Subscribe
Share
Share
Embed
Welcome to The NeuralPod.
The 0-1 machine learning podcast.
Chatting with ML Leaders, Researchers and Engineers who've built models, systems and products 0-1.
Darminder: Fundamentally, the, the,
the idea of leveraging it is to give
you, you know, some form of benefit.
That could be time efficiency, could
be revenue creation, generation, which
actually, you can measure that's,
you can see the impact being created.
and this is nothing new a lot, you tend
to find technology's not the problem.
It's usually the process,
the culture, the people,
Chris: So, uh, dam Inda,
welcome to the neuro Pod today.
It's great to have you on today.
Darminder: Thanks a lot Chris.
Thanks for your time and
thanks for inviting me.
Absolutely.
Chris: we, we've known each other, going
on six, seven years now at, at at least.
I'll place you back into your role at
Fujitsu and, your background is in, data
science, ai, particularly in, critical
infrastructure and defense at, Fujitsu.
And since moved on to, HCL
Tech, around 10 months ago.
yeah.
Would you just like to give
people a quick intro, introduction
to your background and, yeah.
Your career so far.
Darminder: Yeah, absolutely.
I mean, my sort of career started, as
a, as an engineer, way back in time.
And, you know, I had the opportunity
to, do an engineering placement
with, kinetic at the time.
And they were doing some really
good stuff and giving, graduates new
opportunities to come and work for them.
So that sounded like a good idea.
And, was kind of like my, first stab at
understanding, you know, how to model
the world around you in different ways.
you know, in, in the, in, when I
started off, you know, data itself was
a, considered a, a very good asset.
It was a, a nice to have.
you know, what to do with that data was
still a question being asked at the time.
So, you know, fast forward to now, you can
fundamentally see the main differences.
Everything, with AI transformation,
the need for good quality data,
the need for understanding your
data estate, all comes to play now.
So I would say.
I've had a career which has
taught me, how to value data and
Chris: Hmm.
Darminder: and then basically
how to extract value from data.
then you've gotta tie it with a, a
good question, a good use case to
kind of deliver those impact, so as
I, as I've, as I've kind of grown,
over the years, you know, skill
sets, around how to articulate.
key messages, what those
Chris: Hmm.
Darminder: look like, what they
are, how should, how they should be
defined, how do you communicate back
those outcomes to your stakeholders?
And then really work on delivering
that through, good, good data science
and, and, and fundamentally, you
know, AI as well at the same time.
Chris: Yeah, the market's,
changing rapidly, I would
say every six months or so.
And I think just to tie it back to, to
some of what you said there, I think
you've got an interesting career.
You've, you've done your doctorate at
UUCL, but then, also studied, LBS and,
you know, having that blend of research
and also as you say, asking the right
questions, how has that business tied into
your research background helped, you lead?
You know, AI initiatives, particularly
when dealing with stakeholders
that are not as technical as you.
Darminder: Yeah, I mean, the opportunity
to actually, learn, some really good
business skills at London Business School.
So something along the
lines of negotiation, change
management, innovation.
really taught me how to do
certain things really well when
it came to senior stakeholders.
And fundamentally, you've gotta be very
clear upfront about what you're trying
to deliver with senior stakeholders.
and, you know, how do you
break down that technical.
technical knowledge that technical
understanding in, in layman
terms is also very important.
So, having, having that kind of
clear focus is really good upfront.
And, you know, that comes from,
a, a mix of both listening to your
senior stakeholders, understanding
their challenges, their pain points.
And fundamentally trying to work on those,
on those basis to, to tie up, your use
case, what you're trying to propose,
and how you're going to deliver that.
which is quite, quite an interesting
way of packaging all of that into, You
know, coming from a, a, a technical
background, but understanding how to
communicate from non-technical people.
And, and I think you do need a,
a blend of those mix of skills.
when you talk to people, quite
senior, in, in, in enterprises.
Chris: Yeah, and I think you, your
background obviously been at KPMG,
Fujitsu now, now HCL, you know,
huge consultancy aspect to what
you're doing on a day-to-day basis.
And I think that, as you say,
having that business acumen and,
and combining in a technical,
technically the right way, is key.
what do you think?
The, some of the common threads that
would run through, you know, your
consultancy career and, you know,
working in places like Fujitsu and KPMG.
