December 29, 2025 • 28 mins
AI that augments — not replaces — your workforce: practical steps for manufacturers starting today.
In this episode, Murali Sastry, learning and AI strategist, shares a grounded framework for AI adoption in manufacturing that keeps humans in the loop. We cover:
- What vibe coding is and how frontline workers are using it to build dashboards without IT
- Learning in the flow of work: low-risk AI augmentation use cases with real examples
- Why human-in-the-loop governance is non-negotiable in manufacturing environments
- The risks of shadow IT, data ownership, and where AI should never be used on the floor
- Practical steps manufacturers can take tomorrow to build AI capability safely
Whether you're an operations manager, plant engineer, or digital transformation lead, this episode gives you a human-centred approach to manufacturing AI adoption.
Show notes and links for this episode: https://www.manufacturingtech.au/episode/74-ai-on-the-factory-floor-learning-in-the-flow-of-work-and-the-rise-of-vibe-coding/
Listen
Watch
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:02):
Welcome to Manufacturing Tech Australia with Shane Williams
and Paul Mason with actually thelatest manufacturing and tech
news and export innovative solutions to help you improve
your business. Welcome back to Manufacturing
Tech Australia for our last showof 2025.
Today, Shane and I are diving into one of the biggest topics
in manufacturing right now, AI, but with a very practical,
(00:25):
grounded lens. There's a lot of noise out there
about AI replacing jobs, automating everything, and
taking over the factory floor, but our view has always been a
little bit different. We believe the real opportunity
is AI augmenting people, not replacing them, especially in
manufacturing where safety, complexity, and human expertise
matter more than ever. To help unpack that, we're
(00:46):
joined today by Murali Sastry, who is the CTO from Skill Soft
in the United States. Murali works at the intersection
of AI learning and workforce enablement, helping
organizations understand how newtechnologies can genuinely
support people in their day-to-day work.
In this conversation, we explorewhat concepts like Vibe coding
(01:07):
and AI assisted engineering really mean and how AI can
enable learning in the flow of work and what this looks like on
the factory floor, from troubleshooting equipment and
building dashboards to empowering frontline workers
without creating new risks. But before we get into today's
show, please be sure to follow the podcasts and please share it
with a few friends or colleaguesto help us help businesses like
(01:30):
yours to continue to build your manufacturing capabilities.
Now let's get into it. Murali, thanks for joining us on
Manufacturing Tech Australia. Thank you.
It's a great pleasure to be here.
We wanted to have it on today because there's a whole bunch of
talk about AI. Every other topic is about AI,
right? And Paul and I have got this
philosophy that we think the thebig picture in the use of AI,
(01:50):
particularly in the manufacturing and automation
space, is all around this augmenting with the people as
opposed to it being a technologythat sort of drives itself.
So I'd love to get your view on how you see AI changing the way
that sort of workers learn and do their jobs, and particularly
when it comes to augmenting rather than just replacing human
capability. VCAI as a great enabler of human
(02:14):
intent and the ability to augment themselves, scale
themselves, right? Let me start by actually giving
a small example of how how this could apply in the manufacturing
space. I was reading recently reading a
book by Dan Heath called Reset where he talks about this
factory which is shutting down their machinery at noon time
(02:34):
every day, right? And nobody understands why this
was happening. Now imagine we want building a
tool to track all the factory downtimes across a period of
time very quickly just expressing to an application to
to visualize this data, right? I think it's doing things like
(02:54):
this which will give people a tremendous power to look at
systems and data that has historically not existed for
people outside of IT. Typically one would have to go
to IT, put in a lot of requests and then build a lot of these
applications. But what we are seeing more and
(03:15):
more is this notion. I mean, people call this vibe
coding, vibe engineering. We call think of this as AI
assisted engineering. This gives the ability now, not
just for the engineers, but people in all walks of life to
quickly build many kinds of capabilities.
It's going to be a great differentiator, especially teams
(03:38):
that don't have this huge technology investments to build
this capable. Yeah, Vibe coding or vibe
engineering, it's really interesting and I kind of feel
it's similar to generative design that we've spoken about
on the show previously. You feed AIA bunch of
requirements and parameters and ask it to come up with the best
solution. In that case, I think it was to
design the best chair for the lowest cost and ease of
(03:58):
manufacturer, for example. But in this case, vibe coding is
is enabling like natural human language is enabling the
operator on the floor with just natural human language to spin
up an app that visualizes down times in this example and not
having to engage the IT team or get an external developer in.
So I guess how are you seeing these new possibilities change
the manufacturing environment? This new breed of technologies
(04:21):
is very different than what we've seen before, right?
AI fundamentally transforms how you see raw observations into
actionable data. People on factory floors for
example, now can very quickly draft statement of workflows,
troubleshoot even machinery. Say you want a quick way to
(04:42):
build a safety checklist, right?So you could build an app for
example, build a go through machine checklist.
This requires historically used to require a lot of technical
know how. Right now we put this capability
in the hands of people who on the factory floors are most
knowledgeable. I'll give you a recent example.
(05:04):
Even inside skill soft, we buildsystems where for example
intermittently we let's say we, we have some site outage.
Now most of the engineers and ITteams are building production
platform applications with thesecapabilities.
Now we have traditionally systems people, IT
administrators being able to go build a dashboard that says,
(05:28):
hey, we're undergoing through maintenance.
They build a quick web page. We build an app that shows this.
We are leveraging AI internally in skill soft in two ways.
One, to really transform the wayour customers consume learning.
Instead of having to go to a website or an application to
(05:48):
create hours long training, now we're able to deliver training
right in the flow. People are doing their work
through through mobile applications or tablets in bite
sized elements, what we call this as learning in the flow,
right? So let's say you're again office
worker working in a factory where you have a customer that
(06:10):
is very upset with the way that your product is working or
they're failing to understand. Now we've created capability
where looking at those kind of conversations in real time, we
can customize learning experience in terms of
conversing with AI, mimicking that conversation, understanding
in different situations, how youcan train for yourself to have
(06:32):
those conversations. But there is an element of it
where we're building learning experiences.
