Latest / The Innovation Show / Usman Sheikh — How AI Will Reshape Consulting: The End of the Pyramid

Transcript

Usman Sheikh — How AI Will Reshape Consulting: The End of the Pyramid · The Innovation Show

0:00
Before we begin, I wanna thank our sponsor, Kyndryl, who run
0:02
and reimagine the technology that drives the world's infrastructure.
0:07
We're talking about healthcare, we're talking about aerospace, and banking
0:11
amidst many, many other i- industries.
0:14
But also check out the Kyndryl Institute where you'll find
0:17
unbelievable thought leadership, many of the guests we've had on the show.
0:21
You can find both at Kyndryl.com, Kyndryl.com/institute or .com where
0:28
you can find out more about Kyndryl
0:31
today's guest is a previous guest from a webinar we had recently with the
0:35
Kyndryl crew, and it's great to have him.
0:38
He's been writing lucidly about AI for several years, and particularly
0:42
how it's gonna affect knowledge work and professional services companies.
0:47
was saying to him before we came on air, it was a little bit like that quote
0:50
where Wayne Gretzky was asked how does he always know where the puck is gonna
0:54
be, and he said he skates to where the puck is going, not where it is.
0:57
And that's a little bit what's happened with our guest, because
1:00
he's been talking about this stuff.
1:02
Earlier on, people thought maybe, "Is this gonna happen? Is it real?"
1:06
Now they're listening, 'cause they've caught up with the puck.
1:09
He is the founder of High Output Ventures.
1:12
He's the author of a brilliant site called Framebreak, Usman Sheikh.
1:16
Welcome back to the show
1:18
It's great to be on.
1:19
Looking forward to this conversation.
1:20
Always enjoy these where, like you said, the world of professional work is
1:26
changing, and there were signs last year.
1:29
But lots of people have talked about the death of consulting for a very long time.
1:35
and by no means do I think, like, there's going to be the death of professional
1:39
work, but I definitely do think that the way that the firms were structured,
1:43
the way the work was delivered, is fundamentally going to change, and that's
1:48
what I'm most interested in writing about and talking about these days
1:52
I was having breakfast recently with a senior leader, and I was saying to him
1:56
that there's two concurrent shifts or multiple shifts going on, but the, with
2:00
regards to firm, was saying to him, I, we were having breakfast, so I picked up a
2:04
salt cellar, and I was like, " the firm used to be like this, like an IBM." And
2:09
then I pointed to the gr- grains or the flakes of salt that were on his avocado.
2:15
And I was like, "See the flakes of salt there on you?" I said, "So the
2:18
firm, like IBM, had to break up to be more agile and became modular.
2:25
then eventually by doing that, it invited in lots of competitors, and then I poured
2:30
out some salt on the table, and he was kind of looking at me going, "What is
2:33
this guy gonna talk to me about?" And I was like, "Now it's like grains." I
2:37
said, "But AI is ta- " I took some of the grains and I put them in my glass
2:41
of water, and it's like, it's dissolving some of those grains, and it's leaving
2:44
some, it's leaving only very few.
2:48
in there he went, he was looking at me and was like, " did that work?
2:51
Did you understand what I was saying there?" And he goes, "That's a
2:53
brilliant way of understanding it." And I was like, "Oh, yes." So I was
2:57
like, you've written then about that structure and new structures that
3:02
are needed, and I thought maybe we would start with why was the structure
3:06
the structure in the first place?
3:08
And then we'll get onto where you see it going.
3:10
I think professional service firms solved a-- the way that they were
3:15
structured solved a scarcity problem.
3:18
The firm's most valuable asset was its senior leaders' judgments,
3:22
the senior people at the top.
3:24
But the number of hours in the day were limited and, you know,
3:27
they could only service a certain number of client requests.
3:30
In order to solve this problem, what they did was that they put partners
3:33
at the top and built layers of managers and juniors beneath them so
3:37
that they could scale their judgment across a much wider array of clients
3:41
than any single one person could.
3:44
And I think that became the leverage unit that we see today, and
3:49
that's why the structure existed.
3:51
And more than just multiplying the output of the senior leader, what it also did
3:58
was that the base of the firm was the profit engine of these firms as well.
4:03
So the rate that the firm would charge for the work versus the
4:07
rate which they paid the base, the delta between that was the margin.
4:12
And the wider the base became, the more money the firm made.
4:16
And also the, the people at the base of the firm were actually very profitable
4:21
units of margin for the company, and therefore it expanded its way outwards.
4:28
So two things came out of the pyramid so far.
4:30
One was leverage of the senior talent's judgment, and the second
4:34
was the profit and the margin.
4:36
And the third thing, which I think is not talked about as much, is
4:40
that the pyramid also was a training ground for the next generation of
4:46
leaders and signers at the company.
4:48
So when you have a wide base, you can figure out who is good, who is doing the
4:53
work well, and through a series of error correction, supervision through seniors,
4:58
these junior people ultimately became the people who could one day put their
5:02
name on a document and sign it as well.
5:05
And the beauty of this was that all of this training was paid by-- for the client
5:10
under the guise of the billable hour.
5:13
And what AI does is that it pulls all three of these factors, the leverage,
5:22
the margin, and the formation of talent altogether, which then puts a lot of
5:27
pressure back onto the firm because if they are to adapt, they're going
5:33
to have to figure out how to change a model that has worked remarkably
5:37
well for them for a very long time.
5:39
I love the idea of the pyramid and the base and the understanding
5:44
there that a training ground.
5:46
And then it opens up the question, and we'll… I- I go all, all
5:50
over the place, man, here.
5:51
You know
5:52
Right.
5:52
roll.
5:53
But I also fear that, we've seen this with all these firms, that they're
5:57
not hiring for that base anymore.