Darminder: I think one of the
key things you've got to realize
is, the ability to give you, some
flexibility when you come to sort of
problem definitions and understanding
what the challenge points are.
you've got to, present, clear ideas
and innovative at the same time, I
think a lot of senior stakeholders
are interested in understanding
what the state of the art might be.
you've gotta base that on a
vision on, where they are now
and where they want to get to.
the other common thread is really
having, a really good team around
you, I think, because, as you'd
understand domain expertise is
really very important in this space.
And, my background is, as you know, you
said it's coming from a public sector,
government sort of financial side.
But if you're working in sectors, which
are, you're unfamiliar to yourself, then
you do need people who are gonna give you
knowledge around what those domains are,
their knowledge around what it means,
what are the cost drivers, what are the
kind of the drivers that really impact
those industries and those domains.
So
Chris: Hmm.
Darminder: that team around you to kind
of, give you that sort of visibility.
And education around those
sectors are quite key,
Chris: Yeah.
Darminder: to give that backdrop.
Chris: And I think, obviously you've
progressed and you're now into, quite
a high level leadership position.
You know, there's lots of, people who
are trying to get into the market now.
I think,
But, you know, what, what advice could
you give to essentially someone junior
or, or, you know, who's trying to try,
would try and break into your team?
What is it you, you look at?
Darminder: I think, know, when I, when
I'm sort of doing a lot of interviewing
and, and, and interviewing candidates,
I, what I'm looking for is more about
someone who has got a growth mindset.
So, you know, this, this, this space that
we're in at the moment is gonna be a very
fast changing, it's fundamentally gonna
transform the way that we, we deliver
work and, and do work in the future.
so we've, we've, we've gotta embrace
those coding generation tools, as you
mentioned, or any type of AI tool.
and understand how we can work with
it to deliver, know, better outcomes
quickly and more efficiently.
but one of the things which I think
is really important as well is, you
know, really defining what your USP is.
And that's gonna be very important.
It's gonna make a lot of us focus on.
Key areas which we can really,
really become better at doing things.
And that's gonna be a both a mix of
both soft and technical skills and
know, for example, no soft skills could
be like anything to do presentation.
How do you, how do you network?
for example, you know, how do
you, present yourselves and
how do you articulate ideas?
'cause AI will fundamentally deliver those
answers that you're seeking far quickly.
But it won't be able to step into
a role where it will give you or
present you in front of people.
so that's kind of gonna be very important.
And I, I look for that kind
of skill sets and candidates.
They fundamentally have maybe done
a, a small project or a pet project,
which, they've kind of, understood.
the AI gives you those benefits and
articulated that and understood what
the benefits, what those benefits
are, and are able to communicate
what those benefits are clearly.
and as I said, growth mindset.
I think you, we, we are all in the, we're
all gonna be under the, the radar here
about how we grow ourselves in the future.
you know, and having that mindset
about how, what, what can I do next?
What's the, the next,
next tool I can leverage?
is there something fundamentally different
to how I've done something before?
How can I change that, in a different way,
that delivers a better, better outcome?
So these kind of things, which you
are, you, I think what we'll have
is you'll have more time to do
that in the future, I think that's
gonna be very, very important.
Chris: You are, moving
on to AI and defense.
I know, it's a topic that is maybe doesn't
get as much of the, the limelight as,
as, other technologies and, breakthroughs
because, for, well, for various,
reasons including confidentiality, but,
let's talk about that for a second.
What, what unique challenges of deploying,
AI models in in defense have you faced?
Darminder: Yeah, I think that's, a really
interesting question 'cause, the, the,
one of the, one of the unique things about
defense is that you, you, you, you have
an ability to, work on different types
of AI problems, for particular use cases.
everything from.
You know, deploying AI at the edge,
to something which is more attuned
to, back office processes, that you
might find in an, an enterprise.
you know, both, both, both types of
deployment modes are quite different.
You know, the, the closer you
move to the edge, more isolated
the AI becomes because of the.
The lack of perhaps network negativity,
the environment that it's been deployed
in, even to a point perhaps, you
know, adversaries will be impacting
those algorithms in some way.
So all of those challenges present,
really, really interesting.
Modes of working, and makes you think
about how can I, how can I cater for
those challenges, in an efficient way?
fundamentally it's all about
giving the right answer to the
right person at the right time
so that they can make a decision.
And if you go to the back office side, all
about, in my view how you can actually,
and this is quite common in what we've,
what we find in a lot of our enterprise
use cases, you know, bring in AI tool.