We call it learning in the flow of work that gets embedded right
into people's work. And then the other dimension of
it is we ourselves are consumingAI to improve the way we deliver
technologies and the learning modalities to our customers.
(06:54):
This actually starts with deeplyunderstanding process, right?
And the best people who are ableto devise this process
engineering are people who actually know the business.
So the people who drive the business as they move forward
will be equipped with a whole new set of technology
technologies where they can helpthe businesses build these
(07:15):
capabilities move forward. I love what you said there,
learning in the flow so workers can not only have tailored
learning experiences but also inreal time as they're working and
learning. Where I guess most people learn
by doing, not watching, so it makes so much sense.
It kind of reminds me of when you get an office printer, not a
desktop one like a full size office printer and it gets a
(07:36):
paper jam and it's the screen comes up and it's like paper
jam, lift flap and you lift the flap up and it detects that
you've done it and then it changes the step.
OK, now pull this out and you pull it out.
You've never touched the printerbefore.
You've got no idea. Follow the five steps that it
shows you on the screen. Pull the paper jam out, shut it
back up and back up and running.That almost feels like that
learning inflow as a use case. Yeah, that's a great example,
(07:57):
right? So let's take that as an
example. Now what someone can do.
Let's say you have these set of instructions in a paper form,
right? And you have to troubleshoot
through some manual or something.
Now using some of the modern tools available out there, you
can take pictures of this manual, feed it to some of these
newer vibe coding or vibe engineering tools, have it built
(08:17):
an app. Now you can have this app on
your phone. You can go through the same
system and say hey, I'm having this problem, tell me how to fix
it. So it's bringing in the
knowledge from your domain and making it applicable in real
time to solve real world business problems.
It's going to be a whole new world out there.
I think engineers as they move forward will be more of people
(08:39):
who are orchestrating systems then sitting there and actually
building systems. This year, 2025 is called the
year of the agents. I'm sure all of us are hearing
this term agents being thrown around, but in some sense we
will all have these agents that are able to help us do a great
many things. I will come back to some of the
(09:01):
pitfalls, right? So because this technology, we
have to be also very careful in terms of how we apply and where
we apply it. So I'd say at the same time take
all of this with a pinch of salt, right?
Going back, one of the most fundamental things I think we as
skills software in the learning industry been doing and look to
do is the factory worker or knowledge worker, software
(09:24):
engineer. All these lines are becoming
more and more blurred because you have tremendous intelligence
in in your hands. Now people can do a great many
things can apply their domain knowledge at the same time.
So let's say you're building a quick dashboard to do some
quality control for your manufacturing system.
(09:46):
These modern LLMS can help you build all of these dashboards if
the information is presented in some form of caution,
distribution or statistical probability.
Now you might not know what standard deviation is, what
variance is, or if it's we're talking about customer
satisfaction, what is NPS? How is it measured?
(10:06):
So at the same time have to learn a lot of new concepts
which requires a great deal formof curiosity and a newer form of
learning that needs to happen. People need to become more multi
dimensional in terms of the way they learn and build their
skills. And one of the most important
(10:28):
human quality moving forward I think is going to be 1 of change
management, curiosity, creativity and their ability to
apply their domain knowledge to new constructs.
So the knowledge worker, the person who is actually doing the
work on factory floors or manufacturing, they are the ones
(10:48):
who will actually move the needle forward in the industry
with this new tech. OK, so we're talking about
enabling workers on the factory floor to use Vibe coding to
perhaps you know, create a dashboard, extract information
out of the system, maybe troubleshoot or fault, find any
piece of equipment what it's doing wrong.
So I guess maybe it's important at this point in time, maybe I
(11:09):
can get you to explain to our listeners exactly what Vibe
Coding is. Yeah, so the term wipe coding of
listening to another podcast recently apparently is one of
the new terms of most popular terms of this year, 2025.
It started the word wipe startedwith tweet by someone named
(11:29):
Andre Karpati, who was the city of Tesla and part of Open AI and
other places. And he said he's wiping with the
LLMS talking to it, explaining things and it's building code,
right. So it's this new means of
communication where talking to this newer models and explaining
stuff, asking that stuff and youget code back to build products
(11:53):
and applications that are solving this over the year has
tremendously evolved. If you look at the latest models
out there that were just released, even with open AISG
PD52 that got released a few weeks back, you're able to build
end to end applications, really build end to end applications
(12:13):
that are working products without knowing any code.
This at the same time comes withits own pitfalls and challenges
we'll talk about next. But there's this new form of
just talking, sharing documents,even videos, photographs, and
having these development environments that are built so
(12:34):
you don't have to depend on it as much.
But at the same time, one has tobe very careful when you build
these sort of products because once you build it, who is owning
it, Who is maintaining it, How do you maintain it?
There's this whole question of governance and security and
privacy, things that come with it.
(12:56):
Yeah, I think no doubt there's some challenges there.
I studied software engineering and all the sort of basics
around pseudo coding and logic and learning all these different
principles around safety and allthat kind of thing could be
completely missed with Vibe coding where you could literally
use human language to say, buildme an app that does ABC without
taking into account sort of the ramifications or doing the
(13:17):
checks and those kind of things.So I think it's enabling and
empowering in one way, but it's also a little bit frightening in
another if it's in the wrong hands.
When we were chatting earlier offline, we were talking about
how people can use this to sort of prototype.
They can stand up things really quickly, leverage that data and
just maybe make a the pilot system to try it out so that
things don't need to be put intothe hands of the IT guys or
(13:38):
developers, maybe external to the company.
That might take weeks or months and cost a lot of cash.
It can be prototyped internally and putting that power into the
hands of the guys on the floor. And I think that's the trick,
right? You said the right word.
These set of technologies are great for prototyping, ideating,
brainstorming, testing the concept, being a little bit
(14:00):
creative. But yeah, when it comes to you
don't want to wipe code your wayinto a CAD design to a robotic
arm that is building cars, right?
One has to understand very clearly where this technology
could be applied and where you should be very careful.
