00:09Hello, you are watching Awani Review. I'm Cynthia Ng.
00:13We have become used to using AI to help us work faster,
00:18but now it is actually taking on the work itself.
00:22And we have seen some examples where projects used to take months
00:27being completed in just days with smaller teams.
00:30That is the promise of AI agents.
00:34But how practical is this for businesses that may not have large resources like a big tech company?
00:42And also, how much control are we prepared to hand over to these AI agents?
00:48That is what we will be discussing today.
00:51And joining me, we have Namjie Cho from Amazon Web Services.
00:55He is the Director of Technology for Asia-Pacific, Japan and China.
01:00Namjie Cho, great to have you here and really excited to our conversation.
01:04Thank you, Cynthia.
01:05So Namjie, let's start with this because you work with businesses across the region,
01:0911 countries to be specific in adopting new technologies.
01:13So we have spent the last few years talking about how businesses can use AI.
01:20And now we are talking about AI agents actually carrying out the work itself.
01:25How big a shift is this?
01:27I think the shift began when the generative AI hype cycle began back in 2022.
01:33Now, if you look at the journey of AI from when the chat GPT moment happened to today,
01:38there is definitely a shift in the way that AI agents are being used from just asking a question
01:45and getting an answer to moving on to actions and outcomes.
01:50Now, what we are seeing with our customers around the world today is so exciting
01:53because there is a big tech shift happening.
01:55And as part of that, the technology is becoming very accessible for every individual and every business
02:02to complete tasks very, very easily.
02:06Alright, let's talk about real world applications.
02:08How much of this are you seeing using AI agents?
02:11How much of this are you seeing in businesses today?
02:13I would say almost all of our customers who develop software and develop applications
02:18are using some form of AI coding agents today.
02:21So today, I think I can't imagine an organization developing software without an assistance of an AI coding agent like
02:31Amazon Kiro.
02:32We have lots of customers who also run Claude Code or Codex on Amazon Bedrock today.
02:40Okay.
02:41Well, I do also want to talk a bit about Malaysia because we have plenty of SMEs, smaller businesses.
02:46And I think the challenge is that while they are knowledgeable in basic AI applications,
02:54getting from that to authentic AI to AI agents could be quite a journey.
02:59So what have you been observing with their interaction with local businesses here so far?
03:03So in my experience, what I see right now with this AI transformation
03:07is the capabilities of AI is equally available to whether you're a large enterprise
03:12or you're a small to medium business.
03:15So what AI agents and these agentic applications will allow you to do
03:19is actually lifting the floor for everybody to have that equal playing field
03:23to be able to build really innovative applications to transform their business.
03:28Okay.
03:28So this idea that this technology is available to bigger companies or organizations that have more funds
03:37is just not true.
03:38That evidence that I see is that the technology and the capability of AI agents
03:43is just so accessible to everybody.
03:45Now, anybody that comes onto AWS today can sign up to a free account
03:50and give our Amazon Kiro a try almost instantly.
03:53Okay.
03:55I want to hone in on that.
03:56Accessibility.
03:57Cost is lower now, but how low?
03:59So let's say an ordinary business running a fairly small company, 50 employees.
04:05And what is that practical journey for them to get on board with using AI agents from AWS?
04:14Yeah.
04:14What I have seen work really well with these businesses, small to medium businesses,
04:19is that you start small and you start with a use case that's well known
04:25in terms of the problem statement inside your organization, where you know the challenge that you want to solve for
04:30and there are KPIs that already exist.
04:33And we have our desktop agent like Amazon Quick, which is really quick to set up.
04:38You connect it to all your data sources.
04:40And then all you have to do is to have a conversation with your AI agent about the goal
04:45and the outcome that you want to achieve.
04:47And it gets underway in solving your problem almost instantly.
04:50All right.
04:51So one area where we're already seeing some pretty remarkable change is our results in software development.
04:57And I want to bring in an example from AWS itself,
05:02where I think a project that was originally expected to take about 30 developers up to 18 months.
05:08You guys completed it in 76 days with just using AI agents and six senior engineers.
05:16That is quite a difference.
05:18What made that possible?
05:20I'm just trying to understand the productivity and efficiency of it.
05:23I mean, the great example you're talking about there is how Amazon's Bedrock service team
05:27rebuilt their inference engine on the service.
05:31And this project in perhaps before the world of AI agents would have taken anywhere from 12 to 18 months
05:38with 30 engineers.
05:40And for this particular project, six senior engineers were able to complete the project in just 76 days.
05:48Now, what a lot of people, when they look into this example, will say is that,
05:55was it the AI coding agent on his own that was able to achieve that outcome?