To someone.
it's quite easy to do now.
it's the idea of then how you can
train that individual to leverage
that tool efficiently, to give time
back to that individual to create
those time efficiency savings.
and, you know, it's, it's a pro, it's
a, it's a process transformation.
So essentially where, where
you, where you start to deliver.
A transformation in one area
might create a bottleneck
somewhere else in the enterprise.
that becomes apparent to a lot of
people then we are not making as
quickly as decisions we used to.
is that the case even though
we're, we are speeding up the
work further, further downstream.
So I think a lot of When I, when I was
in defense, that that change management,
how you do that efficiently, how do
you bring people along and how do
you, how you do that responsibly that
the, the risks are also mitigated.
again, is quite a, a unique challenge
and that's kind of transferable to
lot of enterprise use cases now.
from that side.
Chris: Yeah.
And you know, in terms we spoke
about having the right data
at the start of the podcast.
I think that, you know, correct me
if I'm wrong, but my impression of
defense is that, you know, it's a highly
complex and also confidential data.
Darminder: Yeah.
Chris: how, how do you handle the data in
the right way where there's a lot, lots
of different considerations involved.
And then how do you, second part
of the question is how do you
take people on that journey?
for the, you know, you can have the
best model and, and tool in the world,
both someone's that using it then.
Darminder: Fundamentally, the, the,
the idea of leveraging it is to give
you, you know, some form of benefit.
That could be time efficiency, could
be revenue creation, generation, which
actually, you can measure that's,
you can see the impact being created.
but it, but, you know, I, and, and this
is nothing new a lot, you know, you tend
to find technology's not the problem.
It's usually the process,
the culture, the people,
Chris: Yeah.
Darminder: In tune with this journey.
And the, the challenge is that this
technology is moving quite fast.
There's always new
innovation coming through.
So, a question for a lot of
people is how do we keep up with
ourselves up to date with that pace?
and what does it mean for our enterprises?
What kind of processes do they change?
so it, it's, that's kind of that cultural
side because you do need to move quite
quickly and agile in this, in this sense.
Chris: Yeah, because you know, let's
say you adopted copilot a year ago.
we have advances with other, tools,
that may not be the best tool to use.
Now is do you, do you see people
adopted it and then they've,
by the time they've adopted it,
they've already been left behind.
Darminder: a lot of enterprises
have already made or have made
previously, historically, a lot of
investment in the, the infrastructure
that they want to do, want to use.
So, that could be either a, a Microsoft
alignment, an a DS environment,
or a GCP or Google environment.
And, for a lot of enterprises, that's
usually quite a hard challenge to
go and shift for something you've.
That's who invested a lot of
time and effort and money in
to move to, to something else.
So you tend to find that the AI tools that
are, that are released in your ecosystem
are the ones that you tend to leverage.
So you mentioned copilot, right?
So when copilot came out, I
think everybody realized that
it, it was something different.
it might not be perhaps.
and, and that depends on your use case.
at the end of the day, a lot of,
a lot of these tools which you are
seeing in the marketplace are, are
doing two, two things fundamentally.
One is, really focusing on and
innovating around that customer
experience, engagement and making
that tool as, as sticky as possible
with users and the user base.
And, and, and fundamentally, number two,
leveraging those tools to actually learn
about how users use those tools and, and.
You know, that data in the, in, in,
in that, in that way to make fire that
further refinements and improvements.
so it's, it's a, it's a real mix and I
think, What We'll, what we'll see is,
is as the, the models, the foundational
models become more and more intelligent.
And, you know, we, we, every
foundational lab will tell you that
they're aiming for some form of a GI.
and, you know, definition of a GI is
still, not, not defined, but what, what
we'll see is what we'll see, what we
will see is more, more intelligence
being baked into the model, you know.
Allows users to fundamentally, apply
their domain expertise and the tools
with the model to do, to deliver
further work and further, innovation.
and I, and I see that model, I would
say that model is, the emergence of
that type of model is, is appearing
now, whether you're in your existing
ecosystem or you are going to be using.
an open AI or a clawed or whatever.
and you know, the, the,
the great use case where.
that model is now fundamentally being
shifted and changed is everything
to do with software engineering.
You know, the, the, the, the, the
idea of how we develop software
prototype quickly release, features,
has all changed considerably.
developers could be a one man team.