6:00
So I was thinking like, "Well, what are they gonna do then?" obviously
6:05
training AI you know, there's some firms bonusing people on using AI
6:10
or punishing them for not using AI.
6:12
But then on top of that, I've seen this concept where if you have a senior
6:17
executive or a senior consultant from one of those firms, they're creating
6:21
a digital twin version of that person.
6:24
Right.
6:24
And
6:24
Yeah
6:25
may be let go in the next few months or years, but there's this digital twin.
6:30
So you have frozen that knowledge at the start, at the top, and then maybe actually
6:35
filling in through outsor- sources of information, new papers, et cetera, your
6:41
future work that you're about to publish in the future, stuff like that feeds it,
6:46
and then you have this, just a- a- a- a- an LLM perhaps that is owned by the firm.
6:54
Yeah
6:55
know that's n- not exactly the right language to put on this,
6:58
but that's the concept that we're
7:00
Yeah.
7:01
There as well.
7:02
Yeah.
7:02
So I think that you correctly outlined two parts.
7:05
One was the digital twins of senior leaders, which we can park on one side.
7:11
And the other part, which I think applies more to the formation base, which is
7:15
that if you stopped hiring as many juniors into the firm, you don't create
7:21
a problem today in the market because you can hire from other firms and other
7:26
people from the stock that exists today.
7:28
But maybe five years from now or seven years from now, the market
7:32
is going to face a glut of that talent which does not exist.
7:38
So I think I really thought about the, the formation problem quite a bit, and
7:44
let's talk about the junior problem first.
7:46
So the first thing that I said was that the base of the the formation was paid
7:50
for by the client in the traditional model because, you know, you learnt on the job.
7:57
Now, when the client, let's say, does not get billed for hours of work,
8:03
and they just get billed for, "This is the outcome that I would like,"
8:06
and you have to figure out how to build that for them, you take that
8:10
abstraction away from the junior base.
8:13
And now what the firm has to do is make an intentional decision as to
8:18
whether they are going to invest into the formation of talent in the future.
8:25
And I'll tell you from personal experience, we have a IT services
8:29
company in the, the portfolio, and this is a discussion we regularly
8:34
have because it just takes a long time to grow that person, nurture
8:40
that person into that pers-- into that person who can sign in the future.
8:44
And the decision now is: Shouldn't we just hire someone from the
8:47
market who is mid-tier and sort of like pay them a little bit more,
8:52
but augment their output with AI?
8:55
And I think this is a very intentional decision that someone
8:59
is gonna have to, to make.
9:01
Because you know, I have friends who are working on building simulators
9:06
for junior talent at some of these firms that, okay, so if you're not
9:09
doing the work, there are simulators.
9:12
In one of our firms, what they make the juniors do is that when the output comes
9:17
out from the LLM, the junior has to assess the output, and then without seeing what
9:23
the senior or the system would recommend, and then they cross-triangulate,
9:28
and they're doing all of this work.
9:29
So all of this stuff is happening right now, but who's paying
9:32
for all of that development?
9:35
It's the firm is paying for that development.
9:38
And I think, like- It's the correct choice for the firm to actually set
9:43
some budget to do this because they set themselves up for durability of
9:48
competitive advantage in the future.
9:50
Because let's say that we do all of this and we skip the thing, five
9:54
years from now, we don't have a new stock of signers, that we can't afford
9:58
new signers or senior people in the market anymore the firm unfortunately
10:04
may have harvested profits for five years, but is going to find itself in a
10:08
really tough spot five years from now.
10:10
So, so I think that's a formation thing.
10:14
The core, if it is that intentionality or talent formation becomes the
10:19
responsibility of the firm and the payoff is much further out, but it's perhaps the
10:26
correct decision to make in the short term
10:29
I see it exactly the same as a sports team and exactly the same as making
10:35
some bets in innovation that m- many don't work out well, but a few may pay
10:40
for all the ones that didn't work out.
10:42
But back to the sports idea, r- because we're in the middle of the
10:45
World Cup, I'm gonna age our episode now, but of course, it's about AI,
10:50
so it's gonna age very quickly.
10:51
But I recently said this about Ronaldo, so Ronaldo, brilliant player,
10:56
absolute unbelievable respect to him for playing so late in his career and
11:00
keep going and be able to perform.
11:03
But there's lo- huge questions over should he have been playing, and for
11:08
me, what that means is he's preventing somebody else from developing, and as
11:12
a result, and this is the choice of ultimately the coach or the management
11:17
team, preventing others from growing.
11:20
And I often think about that.
11:21
If you have a digital twin, say, of a senior consultant, and you're
11:26
relying on that versus to what you said, where I have somebody else going
11:30
through ultimately the struggle that's needed to actually learn, 'cause it's
11:34
through the struggle of learning.
11:35
We were talking about some of the papers we read on AI before we came on.
11:38
I mean, it's a huge struggle to read those papers, to get through, and actually
11:43
new terminology and terminology you're unfamiliar with, but it's only through
11:47
the struggle that you actually learn.
11:49
Right
11:50
one of the things I wanted to say was this concept of the Chinese room that
11:54
we talk about in AI, where you actually, like, push a piece of paper under
11:58
the, the, f- under the, beneath the,
12:01
Yeah, a dork
12:02
and the AI does the work and pushes it back, and you've no idea
12:06
what happened be- after that when it was pushed through the door.
12:11
That more and more, even if you are a junior coming into the bottom
12:15
of that base layer of the pyramid, many are using AI and not going
12:19
through the, the, the discipline and the struggle of the learning.
12:24
And in there, you're like, don't know who actually knows what anymore because all
12:28
you're ultimately selling is the output.
12:30
And it, it's, I, I think the governance of that is huge, and it can be missed so much
12:37
because you think you're doing the work, but it's actually the machine doing the
12:40
work all the time, and nobody's learning.
