humongous, but the real world

362
00:25:00,120 --> 00:25:05,440
applications are still very few.
So that was my core motivation.

363
00:25:05,640 --> 00:25:12,760
How do we create the right set
of information like that source

364
00:25:12,760 --> 00:25:16,800
of how do how does it become the
right source of information so

365
00:25:16,800 --> 00:25:20,520
that the function teams leaders
can understand what this AI

366
00:25:20,520 --> 00:25:24,040
technology is all about?
And secondly, how do they go

367
00:25:24,040 --> 00:25:27,080
about systematically discovering
use cases in their own teams

368
00:25:27,080 --> 00:25:31,520
because that's where they're
struggling that what all can

369
00:25:31,520 --> 00:25:34,320
they really do with AI and where
should they start?

370
00:25:34,320 --> 00:25:36,360
So that's been the core
motivation.

371
00:25:38,120 --> 00:25:41,000
Sorry, I forgot Carmina, what
was the other part of the

372
00:25:41,000 --> 00:25:43,120
question?
I know it was probably and if

373
00:25:43,120 --> 00:25:45,720
you I.
Know that's very helpful it's

374
00:25:46,560 --> 00:25:51,040
but the question was there is
any blind spot that you're doing

375
00:25:51,040 --> 00:25:55,560
your research that you saw and
you recommend the leader to look

376
00:25:55,560 --> 00:26:00,040
at things that for example we
are underestimating,

377
00:26:00,040 --> 00:26:02,600
overlooking.
Absolutely.

378
00:26:02,600 --> 00:26:06,680
I think one of the biggest blind
spots I see is lot of the

379
00:26:06,680 --> 00:26:11,840
leaders are still treating AI as
another technology deployment

380
00:26:12,200 --> 00:26:15,880
and hence they're leaving it
largely to their IT teams or to

381
00:26:15,880 --> 00:26:19,360
the technology vendors to come
and pitch it to them, right?

382
00:26:19,360 --> 00:26:22,240
And that's where you see, right,
MIT came with some statistics

383
00:26:22,240 --> 00:26:26,760
like 95% of the pilots, early
pilots have failed, right.

384
00:26:27,040 --> 00:26:30,480
And one of the key reasons is
when you are going after those

385
00:26:30,480 --> 00:26:34,720
shiny demos by technology
vendors, right, you are not

386
00:26:34,720 --> 00:26:37,760
necessarily picking the right
use cases for your functions,

387
00:26:38,560 --> 00:26:40,040
right?
So I think that's one of the key

388
00:26:40,280 --> 00:26:44,880
kind of blind spots that it's
not just another technology to

389
00:26:44,880 --> 00:26:48,200
be adopted, it's a real world
operating model change.

390
00:26:48,560 --> 00:26:52,720
So leaders need to proactively
think through what parts of

391
00:26:52,720 --> 00:26:57,800
their work could be reshaped and
assigned to some of the AI

392
00:26:57,800 --> 00:27:01,920
agents, while still keeping the
overall control on the decision

393
00:27:01,920 --> 00:27:06,680
making.
But I'm.

394
00:27:06,680 --> 00:27:11,760
Going to ask you to do a little
bit of crystal ball gazing and,

395
00:27:11,760 --> 00:27:15,720
and, and tell us when do you
think the energy industry in

396
00:27:15,720 --> 00:27:19,960
particular will see a
breakthrough in terms of

397
00:27:19,960 --> 00:27:25,320
adoption of of AI?
Are we talking about 10 years or

398
00:27:25,320 --> 00:27:29,360
is it coming in 18 months?
You know, is there sort of, you

399
00:27:29,360 --> 00:27:31,600
know, having done all this
research for your book and

400
00:27:31,640 --> 00:27:34,520
having worked with these
organizations, do you have a

401
00:27:34,520 --> 00:27:39,840
sense of when that time is?
I would say somewhere in between

402
00:27:40,360 --> 00:27:44,640
Rishi, when we can see
meaningful adoption of agentic

403
00:27:44,640 --> 00:27:48,280
AI, I would, my guess would be
like maybe four to five years

404
00:27:48,280 --> 00:27:51,560
time frame because I already see
things in motion.

405
00:27:51,560 --> 00:27:54,960
So some of the low hanging
fruits like we talked about

406
00:27:54,960 --> 00:27:59,200
those context preserving use
cases, right or reducing the

407
00:27:59,200 --> 00:28:02,520
cognitive loads.
I see some of the tools for

408
00:28:02,520 --> 00:28:06,280
those use cases are already
being used and I think that's

409
00:28:06,280 --> 00:28:07,880
the right kind of starting
point.