We'll have developer profiles
doing everything from.
software generation to
test and, and release.
so as you know, you, you can sort of see
how that model could reflect, any other
type of enterprise process in the future.
and this is why it's interesting why these
financial models are trying to get as much
enterprise traction, you know, use cases
with their models as much as possible
because they know that's, that's where
their model will slowly replace maybe the.
Other there incumbents
in this marketplace?
Chris: Yeah, I think that segues nicely
into, what you're working on now at HCL.
a big part of that is AI transformation.
do you wanna just rero to people,
what kind of things you work on
there and, what kind of technologies
your team work on and projects.
Darminder: Yeah, absolutely.
So the, our labs are fundamentally.
front and frontier, everything.
We are leading edge, AI capabilities.
and you know, the, the role that we
have in the labs is fundamentally
as one, as a trusted advisor
that we give to our clients.
you know, clients come to us because
they have a Pacific challenge in.
A, a pain point that
they need to be solved.
Chris: Hmm.
Darminder: were able to, break
down those, those barriers and
unlock the value through a mix of
both, understanding perhaps the
right level of technology to apply.
and, you know, one of the, one
of the, be one of the benefits is
that we are technology agnostic.
We don't really drive a particular.
Type of model or a type of ecosystem.
you know, having the benefit of actually
seeing what's happening across the
whole of that piece, a whole of those
marketplace, you know what's being
released across different vendors and
what's what's gonna happen next gives
us the ability to move quite quickly and
bring that sort of, best of expertise
to our clients so that they're able
to make informed decisions, in their
value sort of transformation journey.
so a couple of things that we're working
on at the moment is, fundamentally we're
everything about trying to, encourage
our people within HCL Tech think about
how they can apply these new emerging AI
coding tools or AI tools in their work.
and we bring those use cases
to kind of demonstrate to our
colleagues how that can be done.
we're also looking at sort of, you
know, that everything to do with how
do we get now agen AI into production.
I think we've gone past a phase
where a lot of our clients are
doing pilots and experimentation.
they're into the game of
actually trying to scale.
Chris: Hmm.
Darminder: and bring these agents to scale
and, and productionize these solutions.
So, and, and that, that presents us a
different set of challenges, which we
are now working with a lot of our clients
on, and, and understanding what those
best practices are because I think once
you can sort of outline best practice,
what does best practice look like, it
becomes a framework that you can then.
Communicate with and actually guide
your clients on a journey with as well.
So we are, we're doing a lot of
that, interesting work at the moment,
which is really, really, really good.
'cause, it's nice, it's interesting
to see how you've done things before
and how that can be changed by now.
The use of ai, in delivering the
same sort of similar outcomes.
Chris: Trying to mutual friend.
Actually, Gil Gilbert from, Accenture
is about, about a year ago, and he was
talking about how with agents, he was
looking at, you know, a process if it
has, you know, eight, eight layers,
eight processes in, eight steps in
that process, how they can knock out
one of the processes with, an agent.
that, that was over a year ago in terms
of, You know, agents, where, where do
people start with it in terms of, you
know, maximum ROII guess because it, a
lot of people have e experimented with it.
I think it's an old stat, but, have
a business school brought out a,
a, a publication a few years back
saying, you know, 89% of generat
generative AI projects fail.
where can you get the, where
do you see the BS OROI with
agents in terms of projects?
Darminder: So I think it's a,
it's, it is a very interesting
question and I think you'll have a
lot of people giving you different
interpretations of what answer might be.
if we were to look at some of our
really good use cases that have.
A huge difference applying
Agen AI to those use cases.
They, start from, a process
challenge, perhaps, a part of that
process, which is, which is acting
like a bottleneck, for example.
and you've then, you've gotta,
you've gotta understand how those
processes, are essentially are linked
together to deliver the end to end.
So you've gotta have regular visibility
what the end-to-end, process looks like.
You know, tho those, those,
those use cases which make,
the really good difference are
the ones which, do two things.
One is, they, they, they, they allow users
to interact with the tool seamlessly.
so one of the, one of the challenges
typically when you use an AI agent is.
we need to sort of, challenge our
users or perhaps, somehow articulate
to our users that, you know, just
don't use it as a, a very simple
search engine or, or a chat bot.
think about how you can utilize
your agent to, deliver much more
meaningful work to yourself.
So, and you've gotta have that
construct in your mind about.
is if I deliver a, a research article to
my scenic stakeholder, it, it, it involved
me doing, some research on the web.
involved me maybe looking
at some of the insights.