12:42
Yeah.
12:43
I wrote a post about this, and I called it fluent and substitutable,
12:49
which goes to the exact word that you just did that you use the machine.
12:54
So I used an example of a firm where the, the junior was struggling to r-get to a
13:00
certain outcome, and they were relying heavily on the machine to produce outputs.
13:03
And they thought that it's just some iterations that I need to
13:08
do, a few more button presses, and I'll get close to that number.
13:12
And it doesn't work that way because you've just converted the LLM into
13:18
a casino or a jackpot machine where you're just hoping that the next button
13:22
press is going to give you the answer.
13:25
And while it could happen, I don't think anyone bases their
13:30
paying their rent on whether the slot is going to pay out or not.
13:33
And similarly my advice to that junior was that you have to do the work by hand to
13:39
understand what the machine is doing so that you can actually grade the output.
13:44
And if you do those difficult reps, you actually understand what's happening
13:50
because if you don't, then everyone who can use your same prompt or eventually
13:55
the machine will just absorb your work because if you don't know what's
14:00
happening, the machine is like, "We're both guessing over here," and the
14:03
machine gets more reps at it than you do, the machine's gonna win at the end.
14:08
I was thinking of that movie WALL-E, you know, the, the I
14:12
think it was a Pixar movie, and
14:14
Yeah
14:14
for watching Pixar, but one of the animation studios, WALL-E, this,
14:18
about this robot, and it's a world where humans don't do very much and
14:21
we're all extremely obese 'cause we've outsourced everything to the machines.
14:26
And I was like, oh, well, that's kinda what a lot of people are doing.
14:29
Where I was talking to a, a team recently, and they were saying that if,
14:34
for example, throw this out there, that you're creating policies, huge policies
14:40
for huge organizations, but actually you're almost just the person punching
14:46
in the data at the start, and it's the machine coming out with the policy.
14:50
So nobody actually understands the policy, and everybody's working towards those
14:54
policies, and you're kinda going, there's nuances in there that are gonna be missed.
14:58
There's gonna be so many thi- things missed.
15:00
There's also gonna be lots of stuff spotted.
15:02
But in there, there's a huge threat.
15:04
So maybe so- some thoughts on that happening, 'cause it, it is happening
15:08
all over the world, before we go into talking about business process
15:12
optimization, outsourcing, offshoring
15:15
Yeah
15:16
of the professional firms use, and where that fits into all of this.
15:19
I think when we look at past moves such as, like you mentioned, offshoring.
15:25
Offshoring essentially changed where the work got done.
15:30
It didn't change anything else.
15:31
It basically changed the jurisdiction, lower cost jurisdiction, and the
15:36
work was still done by people, but it was done somewhere else.
15:39
Workflow software, basically RPA and other aspects, made predictable processes into
15:49
a structured format within encoded rules.
15:51
So it changed how you did the work, but within a certain structure.
15:56
And I think when I think about AI, what it does is that it takes this
16:01
reusable system which now can freeze itself from this constraint of rules
16:07
that were around and it changes who is doing the work now, right?
16:11
So it doesn't produce… What I've talked about is it produced
16:15
judgment-shaped artifacts.
16:17
Like, you know, it's-- it doesn't have judgment in it, but it's ju- shaped.
16:21
Like someone who doesn't know what a good brief looks like will probably
16:25
think AI has created a great brief, but someone who's created briefs
16:28
for a living will know, like, "Okay, this is not a bad brief, but it
16:31
made mistakes in these four areas."
16:33
And I think, like, that's how I see AI being very different from the past
16:40
world, because in the old models, it was still when the machine reached
16:46
or when the process reached a certain exclusion that it couldn't understand,
16:50
you always had to revert back to a senior.
16:53
But now that has expanded, and you need far fewer people sort of like doing that,
16:58
and the machine starts everyone off not from zero, but from later in the cycle.
17:03
And I think that's fundamentally different from past shifts because it has direct
17:09
implications on who's doing the work and how the work gets done compared.
17:15
Because past shifts, a lot of legacy co-owners will say fortified
17:21
the pyramid more than anything.
17:23
They became bigger, like, you know, Accenture has like eight hundred
17:26
thousand people, a lot of them in, um, you know jurisdictions like
17:31
India and Philippines, et cetera.
17:33
But AI does not necessarily play out that way because if you just use it as a
17:39
productivity enhancer, um, I think your long-term competitive advantage deflates
17:49
very quickly because everyone will use the same tools to get to the same answer
17:54
You remind me of an old story Clayton Christensen told in, I think it
17:59
was How Will You Measure Your Life.
18:00
He, he talked about the, he called it the ship of Theseus, but the whole idea of the
18:04
ship of Theseus, for those who don't know, is, like, the ship gets replaced plank by
18:08
plank, and eventually you're kinda going, "There's no planks from the original ship
18:13
so wh- what, is it still the same ship?" he talked about, well, Dell outsourced
18:20
more and more of one of their machines out to China, I think it was, with ASUS.
18:25
And ASUS then eventually, as Dell kinda went down the stairs from
18:30
their side, ASUS were going up the staircase from their side, building
18:33
ta- capability more and more and more, and then actually becomes a competitor.
18:37
And we've seen that with BPOs, for example.
18:41
And actually, where you probably see it more, and I haven't seen anybody really
18:46
call this out, is for example, of the European or even a lot of the US companies
18:52
maybe use India, and now the knowledge work in India is absolutely amazing.
19:00
The people coming from India are amazing.
19:02
A lot of the new mortgages here in Ireland are actually Indian
19:05
people who have moved to Ireland.
19:06
So
19:07
Yeah
19:08
Kinda reversals of that, the same type of thing of the outsourcing.
19:12
Yeah
19:13
in there is a huge threat.
