🦄 Making great presentations more accessible.
This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.
Overview
📖 AWS re:Invent 2025 - A leader's guide to emerging technologies: From insights to rapid action-SNR203
In this video, former CIO Arvind and NASA's first CTO Tom Soderstrom address the dual mandate facing technology leaders: delivering operational excellence while championing innovation. They present eight major technology trends including generative and agentic AI, edge computing with IoT, collaborative robots, medical impacts, data analytics, crypto, quantum computing, and energy solutions. The session introduces TechRecon, an open-source multi-agent AI system that automates technology landscape reconnaissance for CIOs. TechRecon generates emerging technology reports and position papers specific to each company's industry, calculating impact, maturity, and momentum scores to help leaders decide whether to deploy, pilot, or monitor technologies. They emphasize building a "center of engagement" rather than a traditional center of excellence, measuring success through return on attention (ROA) instead of just ROI, and creating a flywheel of innovation funded by reinvesting operational savings.
; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.
Main Part
The Dual Mandate: CIOs Caught Between Operational Excellence and Innovation
What's our AI strategy? I want efficient customer service. I want automated customer onboarding. I want faster accounts payable. The board wants an update on our view and plans for the impact of quantum on our business for the board meeting tomorrow. Show of hands, how many of you felt like that pilot in that scene we just saw?
So as a former CIO, it's been more than a few times that my CEO has walked into the room asking, "Hey Arvind, what's going on with this new technology?" And in that instant, I shift from being a trusted adviser to a scrambling researcher. Now at AWS, as we talk to more and more customers and CIOs, this pattern is repeating more and more often. The reasons are very clear.
Over the last 25 years, a range of technologies have completely transformed how industries in every sector and every function have been transformed more and more by technology, and this is only accelerating. This acceleration is powered by a combination of technologies that are creating more and more impact, whether it's AI or robotics or biotechnology. The cool thing really is that it's not just that these technologies are accelerating independently in isolation, but they're actually feeding into each other, creating an exponential impact that's accelerating impact for businesses as well.
The interesting thing is that this is not just something hidden somewhere. The media is talking about it, vendors are talking about it, there are industry events, and your business leaders are going into those events and learning about it there. They're wondering why they're not hearing this internally in your own organizations. This is a major challenge that we've got to overcome.
The real challenge is that our core work of technology delivery is not getting any easier. Our work is getting more and more critical, more and more central to the business, and delivering excellence is absolutely non-negotiable. But at the same time, technology leaders are also being expected to be the innovation champions, the scouts, the people who bring in and reshape existing organizations to be more modern and digital. This dual mandate is making things more and more complex for our roles.
Building a Digital Kampong: Creating Networks for Technology Experimentation
I was a CIO, but I joined AWS a year and a half ago. Before that, I was a CIO for Kellogg's based in Singapore, and before that with Prudential Insurance and Procter and Gamble. Over the last 10 to 15 years, I've seen this pattern repeat again and again where business leaders are getting more and more interested in technology. The way I've responded to that is I've leaned into it and created a network of people around me in the technology organization and in business that I used to call the digital kampong.
Kampung is a Malay word. I live in Singapore, so Malay is a language that's used there. Kampung is a Malay word for village, and this is a situation where you want people to come together to address this challenge. What I did is work with enthusiasts who were keen to learn new technology and had that learning culture in them. I worked with them to do a lot of experiments to test out new and emerging technologies.
Ten to 15 years ago, a lot of this stuff was very new. One of the first things you see on the top left is we built Alexa skills. Amazon had just launched Alexa then to figure out how consumers would engage with voice systems. We built IoT sensors that helped manufacturing teams figure out how automation can be accelerated, and a number of other such things. Over time, we did more and more of this, eventually including the families of some of our employees because nothing gets people excited more than having their kids learn new technology as well.
There were a number of learnings I personally had that we will share about this now, and I'll invite Tom to share his background on this space.
NASA's Approach to Technology Trends: Predicting Impact Over Predicting the Future
Thank you, Arvind. My name is Tom Soderstrom, and I was NASA's first CTO at NASA's Jet Propulsion Laboratory. The slide you see here was our attempt at how we could spend less money on IT and more on science and engineering. We looked at cloud computing as one of the answers, and yes, it worked. You can see the journey here from 2008 and beyond. The idea was that if we could use cloud and AWS saw a business for it, then they would build it and we would help launch it. That's how we came up with so many of the services that you use today.
The secret here for you all is that if you're interested in something that AWS is interested in, you have an extremely strong voice—not just for what is built, but how it's built. We're going to talk about some trends here. I was the futurist at Jet Propulsion Laboratory as the CTO. Who in here considers it their responsibility to keep understanding what the technology landscape looks like? Show of hands. Good, then we have a gift for you, because it's a difficult job, honestly.
It was fun to hear you talk about Glacier or Alexa, so she's not listening, is she? One of the first things we saw is that when we came up with GovCloud and AWS built it, it created psychological safety for people to innovate in all kinds of ways. Have you ever heard of a two-way door decision? That is the way of using the cloud. We created the first NASA Alexa app, and it's called NASA Mars. It still runs, but this is about technologies.
You have this dual mandate of figuring out what are the technologies that are coming and at the same time keep the plane flying. We're going to approach this in two pieces. We looked at the trends that are coming manually, and we're going to share what they are. At the end of this, I'm going to ask you to bring your cell phones up and see what you care about and which ones you think are interesting. Then we're going to show you how you can do this automatically. This will be our gift to you all.