It involved me looking at, trying
to align those insights to really
good KPIs that my, my senior
stakeholders are looking for.
and then it then senior
stakeholders don't want a.
A one page brief.
They want maybe a, a small
paragraph that summarizes key,
key insight and, and key value.
and, and, and what, what, what the,
what that means for them as a, as
an individual or for their business.
So just described to you four
or five types of processes, and
Chris: Hmm.
Darminder: I think we can, we can
start to think about agents as being.
Kind of like our co-creators in a way
where each of those five processes
could be articulated via some sort
of, giving the agent some sort of good
instructions to leverage it, the right
data to look at, perhaps some of their
processes, could be automated by, by
running, you know, very, very specific,
software functionality in the background.
and then, then it's all about how you
orchestrate that across the whole piece.
And I think, If there's one
innovation where, ROAS really made a
difference, it's, it's, it's really
where the orchestration has worked
really well the orchestration has
allowed, my query, my prompt to be
understood clearly and succinctly,
Chris: Hmm.
Darminder: and for it then to
drive the relevant agents to
go and deliver that outcome.
and of course.
It is not gonna happen from the offset.
It does require refinement.
It does require a little bit of tuning,
I think the idea is that if you can get
to a point where you can do that and
measure that, that process runs repeatably
and is accurate over and over again.
it, it does two things.
It, it delivers ROI and, and
it gives you that trust that
you can give back to your user.
And now that trust is gonna be very,
very important when we get to a
point where we can just leave these
agents to work by themselves and
not worry about what's gonna happen.
Chris: Yeah, I think with the,
the invention of, open claw trust
and observability and, yeah, a lot
of teams are focused in on things
like long horizon memory where the
agent can go off and do lots of
different things without you knowing.
And we we're starting to see,
some good things, but also some,
some horror stories from that.
Darminder: Yeah.
Yeah.
I mean, you mentioned a good point there.
I mean, we started with a
conversation around data.
Now the data requirements for agent
orchestration are slightly different
from we used to traditionally
assume with, you know, predictive
analytics or machine learning.
the need for an agent to remember,
what it was doing a minute ago
compared to perhaps what's important
to you over a longer time period.
Requires two different data constructs.
so how do you bring that into play?
the question is also, it involves a
little bit of different investment
around what you put into play.
and I think one of the good things is
that a lot of the technologies that
we've started to use, you know, like
for example, in the Vector data, the
vector database is for rag, are still
Chris: Hmm.
Darminder: and applicable
to the agent layer as well.
but just for a different purpose.
And, you know, we've always had a
difficulty finding the right use
case for knowledge graphs, right?
We, I, I had a, I had a really hard
time trying to convince a lot of
senior stakeholders that knowledge
graphs would be a really good thing
for them, you know, trying to connect
their data into, a, a relationship
diagram, with, with relationships there.
And now you can see the use
case for that is very prevalent.
The agent layer.
Because they need to understand
how data, how your data is
connected and how it's relevant
Chris: Mm.
Darminder: it links,
across longer context.
so, you know, I think yeah, gen AI is
kind of bringing all of these key facets
of areas that we've all explored in our
careers more and more, you know, more
prevalent to that type of way of working.
So I think that's a good thing in my view.
Chris: Yeah.
And you know, for, for some organizations,
agen AI may be the third step, and
they're still yet to do the, first
step and just, you know, t tying
it back to, you know, cer certain.
Projects and projects just fail.
You're obviously someone who's been in
the consultancy game for a while, and I'd
like to, I think it's got the, the, the
battle scars of implementation and pushing
these things through into production.
What the biggest mistakes
you see organizations make.
Darminder: I wouldn't say there's
a, there's a direct answer to this
question, but, a couple of things
which I think have, have stood out.
we, we, we've, we've, we've talked
about the kind of like the change
process management that needs to
take place, in, in conjunction with
any sort of AI initiative, right?
so that's, that's gonna be important.
So I think you've, you've got to
understand, Where, where, where the
AI will have the biggest impact.
And that could be both an impact to people
and an impact to how you deliver services.
and, and fundamentally, we, every, every
enterprise is doing this for two reasons.
Deliver better customer
experience, or, or revenue creation
and, and revenue generation.
the, the other thing that I've found is
quite, sometimes quite hard to do is, is
really articulate what the problem is.
Very, very, very well.
and sometimes you do have conflicting,
ROI type KPIs that you need
to, to measure against as well.
you, you, you got, you've gotta really
zone in perhaps on and prioritize
what those key KPIs are first, before,
Chris: Yeah.