410
00:28:08,480 --> 00:28:12,400
So probably four to five years.
And which is, which is OK for

411
00:28:12,400 --> 00:28:14,320
this sector, right?
Because unlike many other

412
00:28:14,320 --> 00:28:18,400
sectors like SAS or digital
companies, I think the stakes

413
00:28:18,400 --> 00:28:22,280
are high here.
So it's better to kind of take

414
00:28:22,280 --> 00:28:25,880
more judicious decisions, yeah.
Now it is.

415
00:28:26,960 --> 00:28:30,840
And it's not an industry that's
known for a rapid pace.

416
00:28:30,840 --> 00:28:34,920
So five to six years probably
will be a good achievement.

417
00:28:35,760 --> 00:28:41,600
So we're getting sort of to the,
to, to the end of this episode.

418
00:28:42,280 --> 00:28:45,840
And when, when we, when I
started thinking about this

419
00:28:45,840 --> 00:28:50,080
episode, I actually wasn't
thinking of this as a

420
00:28:50,080 --> 00:28:56,080
conversation on AI.
The, the keywords that were in

421
00:28:56,080 --> 00:28:59,080
my mind were around capacity,
trust, you know, how

422
00:28:59,080 --> 00:29:02,280
organizations actually get work
done in this new context.

423
00:29:02,640 --> 00:29:06,160
Those are the things that I had
in mind and on the podcast

424
00:29:06,160 --> 00:29:08,600
Carmen, you'll obviously
remember is that we we've talked

425
00:29:08,760 --> 00:29:13,040
about building systems and
scaling them over time.

426
00:29:13,440 --> 00:29:17,360
I think with some of the things
that Sabor has described, it

427
00:29:17,360 --> 00:29:20,200
seems like.
With AI.

428
00:29:20,840 --> 00:29:25,080
And you know, with the help of
AI as well as with AI in the at

429
00:29:25,080 --> 00:29:28,400
the centre, there are probably
some systems that need to be

430
00:29:28,840 --> 00:29:32,560
developed, not just thinking
about AI as a tool as as Subodh

431
00:29:32,560 --> 00:29:36,640
was saying earlier, I think that
that phase is coming for this,

432
00:29:36,840 --> 00:29:40,880
for this industry as well for
our business.

433
00:29:40,880 --> 00:29:43,720
I just want to highlight again
that Subodh has written this

434
00:29:44,440 --> 00:29:47,120
amazing book called Agentic AI
for Leaders.

435
00:29:47,400 --> 00:29:51,680
It's available on Amazon.
We'll also live leave a link to

436
00:29:51,680 --> 00:29:54,000
this book in the podcast
description.

437
00:29:54,000 --> 00:29:57,720
So those of who you're
interested in this topic,

438
00:29:57,720 --> 00:30:01,080
please, please feel free to buy
a copy of the book.

439
00:30:01,440 --> 00:30:05,240
So both thanks for joining us
and sharing your perspective on

440
00:30:06,120 --> 00:30:11,160
on this subject.
Again, great conversation and of

441
00:30:11,160 --> 00:30:15,080
course, a great book as well.
Thank you guys.

442
00:30:15,080 --> 00:30:17,360
I really appreciate you guys
inviting.

443
00:30:17,800 --> 00:30:20,600
Thank you so much All right and
to.

444
00:30:20,600 --> 00:30:23,560
Our listeners around the world,
thanks for tuning into

445
00:30:23,560 --> 00:30:27,400
Sustainability Forward.
Make sure to subscribe to the

446
00:30:27,400 --> 00:30:31,160
podcast on any of the platforms
that you listen to podcasts on.

447
00:30:31,840 --> 00:30:34,840
We'll, of course, continue to
keep interesting episodes coming

448
00:30:34,840 --> 00:30:37,200
to you.
Carmine, thank you again for

449
00:30:37,440 --> 00:30:39,800
joining me.
Thank you, Richie, and.

450
00:30:39,800 --> 00:30:44,120
I hope I will not be substituted
soon by energetic podcaster

451
00:30:44,120 --> 00:30:48,000
soon, so I'll invite.
I'll invite you to podcast 10

452
00:30:48,000 --> 00:30:52,320
years down the line for sure.
All right, take care.

453
00:30:52,760 --> 00:30:53,200
Thank you.