This is interesting, right 'cause it's not even a new
problem, but forever in a day. There's been the IT guys trying
(14:22):
to get their hands around and control what's happening in the
environment. And then there's been the the
shadow IT that the smart people that exist in various parts of
the business who I couldn't get support.
Or. Someone wouldn't help me, or I'm
smart enough and I just want to do this myself and I'm clever.
And I could be as simple as building something in an Excel
sheet with a bunch of weird formulas and macros and all
sorts of stuff, right? And it runs great for years
(14:44):
until that person leaves, and then no one's got any idea how
it works or it's written in an old version and you can't
upgrade it or whatever. Then we just grab that concept
and we go, yeah, that was alwayssomeone with a bit of now, so
you could read a book or go online and learn and build
something, and it didn't really have a lot of control around it.
And then we go, OK, now let's create the dial it up to 11
version and then give that to the people who've got even less
(15:05):
now in terms of understanding around security and everything
else and an amazing opportunity for those people to go, I don't
know about engineering, I don't know about how to build code,
but I've got this idea of something and I can demonstrate
it. Amazing, huge opportunity.
But to your point, that's great for prototyping.
The risk for most of these businesses is going to be that
becomes, oh, let's just push that into production.
It's a working piece of softwareand I've worked for
(15:28):
organisations in the past where they, prior to me arriving,
they've engaged a marketing company going, we want to build
an app that does XY and Z and they've got this amazing, oh
look, it's a working application.
You've just got to come in and help us deploy it globally.
And you're like, hang on this, this is some colourful screens
and some buttons that Click to do absolutely nothing behind the
scenes except pretend to show you what it's going to be.
And the customer rethinks. They've bought a fully fledged
(15:49):
ready to deploy solution. And there's this huge gap there.
And it's going to be really fascinating to see how
businesses adopt the whole concept of vibe coding, but
protect their businesses in suchthat they don't end up with a
whole bunch of citizen developedapplications which are full of
holes and flaws in logic and security implications and
pulling all that together, right?
Yeah, yeah. Right now where we are, at least
(16:12):
as we should be handled with great deal of care, establishing
governance. Because even in engineering,
what we're seeing is even other engineers who are actually using
some of these modern editors or code generator.
Once you ship a piece of code toproduction, who owns it?
Is it you, the engineer or is itthe LLM that wants the code?
(16:34):
Because if it breaks, it has to be the engineer that has to fix
it. So everything that gets built
has to be properly vetted and owned.
And I think you've had a great point, right.
So we are actually now seeing the shadow IT problem on
steroids. We should be very clear in terms
(16:54):
of where you use these technologies and where you don't
want to use this technology. So maybe the way to use it is
again, come up with new ideas, new dashboard.
And if you really want to use this in your operations, it
still gives you a much faster path to work with your IT
counterparts to prioritize it rather than waiting for other
(17:17):
resources or other people. So it still makes things move
much faster than historically, but a lot of caution has to be
put in. Probably over index there on the
risk around the shadow it, but there's obviously other
challenges with adopting AI driven augmentation to develop
things like training tools. Do you want to just share a
(17:37):
little bit about anything else you can think of that might be
some of the, I guess, challengeswhen people go all in?
So I would actually recommend 5 clear practical steps, right?
One is start with simple augmentation use cases.
Things like the earlier examplesthat I've given, things that are
low risk that give you some immediate value.
(18:00):
Just maybe some troubleshooting guides or drafts or coming up
with the safety checklist right to use wipe coding to let
process owners, not just engineers create tools.
Make sure that humans review andthe data is clearly understood.
Human in the loop is maybe should be actually the term of
(18:21):
the year, not agents, right haveAI that is there, but the human
is in the center basically orchestrating and make sure that
things are flowing through. Now if you look at the
manufacturing industry and frontline workers, we recently
released special training for frontline workers, right, in
(18:42):
terms how do they deal with different scenarios, have skills
reminders or step by step job aids workers where they are.
So develop tooling you using these AI capabilities in terms
of more fitting those solutions to where people are executing
their work. Another thing that I think
(19:03):
whether you are in IT or you're in the production running
systems business is have clear AI champions train a few people
who really understand these technologies well.
So one of the things that we have done at Skill Soft is
whether you produce some learning content or some other
(19:23):
piece of asset, we have identified people who understand
the process of instructional design pedagogy, how to teach
something, how to design, how tocreate.
And then we have built tools forthem to augment them on how they
do this stuff, right? So understand, draw the
processes, understand clearly the processes and see which
(19:46):
steps could be using AI, right? Then last but not the least,
measure, continuously measure how some of these tools and
technologies are helping you, right?
See if you're getting the ROI and based on that.
Invest and build more. Invest where you believe you're
getting the maximum ROI from. Murali, I wanted to ask you a
(20:07):
question out of left field here,mate.
I know we're chatting about AI augmenting jobs, not replacing
them and ensuring that we have the human in the loop.
But I wanted to get your perspective on all the talk in
the media on AI displacing humanjobs.
So some of the high profile examples that we've seen with
IBM, Duolingo and Dropbox, for example, they're looking to
pause recruiting or even replacecontractors with AI agents.
(20:30):
And there's also been a bunch ofstats around the world from the
World Economic Forum and the IMFestimating that up to 40% of
employers expect to reduce theirheadcount by leveraging AI.
But I kind of feel like this is,you know, all the examples we've
seen are in the tech and the digital space.
I feel like in the manufacturingspace, where at least a few
years away, the work is much more complex.
The process has changed regularly.
(20:50):
They're more safety oriented andmore complex than then sort of
replacing repeatable processes and services.
But I wanted to get your thoughts on this.
I'm going to make a cliche statement, which is the AI is
not going to take the job. The people who can leverage AI
are the ones who are going to take the job.
Change management, updating yourskills, your knowledge.
(21:13):
That's the most important thing.I can't predict five years from
now where we're going to be because there's a lot of
research going on in terms of newer AI models and
technologies. But as things stand today, these
current set of LLMS are still prone to hallucinations.