05:59But the reality is that there's a lot of things that needs to be combined in order to achieve that.
06:04It's the context of that organization that we had and the experience of the engineers who worked on that project.
06:10But I want to emphasize the fact that that example is very much about fundamentally changing the ways that you
06:18work inside a software engineering team.
06:20It's changing the workflow.
06:22It's optimizing every step of that process with AI agents where some of those steps could be what I would
06:30call easy decisions.
06:32So you hand it off completely and that particular step to become autonomous.
06:37Whereas there are some steps that actually require the experience of a very senior engineer to actually make a joint
06:43decision with the AI agent.
06:45But this was some of the most experienced senior engineers at Amazon, right?
06:48So how much of that result could an ordinary business realistically replicate?
06:52Yeah, that's a great question because what I love talking to our customers about is that one of the great
06:57things about what we do with our customers is that we get to learn from all of these experiences inside
07:03Amazon,
07:04as well as what a large enterprise in, say, Australia is doing or what an innovative startup in Singapore is
07:11doing.
07:12And we combine all of that knowledge and bring that force to our customers to share in those learnings, those
07:18methodologies, the best practices.
07:21So one of the things that I get to do with my customers every single day is to bring our
07:26learnings and make sure that they're not reinventing the wheel.
07:30And they're actually not starting from zero, but they have an accelerator that can start on with us.
07:36All right. Taking that example now, but if a team can work with far fewer people, doesn't that change how
07:43many people a company actually needs to carry out a task now?
07:46So I think that's a very commonly held belief in the world of AI today.
07:50And what I would say to you is that what I'm observing from my direct customer engagements across the region
07:58is that these customer teams that are building software and that are building really innovative solutions hasn't actually been reduced.
08:08What they're able to do is that as their AI agents enable them to finish their work quicker, be able
08:14to complete their tasks and outcomes faster.
08:18What they're able to do is to actually take on more experiments, more projects, and they're actually able to solve
08:24their existing problems within the businesses much, much quicker.
08:28It means that there is an acceleration of benefits that they're delivering to their customers.
08:35I do want to press on your point a bit because there is another consequence here while we understand the
08:42remarkable productivity gain.
08:44A lot of the routine work that AI agents are doing right now is precisely how junior employees learn, right?
08:52So if machines are already doing that work, automating the more junior work, the basic work, what then happens to
09:00experience, the talent pipeline?
09:02Just wanted to get your thoughts on what you're seeing in AWS itself.
09:06Yeah, I mean, I would probably start off by saying I would quote our AWS CEO, Matt Garman, when he
09:11was asked this very similar question in that,
09:14are we investing in junior talent that I was on? And he said, absolutely. You know, we believe that like,
09:19you know, these junior talent that come through the ranks that are gaining that human experience is going to be
09:25critical for a long, long time to come.
09:26And I see that across my own organization, where we're making big investments in hiring talent from what we call
09:33early career talent program, all the way up to senior talent.
09:36And we want to make sure that human experience is gained through the knowledge and experience of their peers around
09:45them.
09:45So I strongly believe that having junior talent that comes through an organization and continued investment is critically important.
09:54And so I think that's what our CEO, Matt Garman, has very clearly stated on multiple interviews as well.
10:01Okay. And when you meet businesses, local businesses especially, of course, investing in AI or AI agents while the cost
10:08has come down, it still requires a bit of money.
10:11Right. So I think naturally businesses are asking where are the returns and, you know, am I investing in the
10:18right systems in the right places?
10:21Right. So can you share with us, you know, what are some of the considerations that leaders in the company
10:27have concerns with when you speak with them?
10:30When it comes to valuing the return?
10:32Yeah. I think there are a lot of concerns today that I would acknowledge that as they start to look
10:37at some of these use cases,
10:40I think cost has certainly been a big consideration.
10:43And there is certainly concerns from business leaders around potential cost of AI.
10:50And right now what we're learning is that there isn't going to be one or one or two models that
10:57actually is going to be able to address all use cases.
11:00So we have a service called Amazon Bedrock, Amazon Bedrock. And what we provide on Bedrock is a choice of
11:07models. Like you don't need the world's smartest models to do very simple tasks.
11:12And also there are some really intelligent models that are really optimized for very complex tasks.
11:21But where we have been able to advise our customers on the best use of some of these models is
11:28we understand their business use case and the goals that they're trying to achieve.
11:31And then we work backwards from the outcome that they want to achieve. And right now on Amazon Bedrock, we
11:37have a number of model choices where we have the two front team model providers from Anthropic as well as
11:43OpenAI.