19:14
So we talked about the threat of on-boarders, new starters
19:19
in some of these firms.
19:21
But then think about the other side.
19:23
If y- if you've a BPO out in India or what used to be an emerging market,
19:30
and I don't need that anymore,
19:33
Yeah
19:34
ecosystem, a huge house of cards that could come tumbling down.
19:38
Because somebody told me, you know, if you're working for one of these big
19:42
firms in a BPO over in India, you're extremely well paid by local standards.
19:47
You're getting Ubers to work.
19:49
You're living in luxury locally.
19:53
But if that's gone now, not only do you fall, but so does the actual
19:57
house of cards that's dependent on you
19:59
Yeah.
20:00
And it's a major source of employment within the, the country's base as well.
20:06
And a lot of their exports are directly linked to this export-based revenue that
20:12
gets collected back into the country.
20:14
So it has large macro implications for the country as well.
20:18
And there's this large narrative shift of like, okay, well, we have
20:24
all this trained labor, like you said.
20:27
We're going to convert them into these forward-deployed engineers
20:31
and, you know, suddenly they're going to do this other new types of work.
20:36
And the reskilling narrative is definitely there, and there's also the
20:41
conversation that as the professional services work will get cheaper to produce,
20:48
then there will be a lot more work and therefore, hopefully, a lot of these
20:52
people will find employment within that.
20:55
But I think there's two things in that argument which I wanna click into.
21:02
One is like, invariably, if the answers get cheaper, you're gonna have more
21:07
people asking more answers for it.
21:09
So, so that's there.
21:11
But there are two things moving over here.
21:12
One is the demand for that unit of work, and the second is the human
21:18
labor cost to deliver that unit of work, however much is required.
21:24
And so if the demand rises faster than the human labor cost rises, then employment
21:32
will, will rise because you still need more humans in order to fill the thing.
21:37
But if the human cost falls faster than the demand sort of like rises,
21:43
well, then the reverse happens and, you know… And I don't think it's going to
21:48
be clear-cut across the entire industry.
21:50
I think serv-- certain service lines are going to still be heavy human labor
21:55
required, like high-stakes work is not gonna suddenly be handed over to the
22:00
machines because of a lot of things.
22:02
But low-stakes work, which is heavily checkable, yes, I think a lot of
22:07
that work is going to be a threat.
22:10
And the call center industry is a very good sort of like analogy of this.
22:15
There's a firm called Concentrix, and then I usually compare it to
22:19
a new firm called Sierra, which provides, again, voice agents.
22:23
One uses human labor to pick up the phone, and there's a certain cost towards it.
22:29
And Sierra is multiples in market cap value, and its
22:34
revenue is growing exponentially.
22:36
Like, I think they did a hundred million they reached a hundred
22:39
million in February, and they- Crossed two hundred million in June.
22:44
So it's basically the curve is exponential, and what they provide is call
22:48
center agents for regulated industries.
22:51
And there are no humans involved in this transaction anymore.
22:56
And I think, like, that's where even when the stakes are higher, you can
23:01
see the models moving into this.
23:03
This is just not picking up the phone and booking a reservation for you.
23:07
This is someone calling up a bank and asking them a complex question, and that's
23:11
still being handled by an agent today.
23:14
Just goes to show you that it's not necessary that increase in work is
23:18
necessarily going to lead an increase in employment because value capture
23:23
can either go to the person, to the platform, to the client, or to the firm.
23:28
So… And that's the hard part in this conversation that I'm finding.
23:33
Like, who captures all of this gain that is going to occur now?
23:38
I, I really wanna talk about that,
23:39
' But just on that, what I thought about that before is it's almost
23:43
like, I call it this idea of musical chairs, that game we used to play
23:47
as kids, where the chairs run out.
23:50
And you think about each shift we've gone through from digital to cloud to today,
23:58
there's been shifts, but I just don't see where there's enough chairs at the
24:03
table for people to be all employed when AI is taking so much of the value and
24:10
Bye.
24:10
optimizing so much.
24:11
And as, as, as to, as what we talked about before, it's taken away that
24:15
bottom layer, so there's actually
24:16
Right
24:17
at the bottom layer g- starting, so there's no way
24:19
to get onto the layer at all.
24:21
Yeah
24:22
point there, the valuation of the firm is based on usually revenue per employee.
24:27
So one of the examples is Base44, this company that sold to Wix for
24:32
80 million had eight employees.
24:35
One shareholder, eight employees.
24:37
And you go, well, w- if that's the case, and that's happening more and
24:41
more and more, and we're gonna see, as I said s- to you before, not two
24:46
pizza teams, but two slice companies, where there's enough just to feed
24:50
one person who owns those companies.
24:52
And it's a lot of the companies you've invested in, co-created.
24:55
Right
24:57
does the employment lie?
24:59
And, you know, people are kinda going, "Oh, well, I'm gonna get my kid to
25:02
become a carpenter or go into the trades." You're kinda going, "Yeah,
25:05
but who's gonna buy the houses and employ them?" So th- there's a whole
25:09
ecosystem here needs thinking about.
25:13
And the irony, I thought about this, the irony is somebody in
25:17
government somewhere where he's kinda going, "This is on my desk now,"
25:20
and they're typing it into an LLM.
25:22
Yeah.
25:25
Yeah, I think yesterday it, it was Google DeepMind founder Demis who wrote an
25:32
article on X in fact talking about how this is this massive change happening
25:37
and how we need a collective response in how to deal with the repercussions
25:42
that this is going to have on society.
25:45
Because I think the, the ju-jury is still out on is there gonna
25:48
be net employment increase?
25:52
How are we gonna distribute this?
25:54
I've heard arguments on both sides of the, the thing, and usually
25:59
you're arguing your priors.