Lot of errors. They are statistical machines,
(21:36):
so one small error in the beginning can cascade into large
errors. And especially when it comes to
places like manufacturing where you need 6 ninths of quality
checks. We are not there yet in terms of
AI. Even in the IT and software
areas, we are still not at a point where the AI could replace
(21:57):
humans. AI is a great augmenter.
It's a great tool for people to scale themselves.
This is like people who were using like spreadsheets came in
Think of a world like people whodoesn't want to use Excel.
You're not going to be a productor able to do it right now.
The modern form of a spreadsheetor in Excel is probably using
the LLM. And in terms of job layouts and
(22:18):
all of that, I think there are alot of high expectations that
were set for this technology three years back or four years
back. There was this paper, Chinchilla
paper, which talked about the scaling loss, where they said if
you put enough data in enough compute, AI is going to be so
smart that it could replace the human.
But now if you look at what's happening currently at this
(22:42):
point in industry, a lot of those people who are foremost
thinkers in the space are comingback and saying that the
realization of that with the current set of technologies that
there is a gap 345 years, maybe those gaps could be closed.
But right now, from what we see,these LLMS or these AI set of
technologies are not yet at a place.
(23:02):
Forget about replacing. I think it's more about it's a
tool, it's an augmentation. Humans have to be at the center
of any of these activities to ensure that they're moving
forward. And then another important thing
is the LLMS or the AIS or these tools are only good as the input
that is given to them. And who is giving all the
(23:23):
context and input is the human AI doesn't have the
consciousness yet. Maybe it'll come in, but it
cannot come up with its own creativity and its own ideas to
to solve. But if you give it a great deal
of instructions, very clear instructions, good
introductions, it produce very high quality, good output, and
(23:44):
you can make decisions. But then there has to be a human
who's at the center of it still to guide it.
The human activity is going to change as we move forward.
I think it's a very fun, great world to be in for many of us,
right? To build things in a new way,
solve new problems. But I think replacing humankind
(24:04):
has been said a lot, many times before in history, from the
steam engines to the printing press to all the modern
technology, the evolutions and transformations that have
happened. Yeah, I totally agree with that.
But I guess there's a big difference between using large
language models that are trainedoff data from all over the
place, which comes from who knows where, versus those
(24:25):
specific models trained inside businesses on that specific
business knowledge. And some of the startups I've
been working with, the ones thatare just leveraging that off the
shelf LLMS that anyone can use and beef up their business, not
so important and not so useful. But the ones that are building
specific models for their own, leveraging all their
intelligence that they've gainedover the last two years, five
(24:46):
years, 10 years, putting that intheir own sort of custom models
to extract out all that knowledge out of their own
systems is where they're seeing the value and being able to
leverage it properly. 100%, the biggest innovations and the most
transformative changes that are going to come are things like
alpha fold that Google worked tocreate in terms of understanding
(25:08):
protein folding. There are new models that
understand human DNA, but those are not just build on language
models, right? They are using the historical
AI, machine learning algorithms and a lot of other techniques to
build a combination purpose fit to solve problems and that's
where we will get the return. So I think even in manufacturing
(25:31):
there is a lot of opportunity tofigure out how do you take what
is there in terms of the company's knowledge and convert
it into mechanisms that you could apply to solve business
problems. I'm really enjoying this
conversation, but as is typical of a podcast, we have to wrap it
at some point. And so it's been really
interesting to hear a technologist's perspective on
(25:53):
where does AI fit and what does vibe coding mean for
manufacturing context. So I appreciate you coming on
the show to talk with us. I guess my key takeaways are
that these are great tools that are going to enable the
democratisation of technology solutions to solve on the floor
problems. But we just need to make sure
that we're putting the right governance around that so that
we know what use cases we will use it for.
(26:15):
And we make sure we're not usingit for high risk things and
we're not necessarily trusting the output, but the humans still
in the loop and taking ownership.
I think they're probably some ofthe high level lessons for me.
But I think just for the sake ofthose listening, I'm in a
manufacturing business. My staff are asking me for
access just these tools. What's the practical get started
tomorrow thing you need to do? Yeah.
So I think first step would be to first and foremost work with
(26:39):
your IT partners and team to enable tools, right?
Whether it's Copilot or Google, Gemini models have a very safe
AI system that you can consume inside your company.
So that IP and data is actually not leaking out into the public
LLMS, right. So first, a QPR employee base
with the tools, identify businesses and processes and
(27:02):
look at augmenting or automatingthe processes that give it the
maximum return for the business,right?
Things that could be easily automated, not just easily
automated, but could be automated with precision, right?
So that you don't have a risk toyour business, identify this for
low risk ones. And then as you progress,
upskill and train your people onhow AI could be used for
(27:24):
different roles. So we at Skill Soft have built a
curriculum around job roles and functions where they can use AI.
Have your workforce trained in terms of adoption because every
role could use it in some different ways.
A manufacturing company obviously has marketing, sales.
They could use AI in certain ways.
Then train workforce in terms ofwhat's the safety, what's the
(27:46):
governance, what is the life cycle around some of the AI
tooling to solve that business, business problems have IT
champions, AI champions who within the context of the
environment help take it progress forward because you
need someone to push things along, right?
Always understanding the domain.And so that having the notion of
(28:07):
the AI champions important constantly measure, monitor and
develop governance around this. So that's I think the key steps
I would say. We really appreciate you coming
on the show today, mate, and I know that we'll catch up again
soon, but thanks for sharing your insights and coming on the
show today. Thank you for having me.
Cheers. Thanks for tuning in to
(28:29):
Manufacturing Tech Australia with Shane and Paul recorded on
the traditional lands of the Bunnerong and Mouranjuri people.
The more information jump on theManufacturing Tech dot AU
website. Remember to hit the follow
button to join us again next time as we continue to explore
the intersection of manufacturing and technology.
Welcome to Manufacturing Tech Australia with Shane Williams
and Paul Mason with actually thelatest manufacturing and tech
news and export innovative solutions to help you improve
your business. Welcome back to Manufacturing
Tech Australia for our last showof 2025.