11:44But also you would have heard about the OpenWeights models where they are significantly cheaper, they're still very intelligent.
11:51And so the level of intelligence that these OpenWeights models has is able to achieve customers' goals for a very
11:57low cost.
11:58Okay. I do want to bring up this research by Gartner, the research from Gartner, which forecast that more than
12:0540% of agentic AI projects could be cancelled by the end of 2027,
12:09and largely because of either unclear business value is one of the reasons.
12:15So I do want to get a sense of like how should a business work out and whether it's actually
12:20getting its money worth when they're investing in AI or AI agents.
12:25Yeah. So I would just anchor on that sort of snippet that says unclear business requirements.
12:31Well, there are a few reasons. One of them is a rising cost, unclear business value and inadequate risk controls,
12:36which we'll go to later.
12:36Yeah. I mean, let me sort of pick on the unclear business value because, you know, we've been working with
12:43our customers well over 20 years now.
12:45So, you know, AWS Cloud, it was a business that was established back in 2006. Can't believe it's been 20
12:50years already.
12:52And what we have been doing with our customers all the way through our cloud transformation is that we want
12:59to understand the customer's business need.
13:01What is the problem that they're trying to solve? And then we work backwards from our customer's requirements.
13:07Now, it's very important in whether it's in the cloud transformation, digital transformation or AI transformation, is that you clearly
13:14understand what is the business problem?
13:16What are the measures of success? And then you start working backwards from what the problem that you're trying to
13:22solve for.
13:23And I think often when a new innovative technology lands, people want to play around with it, people want to
13:29experiment.
13:30And often what we find is that if there is an unclear requirement or a KPI for that particular project,
13:38then the chances of success may not be as high as if you had a very well defined business requirement.
13:44You have the KPI for success. And then once you achieve on that sort of project outcome and say, did
13:51we hit our KPIs, you have a clear measure of success.
13:54And I think for us, it's important to work with our customers to help them to teach some of these
13:58key learnings that we've had over the years in making sure that every experiment starts off with a clear business
14:06statement.
14:07We believe that it's very important to have a clear business owner as well, because often when a new technology
14:15lands, there's always an inclination towards, is this a technology problem?
14:20But actually most challenges inside an organization is actually in their business.
14:24Or the human.
14:27People play such a critical role in the success of the business and the technology adoption as well.
14:32So define a measured outcome as one of the key areas to look into to measure success.
14:39But what are the other things that when you speak to a leader of a company, what are some of
14:44a characteristic of a tech leader to ensure that the transformation is well executed?
14:50I think you've hit the nail. It's the people.
14:53And I think what we also learned from our early days cloud experience is that skills and that people change
15:01element is so important.
15:02Because when you look at agentic AI today, even with the tools, your ways of working changes.
15:09And in many cases, you have to be able to go through that change journey in order to adopt the
15:15technology successfully and to achieve on your business goals.
15:20So I think that skills and training and the right set of enablement and people's willingness to go through that
15:26journey from this is how you work today.
15:28But this is potentially how your job and what you do changes in order to get that better outcome.
15:35Okay. Now I want to pivot and talk a bit about trust and control because now we are, I mean
15:41the whole appeal of AI agents is that they can take action without someone guiding them every step of the
15:46way.
15:47But where do you think we should draw the line on AI agents being useful and giving it too much
15:54control? Is that a concern? Is that a worry even at AWS?
15:57Yes.
15:58That's a really great question. And I would start off by saying that no organization starts their first AI project
16:05and hands off full control to an AI agent and gives it 100% autonomy.
16:12And sometimes the mental model around autonomous agents is that you either give them full control or none at all.
16:20But it's not a binary decision in that many of our customers when they start with a coding agent or
16:25Amazon's quick, that is a fully human in the loop AI agent where every important decision requires a permission from
16:34you to proceed to the next step.
16:36Now, this idea of handing control to an AI agent, autonomous agent is an incremental step.
16:44You experiment and you deploy that first use case that is low risk.
16:49You learn how to operate that because agents still need to be operated.
16:53And so once you learn about how to monitor it, how to make sure that it's behaving in the right
16:59way, you have the right evaluation and how you monitor that performance of that agent.
17:04Then you add on top and you iterate you more one step at a time.
17:08So I want to talk about when AI agents are not behaving in the right way.
17:12So when things go wrong, maybe AI agents get things wrong.
17:16Who is responsible? Because you're working with partners, businesses are using your technology.
17:20So is that something that AWS look into?
17:23Absolutely. And I think this is something that we're learning with our customers day by day.
17:27And we get a lot of feedback from our customers around some of the things that these AI agents are
17:32really good at.