26:01
So if you want more employment, then you're gonna say, "Of course, more work
26:05
is gonna create more people," because if my margin unit is dependent on that
26:10
unit, then I do not want less because it creates a lot more problems for me.
26:14
But if I'm a technology-augmented firm which is using this, I'm
26:18
going to say the, the, the latter.
26:22
Ramp, in fact which is this card-- expense card plus now accounting software
26:28
company, wrote a paper recently where it found that high growth companies when
26:33
they adopted AI aggressively actually saw headcount rise within the firm.
26:39
But I think, like, what the paper says in one line was that this is
26:43
limited to high growth firms in sort of information technology sectors.
26:48
And so your, your concern is valid, and I think a lot of people are seeing that.
26:54
It's just very difficult because I think jobs are a bundle of tasks, and
26:59
what we spoke about earlier was that if you unbundle the job, there are
27:03
certain parts of the job that sort of get automated away, and the parts that
27:09
remain require you to either have high judgment or the ability to have some
27:14
credential to stand behind the answer.
27:16
You might have a, a, a law background or a audit certificate or a CFA or a doctor.
27:23
I think that provides you position, but someone who doesn't have those
27:29
gets put in a very difficult spot.
27:31
Um, and like the first work that we were doing, sort of now firms need to invest
27:37
in you to build those skills where it's not a byproduct of how I do the work.
27:43
That in itself creates, um, tension within the firm because do you optimize
27:49
for the short, mid, long, or altogether, or how do we do, or it's very uncertain.
27:54
I don't want to invest now.
27:56
Maybe I'll invest two years from now.
27:58
And these are all conversations I'm having with portfolio
28:00
companies, and it's, it's not easy
28:05
You know, I saw this shift when I, I was a professional rugby player, as
28:10
you know, and I went into the working world in 2008, which was terrible
28:14
timing if you look at it that way.
28:15
But I was, I actually purposely went in at that stage because I was like, "Oh,
28:20
well, this is gonna be difficult, and I may as well go into an environment
28:23
that's difficult for everyone," and used the kind of perseverance and the,
28:28
the resilience that I'd built up from sport to work in the working world.
28:32
But one of the places I first saw was I saw salespeople really struggling be-
28:38
because before they'd actually worked as booking agents, or they were esser-
28:43
essentially just minding their clients, and now they have to actually sell, and
28:48
they'd actually forgotten how to sell.
28:50
And before we, we talk, 'cause I really wanna share the shape of the new
28:53
firms, you talk about five different structures, shapes that a, a firm can
28:56
have and the evolution of those shapes.
28:59
But the, the atrophy of those skills if you don't use them, and one of
29:04
the things I see people, and this is why I was talking about that
29:06
mental atrophy, that you may become mentally obese because you're not u-
29:11
actually using that muscle anymore.
29:13
There's a huge onus on people to become constant learners or be lon- constant
29:18
learners, and I know it's one of the motivations behind Framebreaker through
29:23
your writing, my writing as well.
29:24
It's not only for the audience to share this stuff, but it's actually for me to
29:28
actually understand it, and going through the struggle of figuring out arguments,
29:32
does that work, does it land, my
29:35
Right.
29:36
my salt, metaphor at the start.
29:37
D- does it make sense?
29:39
And actually going through the process of that, thinking, sleeping on it,
29:43
thinking about it over several days.
29:45
And when people are using agents more and more and LLMs more and more and
29:50
co-working more and more, the- all they are are this person who shares the
29:54
information that they've just got from the LLM, even to a point of, "Upload
29:58
the PDF, give me the main points." like, when you read a book I, I used to, I
30:04
used to get this from people to go, "Oh, why don't you just use one of these
30:07
summary book- you know, like Blinkist or one of these tools to summarize books
30:12
for you instead of going through it?"
30:13
And I kind of go, Whenever I did that and then I read the book, I used to do both,
30:18
I'd read it and then read the book, I'd pick out totally different highlights."
30:22
Yep
30:22
much of the book was so personal to m- my own experience, my prior
30:27
information that I'd sourced, that I'd build information totally different.
30:32
So if we all use the same agents to produce the same results, to get
30:37
the same highlights, no innovation in there 'cause there's no spotting
30:42
anomalies and information that's appealing to you or something that
30:46
might spark a metaphor for you, and I just think that that's a huge shame.
30:50
And I wondered what your advice was for people who are… 'Cause
30:53
we're all doing it a little bit.
30:55
Yeah.
30:56
towards that.
30:57
What is your advice?
30:58
What, what are you seeing as a best use of the tools that are out there?
31:03
Yeah.
31:03
I like the, the way that you said that, you know, you were doing both.
31:07
I sometimes use that as a filter of whether I want to
31:10
read this completely or not.
31:13
Because also, to be honest, a lot of books these days, especially the newer books
31:18
these days, they fill a lot of pages, but there are a few ideas that are stacked.
31:23
So you wanna figure out whether you wanna go through them or not.
31:27
But I completely agree with you that there are certain books where just the
31:31
highlights take all-- they abstract everything away from the, the fabric
31:38
of how everything stitches together.
31:40
That if I just-- If you saw Alexander's Business Model Canvas, and you
31:44
just saw that snapshot and you say, "Okay, here are these boxes."
31:47
You put your firm on this, and then you try to figure out w-- how to maneuver it.
31:51
I think, like, you know, it's, it's quite shallow, but if you wanna understand
31:55
what is the, the, the core behind each one of this, how do they each
31:59
interact, what are the different parts?
32:01
I don't think that you get that.
32:02
And that's how I, that's how I tend to read these days as well.
32:06
I think certain older books that you'll just read through because
32:10
there's just so much in those books.
32:12
But especially when I read newer books, I do this where I go to the parts which
32:16
is also most relevant to me so that I can pick that up, and then if that grabs me,
32:22
then I'm reading a lot more of the book.