Today, Shane and I are diving into one of the biggest topics
in manufacturing right now, AI, but with a very practical,
(00:25):
grounded lens. There's a lot of noise out there
about AI replacing jobs, automating everything, and
taking over the factory floor, but our view has always been a
little bit different. We believe the real opportunity
is AI augmenting people, not replacing them, especially in
manufacturing where safety, complexity, and human expertise
matter more than ever. To help unpack that, we're
(00:46):
joined today by Murali Sastry, who is the CTO from Skill Soft
in the United States. Murali works at the intersection
of AI learning and workforce enablement, helping
organizations understand how newtechnologies can genuinely
support people in their day-to-day work.
In this conversation, we explorewhat concepts like Vibe coding
(01:07):
and AI assisted engineering really mean and how AI can
enable learning in the flow of work and what this looks like on
the factory floor, from troubleshooting equipment and
building dashboards to empowering frontline workers
without creating new risks. But before we get into today's
show, please be sure to follow the podcasts and please share it
with a few friends or colleaguesto help us help businesses like
(01:30):
yours to continue to build your manufacturing capabilities.
Now let's get into it. Murali, thanks for joining us on
Manufacturing Tech Australia. Thank you.
It's a great pleasure to be here.
We wanted to have it on today because there's a whole bunch of
talk about AI. Every other topic is about AI,
right? And Paul and I have got this
philosophy that we think the thebig picture in the use of AI,
(01:50):
particularly in the manufacturing and automation
space, is all around this augmenting with the people as
opposed to it being a technologythat sort of drives itself.
So I'd love to get your view on how you see AI changing the way
that sort of workers learn and do their jobs, and particularly
when it comes to augmenting rather than just replacing human
capability. VCAI as a great enabler of human
(02:14):
intent and the ability to augment themselves, scale
themselves, right? Let me start by actually giving
a small example of how how this could apply in the manufacturing
space. I was reading recently reading a
book by Dan Heath called Reset where he talks about this
factory which is shutting down their machinery at noon time
(02:34):
every day, right? And nobody understands why this
was happening. Now imagine we want building a
tool to track all the factory downtimes across a period of
time very quickly just expressing to an application to
to visualize this data, right? I think it's doing things like
(02:54):
this which will give people a tremendous power to look at
systems and data that has historically not existed for
people outside of IT. Typically one would have to go
to IT, put in a lot of requests and then build a lot of these
applications. But what we are seeing more and
(03:15):
more is this notion. I mean, people call this vibe
coding, vibe engineering. We call think of this as AI
assisted engineering. This gives the ability now, not
just for the engineers, but people in all walks of life to
quickly build many kinds of capabilities.
It's going to be a great differentiator, especially teams
(03:38):
that don't have this huge technology investments to build
this capable. Yeah, Vibe coding or vibe
engineering, it's really interesting and I kind of feel
it's similar to generative design that we've spoken about
on the show previously. You feed AIA bunch of
requirements and parameters and ask it to come up with the best
solution. In that case, I think it was to
design the best chair for the lowest cost and ease of
(03:58):
manufacturer, for example. But in this case, vibe coding is
is enabling like natural human language is enabling the
operator on the floor with just natural human language to spin
up an app that visualizes down times in this example and not
having to engage the IT team or get an external developer in.
So I guess how are you seeing these new possibilities change
the manufacturing environment? This new breed of technologies
(04:21):
is very different than what we've seen before, right?
AI fundamentally transforms how you see raw observations into
actionable data. People on factory floors for
example, now can very quickly draft statement of workflows,
troubleshoot even machinery. Say you want a quick way to
(04:42):
build a safety checklist, right?So you could build an app for
example, build a go through machine checklist.
This requires historically used to require a lot of technical
know how. Right now we put this capability
in the hands of people who on the factory floors are most
knowledgeable. I'll give you a recent example.
(05:04):
Even inside skill soft, we buildsystems where for example
intermittently we let's say we, we have some site outage.
Now most of the engineers and ITteams are building production
platform applications with thesecapabilities.
Now we have traditionally systems people, IT
administrators being able to go build a dashboard that says,
(05:28):
hey, we're undergoing through maintenance.
They build a quick web page. We build an app that shows this.
We are leveraging AI internally in skill soft in two ways.
One, to really transform the wayour customers consume learning.
Instead of having to go to a website or an application to
(05:48):
create hours long training, now we're able to deliver training
right in the flow. People are doing their work
through through mobile applications or tablets in bite
sized elements, what we call this as learning in the flow,
right? So let's say you're again office
worker working in a factory where you have a customer that
(06:10):
is very upset with the way that your product is working or
they're failing to understand. Now we've created capability
where looking at those kind of conversations in real time, we
can customize learning experience in terms of
conversing with AI, mimicking that conversation, understanding
in different situations, how youcan train for yourself to have
(06:32):
those conversations. But there is an element of it
where we're building learning experiences.
We call it learning in the flow of work that gets embedded right
into people's work. And then the other dimension of
it is we ourselves are consumingAI to improve the way we deliver
technologies and the learning modalities to our customers.
(06:54):
This actually starts with deeplyunderstanding process, right?
And the best people who are ableto devise this process
engineering are people who actually know the business.
So the people who drive the business as they move forward
will be equipped with a whole new set of technology
technologies where they can helpthe businesses build these
(07:15):
capabilities move forward. I love what you said there,
learning in the flow so workers can not only have tailored
learning experiences but also inreal time as they're working and
learning. Where I guess most people learn
by doing, not watching, so it makes so much sense.
It kind of reminds me of when you get an office printer, not a
desktop one like a full size office printer and it gets a
(07:36):
paper jam and it's the screen comes up and it's like paper
jam, lift flap and you lift the flap up and it detects that
you've done it and then it changes the step.
OK, now pull this out and you pull it out.
You've never touched the printerbefore.
You've got no idea. Follow the five steps that it
shows you on the screen. Pull the paper jam out, shut it
back up and back up and running.That almost feels like that
learning inflow as a use case. Yeah, that's a great example,
(07:57):
right? So let's take that as an
example. Now what someone can do.