17:33But also our customers give us feedback that there are certain things within deploying AI agents can actually be quite
17:41complex.
17:42For example, in order to deploy an AI agent into your business environment, you need to make sure that you
17:48think about the identity these agents undertake.
17:51You want to make sure that they have the right knowledge base, the context to make the right decisions.
17:56You also want these agents to have some memory so that you're not repeating the same thing each and every
18:01time.
18:01So in order to address some of these complexities, and this is something that is actually quite difficult to set
18:08up for an organization.
18:10So what we did was we listened to our customers and then we built a service called Amazon Bedrock Agent
18:16Core,
18:16which actually packages up all of these capabilities into a service that our customers can safely deploy these agents.
18:23Has things gone wrong before? How do you mitigate that?
18:28So I think it's important for me to emphasize that this is a very new technology and we're learning as
18:34we go.
18:36But what we, you know, I think security and trust has always been really on the top of our priority
18:43list.
18:44And so we want to make sure that anything that we develop and we make available to our customers is
18:51in response to their business need.
18:53And I think one of the things that we want to make sure is that as customers scale and adopt
18:59agents inside their businesses,
19:02it's important that they do it with confidence and trust.
19:05So we now have some services that we offer to our customers that actually address some of the domain of
19:13security, for example,
19:14where it helps customers to do the really hard things that were not easy to do in the past.
19:20But things like we have a service called AWS Continuum where it does code scanning, vulnerability scanning.
19:28It does penetration testing.
19:31And these are all capabilities that were really hard to implement inside an organization until agentic.ai came along.
19:39All right. So I know you look into 11 countries as under your remit.
19:43But I want to bring it back to Malaysia because Malaysia is seeing growing AI adoption.
19:48But research shows, this is AWS's own research, is that AI adoption in Malaysia has reached 38%,
19:55but two thirds of adopters are still using fairly basic applications.
20:00Let's talk about what's stopping them from going further, especially when they try to move beyond that pilot stage.
20:08Okay. I think this right now, there are varying different degrees of adoption and adoption maturity in the region.
20:19And what I would say is that the really easy place to get start for most organizations is an AI
20:27agent like Amazon Quick or Kiro.
20:29And then as they learn more about the capabilities of these agents, there is steps to maturity in getting these
20:39agents to solve more complex problems inside your organization.
20:43Let me put it this way. What do you think businesses underestimate when they start using something like an AI
20:49agent?
20:51That's a really good question. Underestimate. I would say most organizations that are trying to adopt AI today is so
20:58new to everybody that it's a learning journey.
21:01And I think organizations that I've seen succeed and that are moving quicker with AI than those that are not,
21:07ones that are actually focusing on people training enablement and making sure that it's actually tied to a business outcome
21:15and also measurable business outcomes with clear KPIs.
21:19So the end-to-end workflow from the inception all the way to the delivery of the outcome is well
21:25planned and measured.
21:26And I think that's something that these customers who have reached that maturity have gone through that journey from early
21:32adoption of AI to where they are today.
21:35So it's the maturity journey that they've gone on to, to reach that where they are today.
21:40In Malaysia, where do you see, what stage are you think most businesses are at in terms of adoption?
21:45I wouldn't generalize and say they're all in certain one or the other place, but I come across some very
21:53innovative startups to large enterprises that are adopting AI at scale here in Malaysia.
21:59I see lots of really dynamic environment with SMBs as well as SMEs and startups that are just starting out
22:07their journey.
22:07But what's really interesting trait amongst all of them is that they're all moving very, very quickly because you can
22:14pick these things up and just give it a problem.
22:17And you realize that it can solve some problems that you haven't been able to solve before.
22:21And the next step, they're experimenting on their next use case and the next use case.
22:25So that's what's so exciting to see.
22:27Okay.
22:28Well, just a final minute that we have now, if you're advising a Malaysian business today, what would you tell
22:33them about investing in AI agents if they are just getting stuck?
22:38If they're just getting started?
22:40If they're just getting started?
22:40Yeah.
22:40The barrier to entry, or I would say the floor has been lifted up for everybody and you can get
22:46started in minutes and you can experiment very safely.
22:50And as I said earlier in our conversation, I think like the mental model of people has to change from,
22:57it's not just good at answering questions.
23:00It can take actions and it can drive outcomes and it can do it very quickly.
23:05So I would say just, you know, learn and just get hands on.
23:08Okay.
23:09Well, thank you so much Namjie for spending some time to talk to me about the opportunities of AI agents.
23:14And great to have you.
23:15Hope to have you again soon.
23:16Thank you, Cynthia.