Darminder: actually
embarking on a project.
And fundamentally we
come back to data, right?
Do we have the right data to, to
deliver that, that, that transformation.
we, we mentioned defense.
we, we did say that data is, a challenge.
it's the same thing with enterprise.
A lot of data is siloed in,
in different, business units
and business departments, so.
You know, how do I get
that data connected?
we, we grew up through a transition of
trying to collect data and put it in one
big place called a data lake, you know,
Chris: Hm.
Darminder: but that, that raised
challenges around it became more
expensive and very difficult to maintain.
But now with the advent of, let's
say MCP, for example, with Agen ai.
you can, you can sort of leave
data where it is and, and
have a more distributed model.
and how allow a MCP to connect
your data and have the agents
query the right data when required?
and, and what that's gonna do is what we
sort of call a data mesh architecture.
working on very specific problems
to their processes and, domain
knowledge will be the custodians
of their data, for themselves.
and then as long as an enterprise has
a good governance model to make that
data available for others to use,
that's gonna unlock a lot of value.
in the organization moving forward.
so it's kind of like a, of a theater, sort
of starting at the business problem, the
application layer, then moving downwards.
And I've kind of articulated a
couple of challenges along the way.
Chris: Thank you so much for sharing that.
And a question that comes to mind for
me there is, you know, security of these
agents and how much of organizations that
you are dealing with thinking about, you
know, we touched on adversarial attacks
earlier and, how, how much do people worry
about an agent going off into their data
or, or leaving it to do like a long task?
Darminder: Yeah.
And, and, and you know, I did mention
about the transformation that you
bring being one of responsible,
responsibility and, and risk
Chris: Hmm.
Darminder: So you do need very
good, effective guardrails.
You know, a classic example is not
releasing personal identifiable
information, for example, getting
personal information that's not
relevant to, the agent's process.
So you've gotta really put those checks
and balances in guard rails and play.
Chris: Hmm.
Darminder: lot of the, a lot of the
agentic frameworks and, and technologies
that allow you to develop and, and
actually, deploy agents, have a very
good pull component within that that
does a lot of the guardrails and,
and responsibility aspects with that.
and, you know.
similar to what we were talking
about, adversarial attacks, you
still need an ability to do some
form of red teaming against those
agents in a safe sandbox environment.
That, that you can then really, really hit
it with interesting edge use cases, right?
Edge, edge, edge considerations, which
really then allow you to understand
does the agent really perform well or
not, where the limitations might be.
So you're able to kind of put,
risk-based, scoring on it.
You know, that it, that in this way
could act as a way of, doing some
self-learning with agents, kind of
driving that data back to kind of then,
give it a much more informed view of
what's correct and what's not correct.
but yeah, I funda this is something
which is a, a very core component and
that's gonna be hugely driven by the
regulatory landscape out there as well.
Chris: Mm.
Darminder: AI services into
different marketplaces geographically
in both, nationally as well.
Chris: Nice.
I think that leads, nicely onto the
final part of the podcast, which is
just, talking about the future and the
change of pace seems to be getting more
rapid, you know, for every six months.
But, From a personal and even
business perspective, what trends
are exciting you the most right now?
Darminder: Yeah, absolutely.
I think, like you mentioned OpenCL, right?
these, these orchestration, frameworks,
which, you know, are going from.
a place where we could give it a query and
it understands how to, essentially, put
Chris: I.
Darminder: put a process
in play to deliver outcome.
So I think we're getting to a point where.
we need to develop skill sets that can,
articulate our own domain expertise, the
experience that we've built throughout
our years of working into, artifacts
that agents can understand now.
So we look at everything which,
that everybody leverages their
own agent skills, for example,
as, as mark down files, right?
we see people creating.
Pacific Markdown files that
describe, their own personal profile.
Profile or personality.
the way that they communicate, the
way that they want to articulate,
how they write information.
all it is, is writing that down
in a markdown file and allowing
an agent to, to interpret that.
so, so based, working, developing,
articulating great skills, leveraging
that can actually then, improve
how your agent performs is gonna
be really, really important.
along with tools as well
that the agent can use.
and I think from a user
perspective, it's trying to.
articulate the best way,
of understanding yourself.
What is the right tool to use?
What is the right skill to bring in?
and, and then you, you, you can start
to then, you know, play around with
these concepts to kind of give you the
best result that you're after, right?