Let's say you have these set of instructions in a paper form,
right? And you have to troubleshoot
through some manual or something.
Now using some of the modern tools available out there, you
can take pictures of this manual, feed it to some of these
newer vibe coding or vibe engineering tools, have it built
(08:17):
an app. Now you can have this app on
your phone. You can go through the same
system and say hey, I'm having this problem, tell me how to fix
it. So it's bringing in the
knowledge from your domain and making it applicable in real
time to solve real world business problems.
It's going to be a whole new world out there.
I think engineers as they move forward will be more of people
(08:39):
who are orchestrating systems then sitting there and actually
building systems. This year, 2025 is called the
year of the agents. I'm sure all of us are hearing
this term agents being thrown around, but in some sense we
will all have these agents that are able to help us do a great
many things. I will come back to some of the
(09:01):
pitfalls, right? So because this technology, we
have to be also very careful in terms of how we apply and where
we apply it. So I'd say at the same time take
all of this with a pinch of salt, right?
Going back, one of the most fundamental things I think we as
skills software in the learning industry been doing and look to
do is the factory worker or knowledge worker, software
(09:24):
engineer. All these lines are becoming
more and more blurred because you have tremendous intelligence
in in your hands. Now people can do a great many
things can apply their domain knowledge at the same time.
So let's say you're building a quick dashboard to do some
quality control for your manufacturing system.
(09:46):
These modern LLMS can help you build all of these dashboards if
the information is presented in some form of caution,
distribution or statistical probability.
Now you might not know what standard deviation is, what
variance is, or if it's we're talking about customer
satisfaction, what is NPS? How is it measured?
(10:06):
So at the same time have to learn a lot of new concepts
which requires a great deal formof curiosity and a newer form of
learning that needs to happen. People need to become more multi
dimensional in terms of the way they learn and build their
skills. And one of the most important
(10:28):
human quality moving forward I think is going to be 1 of change
management, curiosity, creativity and their ability to
apply their domain knowledge to new constructs.
So the knowledge worker, the person who is actually doing the
work on factory floors or manufacturing, they are the ones
(10:48):
who will actually move the needle forward in the industry
with this new tech. OK, so we're talking about
enabling workers on the factory floor to use Vibe coding to
perhaps you know, create a dashboard, extract information
out of the system, maybe troubleshoot or fault, find any
piece of equipment what it's doing wrong.
So I guess maybe it's important at this point in time, maybe I
(11:09):
can get you to explain to our listeners exactly what Vibe
Coding is. Yeah, so the term wipe coding of
listening to another podcast recently apparently is one of
the new terms of most popular terms of this year, 2025.
It started the word wipe startedwith tweet by someone named
(11:29):
Andre Karpati, who was the city of Tesla and part of Open AI and
other places. And he said he's wiping with the
LLMS talking to it, explaining things and it's building code,
right. So it's this new means of
communication where talking to this newer models and explaining
stuff, asking that stuff and youget code back to build products
(11:53):
and applications that are solving this over the year has
tremendously evolved. If you look at the latest models
out there that were just released, even with open AISG
PD52 that got released a few weeks back, you're able to build
end to end applications, really build end to end applications
(12:13):
that are working products without knowing any code.
This at the same time comes withits own pitfalls and challenges
we'll talk about next. But there's this new form of
just talking, sharing documents,even videos, photographs, and
having these development environments that are built so
(12:34):
you don't have to depend on it as much.
But at the same time, one has tobe very careful when you build
these sort of products because once you build it, who is owning
it, Who is maintaining it, How do you maintain it?
There's this whole question of governance and security and
privacy, things that come with it.
(12:56):
Yeah, I think no doubt there's some challenges there.
I studied software engineering and all the sort of basics
around pseudo coding and logic and learning all these different
principles around safety and allthat kind of thing could be
completely missed with Vibe coding where you could literally
use human language to say, buildme an app that does ABC without
taking into account sort of the ramifications or doing the
(13:17):
checks and those kind of things.So I think it's enabling and
empowering in one way, but it's also a little bit frightening in
another if it's in the wrong hands.
When we were chatting earlier offline, we were talking about
how people can use this to sort of prototype.
They can stand up things really quickly, leverage that data and
just maybe make a the pilot system to try it out so that
things don't need to be put intothe hands of the IT guys or
(13:38):
developers, maybe external to the company.
That might take weeks or months and cost a lot of cash.
It can be prototyped internally and putting that power into the
hands of the guys on the floor. And I think that's the trick,
right? You said the right word.
These set of technologies are great for prototyping, ideating,
brainstorming, testing the concept, being a little bit
(14:00):
creative. But yeah, when it comes to you
don't want to wipe code your wayinto a CAD design to a robotic
arm that is building cars, right?
One has to understand very clearly where this technology
could be applied and where you should be very careful.
This is interesting, right 'cause it's not even a new
problem, but forever in a day. There's been the IT guys trying
(14:22):
to get their hands around and control what's happening in the
environment. And then there's been the the
shadow IT that the smart people that exist in various parts of
the business who I couldn't get support.
Or. Someone wouldn't help me, or I'm
smart enough and I just want to do this myself and I'm clever.
And I could be as simple as building something in an Excel
sheet with a bunch of weird formulas and macros and all
sorts of stuff, right? And it runs great for years
(14:44):
until that person leaves, and then no one's got any idea how
it works or it's written in an old version and you can't
upgrade it or whatever. Then we just grab that concept
and we go, yeah, that was alwayssomeone with a bit of now, so
you could read a book or go online and learn and build
something, and it didn't really have a lot of control around it.
And then we go, OK, now let's create the dial it up to 11
version and then give that to the people who've got even less
(15:05):
now in terms of understanding around security and everything
else and an amazing opportunity for those people to go, I don't
know about engineering, I don't know about how to build code,
but I've got this idea of something and I can demonstrate
it. Amazing, huge opportunity.
But to your point, that's great for prototyping.