32:24
And but, you know, I, I don't know whether the, the audience will see this,
32:30
but this Mike Kirsten had written this book called Project to Product, and
32:35
now his book just came out yesterday called Output to Outcome, which is very
32:38
relevant to what we're talking about over here because we're moving from a
32:42
society which is was driven a lot by output, like, especially software firms.
32:47
Output was the code and everything, but now we're talking about these outcomes.
32:51
But how do we reorient the firm around the outcomes?
32:54
And he has a structure in the book, and I read a synopsis of it.
32:57
Someone that I really respect r- recommended the book, and here is the
33:01
book, and I started reading it this morning, and it was a great sort of, like
33:05
read so far because it connects with a lot of things which are relevant to me today.
33:10
So that's, that's my process as to how I'm doing things.
33:14
And it's becoming harder because there's just so much to read
33:18
these days that you have to figure out what's really important.
33:21
Seth Godin years ago saying that he, he'd put the summary… You know, the way you
33:26
see at the end of a chapter, it usually goes chapter takeaways, that he'd put
33:29
it at the start of the, of the chapter.
33:31
Yeah
33:32
through learning w- how the best optimization of learning is actually
33:37
you, if you have a, a concept of what you're gonna learn early, then you
33:43
kind of read to almost validate that.
33:46
But if you're led down the wrong path, so this is what I was saying
33:49
about the AI could actually lead you down the wrong path and go, "These
33:52
are the salient bits. This is what's relevant," and then you read it.
33:55
You're missing the richness of it, you know?
33:57
And, and when I write, I, want to bring the jour- the reader on a journey.
34:03
So actually, in a way, you're warming up their mind.
34:06
Like, the way I see it is the metaphor is creating, like, Velcro, and then
34:11
the idea, which sometimes is complex, is what you throw against the Velcro,
34:15
and it, it sticks to the metaphor, and
34:18
Yeah.
34:20
Yeah
34:20
d- you can't shortcut that.
34:22
Like, that has to go
34:23
Yeah.
34:23
that journey, and it's like, when I run a workshop, I, I kinda do the same thing.
34:27
You kinda gotta go through and bring them on a journey, not too long, but bring
34:31
them on a journey where, where it sticks.
34:33
But let's, let's talk about the, the move of what y- the way you s- you
34:38
phrased this, and maybe we'll qualify and share some of what you mean by this.
34:42
So as leverage moves towards an intelligence core,
34:47
Right
34:47
then different structures of the firm emerge, and there's
34:50
five that you identify.
34:52
So maybe let's talk about what, what do you mean by leverage, what do you mean
34:55
by an intelligence core, and then these
34:57
Yeah
34:58
shapes that emerge
34:59
Right.
34:59
Let me go through it quickly.
35:02
We talked about leverage first when we talked about the pyramid
35:05
and traditional structures, which the source of leverage was labor.
35:10
If you wanted to do more work, you would add more labor, and that is how
35:15
this firm progressed or grew bigger.
35:19
Now, the intelligence core is essentially taking the
35:25
learnings of how work gets done.
35:27
So let's make this real.
35:29
A client comes, and they want us to produce ads for their e-commerce brand,
35:34
and we have to put them up on Facebook for them and test it out for them.
35:39
So let's say that we tell the LLM to generate us, "Here's a
35:43
brand, here's all this context.
35:45
Generate some ads for us." So the LLM will push the ads out, and now
35:50
someone with knowledge of both the brands and advertising will say like,
35:55
"Okay, let's edit this, let's try this, and let's put this to work."
36:00
They'll put the ad. There'll be a feedback loop where the ad platform
36:05
will say, "Oh, this ad performed well, didn't perform well." Let's say that the
36:09
ad came back, it didn't perform well.
36:10
We made three tweaks to it, and then the ad did better.
36:14
Okay, so that's a simple loop that I just created over there that, you
36:18
know, you could replace that with tax or anything else, or a legal contract.
36:23
Most of the times, you just log the winning ad variation and say, "This
36:27
ad did better for every, And it's just logged in a library or a best
36:32
practice or a template or whatever.
36:35
An intelligence score is when the output was created by the machine, someone
36:41
with expertise was able to make error corrections to it, and then was able
36:47
to assess what happened with it or not.
36:50
And then they would feed all those error corrections back into a
36:54
machine which learns all of this.
36:56
So, so for all the things that you would discard, this is putting all
37:00
of that context back into the core.
37:02
And now the next time the machine says, "Oh, here's a health brand which
37:07
is targeting the same demographic," the next iteration that the machine
37:11
will produce will be of higher quality because it has some learning feedback
37:16
loops from before that makes it better.
37:19
This is when you don't rely on a person for a source of leverage, you
37:23
rely on the machine to produce this artifact which abstracted that thing.
37:29
This also goes very close to your digital twin example that you used earlier
37:34
because, you know-- But there's a key differentiation I wanna make here.
37:39
Digital twin is a snapshot of your state of mind today, and that
37:46
decays at a really fast pace in the world that we live in today.
37:50
The loop that I'm talking about is this constant flow of information and
37:56
work that is happening at the firm, error corrections that are happening.
38:00
And let's say that we've never worked with a shoe brand targeting women in
38:04
Europe, and today we're gonna do that, but we haven't created ads for it.
38:08
So someone is going to set that up first, do that, put it into the thing, and
38:13
the core becomes smarter and smarter.
38:17
Once this works really well, and you can start to orient the entire firm using
38:23
this to ask it, "Here is the client. We produce this outcome. Can you produce this
38:29
for me?" And it produces output for them.
38:32
This loop creates, the feedback loop happens, and the firm changes
38:35
structure because instead of adding an additional person to the team, you
38:40
can scale work depending on how much demand you can generate and how quickly
38:44
you can verify the work coming out.