So, I think that way of working is gonna.
You, you know, fundamentally,
tr you know, change everybody's
perspective of how they, how they
work with agents in the future.
and as we could say, as intelligence
builds up within the foundational model,
I think a lot of those capabilities
might come into the foundational model.
we, we might be just a prompter
way from, work being arti, work
being given, and done autonomously.
Chris: Hmm.
Darminder: one of the key things is
that I think, One of the areas which
I think needs further refinement
is everything that you provide an
agent has to be the right context,
the right level of information.
That context is specific to your
problem, your domain, your expertise.
and that's something which we should
hold onto as much as possible and
refine, because I think that's
gonna be moving forward our key
USP, in the future of work, I think.
Chris: Nice.
And, from a personal perspective,
are you using dispatch or, open
cloud to, to, in your personal life?
Darminder: Yeah, so I've got my own open
call running in, My own, my own laptop.
Not, not the enterprise laptop, but
definitely, definitely working with
Open, definitely leveraging, all of
the coding agent tools which are out
there, crawl code, code X, for example.
and, and you know, as we start to
use some of these tools, really
understanding where certain tools give
better benefit than others and what,
how they sort of work in different ways.
so I think, yeah, it's, it's,
it's all about a learning.
I think.
we, we do need to do, on the game
of actually trying to keep ourselves
continuously learning as new capabilities,
new functionalities emerge, in this kind
of, in this new, new areas of, of work.
I think,
Chris: Yeah.
And do you know, do you think that,
you hear a lot on the internet and,
you know, through various, quote
unquote experts that, we're not
that far from autonomous employees.
do you think that is teams
we've augmented AI employees?
I, I is real and will be here soon.
Darminder: I personally think
it's moving in that direction.
so digital avatars of certain work,
certain colleagues, certain profiles,
is, is already here to some degree.
and I, and I guess it's, it's really then
a trust game because if you can prove
that your agents are going to be reliable
deliver the same outcome repeatedly,
on over again, I think that's gonna.
move us to that kind of
trajectory going forward.
I think we're still in that
phase of understanding how do
we take agents to production?
How do we make them reliable?
How do we ob, how do we observe
that they're delivering the right
outcomes over and over again?
then communicating that.
across to users.
Chris: Hmm.
Darminder: I think the
directory is certainly there.
I'm not gonna put a timeframe on that,
but I think that that's the, general
movement in that directory, no doubt.
Chris: Cool.
And final question.
I know intelligence and knowledge, is
becoming a bit of a commodity with, with
all these tools, but, you know, the, any
papers from a machine learning and AI
perspective that have really challenged
the way you've thought about, you know,
the industry or, or books that have shaped
your leadership that you'd like to share?
Darminder: Yeah.
absolutely.
So, I mean, uh, I think, um, any,
Publication that talks to how you
do change management efficiently
and effectively, is a good one.
and there's a, and there's a,
there's a number out there.
I think from a, from a paper perspective,
it's, it's great to see how the open
source community are with the kind of
proprietary models which are out there.
So, a recent paper that came out where,
you know, the, the Gemma Gemma models just
has been released, it's now outperforming.
Quinn models in, in,
Chris: Hmm.
Darminder: in certain task-based,
performance benchmarks, just shows you.
how these models are competing
with foundational models,
which are proprietary.
So I think it, it's, it's greatest
to, see those papers and, and that
research being, being, being publicized
because it opens up opportunities
to how you deploy and experiment
with AI models in different ways.
A great example would be if you, if you're
really concerned about data sensitivity
and proprietary data and you don't want
it to leave your, your laptop, what,
why don't you, why don't you deploy
a, a local, a local model through some
type, an alarm type framework and work
on that data, without worrying about
it being pinged across an API, so.
If you're aware of what the leading
open source models are, they might
have the same equivalent, performance
to perhaps a claw or an open ai.
So I think that's one area
that, you should, I think we
should focus on a bit more.
Chris: Nice.
Well, I'll include the link to
the paper in the podcast as well
as your, LinkedIn profiles and,
you know, what links to your work.
Thank you so much for coming on the
podcast, Amin, and sharing your knowledge.
really appreciate it.
It's been been a real pleasure
to, to have you on at last.
Darminder: Yeah, absolutely.
Chris, great to catch up and likewise.
It's been a good discussion.
Chris: Thank you so much.
Darminder: Thank so much.