The risk for most of these businesses is going to be that
becomes, oh, let's just push that into production.
It's a working piece of softwareand I've worked for
(15:28):
organisations in the past where they, prior to me arriving,
they've engaged a marketing company going, we want to build
an app that does XY and Z and they've got this amazing, oh
look, it's a working application.
You've just got to come in and help us deploy it globally.
And you're like, hang on this, this is some colourful screens
and some buttons that Click to do absolutely nothing behind the
scenes except pretend to show you what it's going to be.
And the customer rethinks. They've bought a fully fledged
(15:49):
ready to deploy solution. And there's this huge gap there.
And it's going to be really fascinating to see how
businesses adopt the whole concept of vibe coding, but
protect their businesses in suchthat they don't end up with a
whole bunch of citizen developedapplications which are full of
holes and flaws in logic and security implications and
pulling all that together, right?
Yeah, yeah. Right now where we are, at least
(16:12):
as we should be handled with great deal of care, establishing
governance. Because even in engineering,
what we're seeing is even other engineers who are actually using
some of these modern editors or code generator.
Once you ship a piece of code toproduction, who owns it?
Is it you, the engineer or is itthe LLM that wants the code?
(16:34):
Because if it breaks, it has to be the engineer that has to fix
it. So everything that gets built
has to be properly vetted and owned.
And I think you've had a great point, right.
So we are actually now seeing the shadow IT problem on
steroids. We should be very clear in terms
(16:54):
of where you use these technologies and where you don't
want to use this technology. So maybe the way to use it is
again, come up with new ideas, new dashboard.
And if you really want to use this in your operations, it
still gives you a much faster path to work with your IT
counterparts to prioritize it rather than waiting for other
(17:17):
resources or other people. So it still makes things move
much faster than historically, but a lot of caution has to be
put in. Probably over index there on the
risk around the shadow it, but there's obviously other
challenges with adopting AI driven augmentation to develop
things like training tools. Do you want to just share a
(17:37):
little bit about anything else you can think of that might be
some of the, I guess, challengeswhen people go all in?
So I would actually recommend 5 clear practical steps, right?
One is start with simple augmentation use cases.
Things like the earlier examplesthat I've given, things that are
low risk that give you some immediate value.
(18:00):
Just maybe some troubleshooting guides or drafts or coming up
with the safety checklist right to use wipe coding to let
process owners, not just engineers create tools.
Make sure that humans review andthe data is clearly understood.
Human in the loop is maybe should be actually the term of
(18:21):
the year, not agents, right haveAI that is there, but the human
is in the center basically orchestrating and make sure that
things are flowing through. Now if you look at the
manufacturing industry and frontline workers, we recently
released special training for frontline workers, right, in
(18:42):
terms how do they deal with different scenarios, have skills
reminders or step by step job aids workers where they are.
So develop tooling you using these AI capabilities in terms
of more fitting those solutions to where people are executing
their work. Another thing that I think
(19:03):
whether you are in IT or you're in the production running
systems business is have clear AI champions train a few people
who really understand these technologies well.
So one of the things that we have done at Skill Soft is
whether you produce some learning content or some other
(19:23):
piece of asset, we have identified people who understand
the process of instructional design pedagogy, how to teach
something, how to design, how tocreate.
And then we have built tools forthem to augment them on how they
do this stuff, right? So understand, draw the
processes, understand clearly the processes and see which
(19:46):
steps could be using AI, right? Then last but not the least,
measure, continuously measure how some of these tools and
technologies are helping you, right?
See if you're getting the ROI and based on that.
Invest and build more. Invest where you believe you're
getting the maximum ROI from. Murali, I wanted to ask you a
(20:07):
question out of left field here,mate.
I know we're chatting about AI augmenting jobs, not replacing
them and ensuring that we have the human in the loop.
But I wanted to get your perspective on all the talk in
the media on AI displacing humanjobs.
So some of the high profile examples that we've seen with
IBM, Duolingo and Dropbox, for example, they're looking to
pause recruiting or even replacecontractors with AI agents.
(20:30):
And there's also been a bunch ofstats around the world from the
World Economic Forum and the IMFestimating that up to 40% of
employers expect to reduce theirheadcount by leveraging AI.
But I kind of feel like this is,you know, all the examples we've
seen are in the tech and the digital space.
I feel like in the manufacturingspace, where at least a few
years away, the work is much more complex.
The process has changed regularly.
(20:50):
They're more safety oriented andmore complex than then sort of
replacing repeatable processes and services.
But I wanted to get your thoughts on this.
I'm going to make a cliche statement, which is the AI is
not going to take the job. The people who can leverage AI
are the ones who are going to take the job.
Change management, updating yourskills, your knowledge.
(21:13):
That's the most important thing.I can't predict five years from
now where we're going to be because there's a lot of
research going on in terms of newer AI models and
technologies. But as things stand today, these
current set of LLMS are still prone to hallucinations.
Lot of errors. They are statistical machines,
(21:36):
so one small error in the beginning can cascade into large
errors. And especially when it comes to
places like manufacturing where you need 6 ninths of quality
checks. We are not there yet in terms of
AI. Even in the IT and software
areas, we are still not at a point where the AI could replace
(21:57):
humans. AI is a great augmenter.
It's a great tool for people to scale themselves.
This is like people who were using like spreadsheets came in
Think of a world like people whodoesn't want to use Excel.
You're not going to be a productor able to do it right now.
The modern form of a spreadsheetor in Excel is probably using
the LLM. And in terms of job layouts and
(22:18):
all of that, I think there are alot of high expectations that
were set for this technology three years back or four years
back. There was this paper, Chinchilla
paper, which talked about the scaling loss, where they said if
you put enough data in enough compute, AI is going to be so
smart that it could replace the human.
But now if you look at what's happening currently at this
(22:42):
point in industry, a lot of those people who are foremost
thinkers in the space are comingback and saying that the
realization of that with the current set of technologies that
there is a gap 345 years, maybe those gaps could be closed.