38:47
And the firm could potentially morph from a pyramid to what I'm calling a ring.
38:53
And the ring would have the intelligence core at the center,
38:57
and then you would have this new structure of people around it.
39:01
And in a nutshell, that's this transformation.
39:03
And I go through in, in detail for how it could happen to various models.
39:08
But I think as an abstraction, this should provide you with an analogy of
39:12
what I mean by the leverage, intelligence core, and how this new shape comes about.
39:19
It's funny you were saying there about the, have a snapshot of
39:22
someone's, essentially their mental quiver at a certain point in time.
39:29
But what you see also is, you see people get to a certain level in organizations
39:33
and actually stop learning, and actually thinking, "I, I have it now. I'm here
39:37
now. I, I, I've, I'm in port." And you're kind of going, well, there
39:40
is no safe port or safe haven with
39:43
Yeah.
39:44
anymore.
39:45
You have to consistently learn and find a way to learn, which is why learning
39:49
how to learn is, is so important.
39:51
And unlearn stuff as well because I think that's where, where people have dug their
39:57
own grave in a way with AI, in that y- you gotta be the one who's constantly
40:02
provoking yourself to learn, and it, it's uncomfortable, and it's a pain.
40:06
Like, I would say, like, some of the papers we read are boring.
40:09
Like, they are tough academic reads.
40:12
They're tough to get through.
40:13
But it's only by doing that that I actually can learn and, and
40:16
challenge your own thinking as well.
40:18
But you, you mentioned there, so the, the shape changes.
40:22
The, the gain then shifts.
40:26
And the, so you, you were saying earlier on about, okay, the, the, amount of w-
40:32
effort I put in versus what's output is, is, is my profit, essentially, my margin.
40:39
But there's a surplus created by AI, and then maybe the client's
40:44
kind of going, "Hey, wait a second.
40:45
You've been learning on my dime.
40:47
Right.
40:47
created value on my dime."
40:49
So how do we partition out this new value that's been created
40:53
essentially by this, this new party, this new entity, this alien AI?
40:59
Where, is my share in that?
41:01
Yeah.
41:01
And it's happening.
41:04
You know, I call it the AI discount, which is happening.
41:07
And, you know, firms like-- if firms like KPMG can go to its auditor and
41:11
say, like, "Hi. So I think that I should get a discount on my audit,"
41:14
which was highly reported it, it shows that clients are definitely thinking.
41:20
So let's talk about the surplus.
41:22
So we now, let's say, with the help of this technology, can produce
41:27
those outcomes using less hour, less time to produce the same output.
41:31
In the short run, let's say clients don't adapt, but ultimately, clients will
41:35
start asking the question that, "Hey, you know, you must be using some sort of
41:40
augmentation, so I should get a discount."
41:42
And so the surplus drains in a few ways, but there are two primary ways.
41:47
One will drain to the client, and it will drain to the client when
41:52
the work can be easily checkable and the stakes are relatively low.
41:57
And that work drains very quickly because the client's like, "You're
42:01
doing this, but I could get this model to do this, and I can also test it."
42:06
So, you know, it's only when the stakes rise where the cost of
42:11
being wrong is wrong that they'll come to you, but they'll still
42:13
argue with you on this discount.
42:17
The second leakage is now you have the new cost of the business is using a model, a
42:25
frontier model, a harness, or a control plane to build-- get your answers out of.
42:30
Basically, you have to use-- either you use your own platform or you use someone
42:34
else's platform to generate these answers.
42:37
And that platform owner, depending on their position, is also going
42:42
to extract their rent out of the surplus because they're like, "Hey,
42:46
you used these many tokens. Guess what? You owe me this much money."
42:49
Or we have a firm that I'm working with which sells something called
42:54
routines, which are loops that they sell to small businesses.
42:58
And they say, "We don't charge you any difference on top of the token cost,
43:02
but every time you run this loop, we're going to charge you a small amount of
43:06
money." And that's again an example of rent extraction coming out of to the
43:13
platform that you're losing money to.
43:16
And over the last two weeks, there has been a lot of talk by the CEO of of, of
43:23
Palantir and lately Satya from Microsoft, which are saying that you need to become
43:29
sovereign in how you deal with your data, how you deal with the models, how you
43:34
host them, and you've got to do this.
43:37
And while both of them are arguing from their priors because they're both
43:40
saying like, "Listen, you could use the Microsoft platform, or you could
43:43
use the Palantir platform," but they're really scared that platform capture
43:48
happens from the frontier models who will then sort of like cut them out,
43:52
and that is going to be the challenge.
43:56
And that's where it leaks out, where the rent Is slow to leak out is where the
44:04
output does not have a quick answer back.
44:08
Like, let's enter China.
44:10
I don't know.
44:11
Or let's launch this new product.
44:12
Will it be successful?
44:13
I don't know.
44:14
And or we'll find out in a couple of years, and the cost
44:18
of being wrong is really high.
44:20
In those situations, which is traditional MBB strategy work and all of this, I
44:26
think it's going to be slower or let's say audit companies for Fortune five hundred.
44:31
I don't think that their rent will go very quickly because
44:35
very few people can do that work.
44:39
And so your position in the marketplace is going to depend or decide on
44:44
how much of that surplus you keep.
44:46
If you're competing in a place where you get answers back very quickly, the work
44:51
can be checked, the stakes are relatively low, you're going to be in a world of pain
44:56
unless you figure out these locks that I talk about which can defend the business.
45:01
Otherwise you get competed or deflated away in this environment
45:07
I, I'm loving your new vocabulary, man.
45:09
You're, you're introducing this value drain and et cetera.
45:14
These locks, almost like a moat of the future.
45:17
A lock is smaller than a moat because the companies are smaller as well.