But right now, from what we see,these LLMS or these AI set of
technologies are not yet at a place.
(23:02):
Forget about replacing. I think it's more about it's a
tool, it's an augmentation. Humans have to be at the center
of any of these activities to ensure that they're moving
forward. And then another important thing
is the LLMS or the AIS or these tools are only good as the input
that is given to them. And who is giving all the
(23:23):
context and input is the human AI doesn't have the
consciousness yet. Maybe it'll come in, but it
cannot come up with its own creativity and its own ideas to
to solve. But if you give it a great deal
of instructions, very clear instructions, good
introductions, it produce very high quality, good output, and
(23:44):
you can make decisions. But then there has to be a human
who's at the center of it still to guide it.
The human activity is going to change as we move forward.
I think it's a very fun, great world to be in for many of us,
right? To build things in a new way,
solve new problems. But I think replacing humankind
(24:04):
has been said a lot, many times before in history, from the
steam engines to the printing press to all the modern
technology, the evolutions and transformations that have
happened. Yeah, I totally agree with that.
But I guess there's a big difference between using large
language models that are trainedoff data from all over the
place, which comes from who knows where, versus those
(24:25):
specific models trained inside businesses on that specific
business knowledge. And some of the startups I've
been working with, the ones thatare just leveraging that off the
shelf LLMS that anyone can use and beef up their business, not
so important and not so useful. But the ones that are building
specific models for their own, leveraging all their
intelligence that they've gainedover the last two years, five
(24:46):
years, 10 years, putting that intheir own sort of custom models
to extract out all that knowledge out of their own
systems is where they're seeing the value and being able to
leverage it properly. 100%, the biggest innovations and the most
transformative changes that are going to come are things like
alpha fold that Google worked tocreate in terms of understanding
(25:08):
protein folding. There are new models that
understand human DNA, but those are not just build on language
models, right? They are using the historical
AI, machine learning algorithms and a lot of other techniques to
build a combination purpose fit to solve problems and that's
where we will get the return. So I think even in manufacturing
(25:31):
there is a lot of opportunity tofigure out how do you take what
is there in terms of the company's knowledge and convert
it into mechanisms that you could apply to solve business
problems. I'm really enjoying this
conversation, but as is typical of a podcast, we have to wrap it
at some point. And so it's been really
interesting to hear a technologist's perspective on
(25:53):
where does AI fit and what does vibe coding mean for
manufacturing context. So I appreciate you coming on
the show to talk with us. I guess my key takeaways are
that these are great tools that are going to enable the
democratisation of technology solutions to solve on the floor
problems. But we just need to make sure
that we're putting the right governance around that so that
we know what use cases we will use it for.
(26:15):
And we make sure we're not usingit for high risk things and
we're not necessarily trusting the output, but the humans still
in the loop and taking ownership.
I think they're probably some ofthe high level lessons for me.
But I think just for the sake ofthose listening, I'm in a
manufacturing business. My staff are asking me for
access just these tools. What's the practical get started
tomorrow thing you need to do? Yeah.
So I think first step would be to first and foremost work with
(26:39):
your IT partners and team to enable tools, right?
Whether it's Copilot or Google, Gemini models have a very safe
AI system that you can consume inside your company.
So that IP and data is actually not leaking out into the public
LLMS, right. So first, a QPR employee base
with the tools, identify businesses and processes and
(27:02):
look at augmenting or automatingthe processes that give it the
maximum return for the business,right?
Things that could be easily automated, not just easily
automated, but could be automated with precision, right?
So that you don't have a risk toyour business, identify this for
low risk ones. And then as you progress,
upskill and train your people onhow AI could be used for
(27:24):
different roles. So we at Skill Soft have built a
curriculum around job roles and functions where they can use AI.
Have your workforce trained in terms of adoption because every
role could use it in some different ways.
A manufacturing company obviously has marketing, sales.
They could use AI in certain ways.
Then train workforce in terms ofwhat's the safety, what's the
(27:46):
governance, what is the life cycle around some of the AI
tooling to solve that business, business problems have IT
champions, AI champions who within the context of the
environment help take it progress forward because you
need someone to push things along, right?
Always understanding the domain.And so that having the notion of
(28:07):
the AI champions important constantly measure, monitor and
develop governance around this. So that's I think the key steps
I would say. We really appreciate you coming
on the show today, mate, and I know that we'll catch up again
soon, but thanks for sharing your insights and coming on the
show today. Thank you for having me.
Cheers. Thanks for tuning in to
(28:29):
Manufacturing Tech Australia with Shane and Paul recorded on
the traditional lands of the Bunnerong and Mouranjuri people.
The more information jump on theManufacturing Tech dot AU
website. Remember to hit the follow
button to join us again next time as we continue to explore
the intersection of manufacturing and technology.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
Hey Jonas!
Hey Jonas! The official Jonas Brothers podcast. Hosted by Kevin, Joe, and Nick Jonas. It’s the Jonas Brothers you know... musicians, actors, and well, yes, brothers. Now, they’re sharing another side of themselves in the playful, intimate, and irreverent way only they can. Spend time with the Jonas Brothers here and stay a little bit longer for deep conversations like never before.
Betrayal Weekly
Betrayal Weekly is back for a new season. Every Thursday, Betrayal Weekly shares first-hand accounts of broken trust, shocking deceptions, and the trail of destruction they leave behind. Hosted by Andrea Gunning, this weekly ongoing series digs into real-life stories of betrayal and the aftermath. From stories of double lives to dark discoveries, these are cautionary tales and accounts of resilience against all odds. From the producers of the critically acclaimed Betrayal series, Betrayal Weekly drops new episodes every Thursday. If you would like to share your story, you can reach out to the Betrayal Team by emailing them at betrayalpod@gmail.com and follow us on Instagram at @betrayalpod and @glasspodcasts. Please join our Substack for additional exclusive content, curated book recommendations, and community discussions. Sign up FREE by clicking this link Beyond Betrayal Substack. Join our community dedicated to truth, resilience, and healing. Your voice matters! Be a part of our Betrayal journey on Substack.