45:21
And, and to that, that point, maybe we'll finish on this, so if delivery of a firm
45:27
moves from labor and th- that bottom base, the, the, the structure of the org like
45:33
that, to compute, computation essentially,
45:36
All right
45:36
how does a firm intentionally produce the next generation?
45:41
And remember when we did a webinar recently with Kyndryl and
45:44
Ismail Amela, he mentioned this.
45:46
He's like, going, "Imagine you have a review, a, a feedback review,
45:52
and it's with, it's with an agent, and you're treating this agent like
45:55
it's literally like a colleague."
45:57
Right
45:58
it came to mind when I was thinking about this, well, how, how do I both
46:03
create that new bench for the future?
46:05
'Cause actually, when you look at books like, for example, Stall Points, a book
46:09
about when companies get to these points of stall of revenue, what was the failure?
46:15
And of companies, it was an internal reason.
46:20
It wasn't… 13% was actually regulatory change, sociopolitical
46:24
landscape, stuff like that.
46:26
was internal.
46:27
One of them was a, a talent bench shortfall, so no future talent that we
46:33
talked about, that academy, like the
46:36
Yeah.
46:36
story, that was able to carry the new strategy for the future.
46:40
So you may come up with a new strategy.
46:42
We saw this in the shift from analog to digital in many companies.
46:45
For example, media companies, nobody had socio-media- social media literacy.
46:50
They weren't digital natives.
46:52
So you had this period that was this messy middle of transition.
46:57
What are we gonna see in an AI world with that, particularly with
47:00
professional fir- service firms?
47:02
I think in the short run, it's going to be really difficult for
47:07
smaller firms to build this bench out because to build a bench, you
47:14
essentially need to make an investment.
47:16
And when the investment was subsidized by the client you were fine.
47:21
Like, I have lots of friends who run very large software companies or you know.
47:27
But the way that it worked was the junior came in, they error corrected, they
47:31
worked with-- They figured out how the supervisor made the, the adjustments and
47:35
the error corrections, and ultimately they got harder and harder work, but
47:39
it was all paid for by the client.
47:41
Now we're talking about a world where that assumption is no longer true, and the
47:46
firm needs to make an intentional decision whether or not to invest in a bench.
47:52
And I think mechanically it is very possible.
47:56
I have friends of mine who are running a simulation company which provide
48:00
simulation tools for companies to at least get juniors started off, you know, from
48:05
zero to one and run through the simulator.
48:09
They allow you to quickly get started.
48:12
But I think, like, the next layer on top is something called Shadowlands, and
48:18
I've, I've written quite a bit about this.
48:20
But essentially, mechanically, it is very possible in this new world to do this.
48:24
There just comes a cost to, to get it done.
48:27
And I think that's going to be the true decider as to whether or not
48:33
leadership today has the incentive to do that if they're optimizing.
48:39
Because let's say that you are in a partnership, and you're gonna leave
48:43
the partnership in three years.
48:44
Are you going to invest in developing a brand-new set of talent?
48:49
And I read something recently about Cognizant and some
48:53
of these large BPO firms.
48:55
They're actually making some headways into this by making very large
48:59
investments into this pool today.
49:03
And I'll be writing about that soon.
49:05
But I can see firms intentionally sort of like betting with us because you have
49:11
to take margin, and especially for a publicly traded firm, that means you're
49:16
gonna make less money today, but hopefully you're gonna pay off it in the later day.
49:22
And this all comes down to incentive structures as to whether firms are
49:25
gonna make this decision or not.
49:28
And fortunately the, the onus therefore falls on the junior to try to figure
49:34
out new ways how to run apprenticeships.
49:37
But within the firm, it is a incentive discussion more
49:41
than anything else over here
49:43
it's a funny time and I, you know, I always think about that term, the
49:47
French term, après moi le déluge, which means after me the flood.
49:51
So you have partners in firms coming towards retirement.
49:55
no incentive for them to go and re- re- reinvent, reshuffle,
50:00
Right
50:00
the entire organization.
50:03
So it's a really messy, messy time, and thank you for providing
50:09
shedding some light on it for people.
50:11
For people who wanna find out more about Framebreak, about your work,
50:14
where's the best place to find you?
50:16
Yeah, I think it's framebreak.com and I write weekly over there, and a lot of the
50:22
concepts that we talked about over here are discussed on the blog in great detail.
50:28
I sometimes write on LinkedIn as well, and those are the best places to connect.
50:33
And I look forward to sharing more in the coming weeks as I get closer
50:38
to releasing a lot of the thesis work that I've been working on
50:41
And for those who enjoyed this, who learned from it, hope you have.
50:46
I'm also learning.
50:47
I told you I'm putting these building blocks in place, learning,
50:50
looking back over old, like, really early books like that Design Rules
50:54
books really helped me understand.
50:56
More difficult to read.
50:57
I told Kim and Carliss Baldwin that it was difficult to read, but really
51:01
worth the struggle to get through.
51:03
But the good news is, for me anyway, I'm gonna be collaborating
51:06
more with Usman in the future.
51:07
We might have a series, special series coming out or podcast, standalone
51:12
podcast, but we'll be collaborating in some way in the future, and
51:15
really looking forward to that.
51:16
Finally, wanna thank our sponsor, Kyndryl, who run and reimagine
51:20
the technology for the world's leading businesses and industries.
51:24
They work in aerospace, they work in healthcare, keep propping up
51:27
some of those massive industries, and it's a pleasure to have them
51:31
as sponsor of the Innovation Show.
51:32
You can also find the Kyndryl Institute where you'll find brilliant authors
51:36
like Usman, and many of the guests that we've had on the show before.
51:39
You can find that at Kyndryl.com/institute.
51:43
It's been an absolute pleasure having you on the show.
51:45
Author of Framework and founder of High Output Ventures, Usman
51:50
Sheikh, thank you for joining us
51:52
Thanks for having me Aidan