What are the major technology trends? You are already looking at it, and the point is not to predict what's coming in the world. The point is to predict for your company how it will make you more productive, more profitable, and how it will save money using these new technologies. If you build it, they will come if you guess right. If you don't want to build it, then somebody else will build it, and that's how we worked at NASA.
When you are looking at the future and you are wrong, it's easy to be ridiculed. You can see some of these fun ones here. I'm very happy that he was wrong about the airplane, otherwise we'd be speaking to a very empty room. If you predict right, you can get a first mover advantage. You can excite your people to participate. You can impact the creators of what's being built. You can also detect very early that you're way off and this trend doesn't matter.
Generative and Agentic AI: From Experimentation to ROI
We'll go a little deeper into each of these, but these are the eight that we saw were interesting. Two of them are more questionable than the others, and I think you can guess which ones they are, but we'll find out. Cloud and generative AI belong together. You really cannot be successful in generative or agentic AI without using the cloud. It just needs that much horsepower.
In 2020, we were experimenting and trying to see what we could do with generative AI and beyond. In 2025, we're generating ROI, return on investment. Where are we? With a show of hands, how many people here have in production a generative AI that is generating an ROI? Raise your hands very proudly if you do. It looks to be about five percent. What about agentic AI? Do you have something in production? One person, two, three, four. I feel like a salesman here.
Is it generating ROI? Why? Next year when we ask the same question, a lot of hands will go up. That's the nature of these technologies. The estimate is fifteen point seven trillion, which is a big number. New jobs will be created. We think about AI stealing our jobs, but it's actually going to create new jobs. Your opportunity is to create those jobs and staff them with your own people to see if they pay out. If they do, you're way ahead of the game, and if you're right, then you couldn't afford to hire those people later anyway.
So it's a good way to just brainstorm what are the new roles that will come. Now when we think about it, I thought about this long and hard, what is AI in Rio?
So remember hailing a taxi. Actually, is anybody here from New York? You probably do that today. You trusted that they would stop for you. Then later on you trusted that if you clicked your smartphone, they would come and get you. What we're seeing is this judgment is now that the self-driving car, you will trust that the car will get you there safely. So it's all about judgment. This happens to be Zooks. Has anybody been in a Zooks? It's here in Las Vegas. It's fun. I highly recommend you do it, but it's this trust.
And when we look at AI at scale, anything from anybody use Rufus to help you shop, it's awesome. Ask for a coffee maker that is easiest to clean, and it'll give you an example. The other thing that we don't see is regular machine learning is not going away. For instance, Rufus on the back end filters out 275 million fake reviews, so you don't have to look at them. It turns out the reviews is one of the most popular things. So generative AI, the benefit of this is that this is working at scale.
So it's AI, it's moving forward, we're already there, but what's coming? It's where AI is becoming a personal assistant. It's where the future of education. Anybody have young children who ask why over and over and over? Well, AI doesn't get tired. AI will engage the student and it will help them learn the way they want to do.
Amazon CTO Verna Vogels said, and I'll quote this, when you use a tool to engage curiosity instead of enforcing compliance, schools spring to life and that changes everything. I think that's really key. And for the teachers, they can now use generative AI and agentic AI for the drudgery, grading the reports, and so on. So it's really about how we use AI. So you'll hear a lot about generative AI and agentic AI at this conference because it is the future.
Eight Major Technology Trends: IoT at the Edge, Robotics, and Medical Impacts
If you went through the cloud evolution like I did, this is at least as big, if not bigger, and it's really exciting. If you went through the cloud evolution, take those lessons learned, and you can now go through the agentic revolution. So productive at the edge. Anybody heard the word IoT? You're tired of it, it didn't pay out, didn't have any ROI. Well, that's changing. Because we have 20 billion devices today, by 2030, we'll have 40 billion devices, and they will be connected, and they cannot send data all the time.
So what we're going to see is that the return on investment between business outcome and being productive at the edge is going to be key. Today we talk about large language models. Tomorrow we're going to talk about nano-language models where the AI will run on the device. Better silicon, smaller models, and it will create a digital twin. I'll talk about that in a second. But what are some people who are doing this well?
Kone, anybody know who that is? Anybody ever ride in an elevator or an escalator? They keep the cities running. They have 30,000 maintenance technicians. It's a Finnish company. They go on 80,000 customer calls per day, and 30,000 customer calls per month is filtered out because of agentic AI. They have everything at their fingertips. When they go to fix an elevator, they know the version, they know the history, what usually goes wrong, and they're much more productive.
So IoT at the edge is really powerful. We also have AI in the jungle. I just like saying that. AI in the jungle is Hexagon, a Swedish-Swiss company, has a subsidiary called R Evolution, and it's about measuring biodiversity. So they have a device that sits in the jungle and measures biodiversity over time. It's sitting at the edge and it reports back when it has connectivity. And that's nice because we do want to save the environment, but if you're a mining company and you want to do a mine, you have to prove that you are actually not disturbing biodiversity.
So there's business reasons for AI and IoT at the edge. So we're going to see this massive digital scale as twins, digital twins. And what it means is that they're collecting the data at the edge, they're reporting it back into the cloud. And now you have a copy in the cloud of what is in the physical space. So now you can start doing what if analysis. You can inject faulty data.
And have a conversation with all of that data in the digital twin using generative AI. Anybody who's been affected by a wildfire or knows somebody affected by a wildfire? I live in LA, so I'm very affected, but it's coming faster and faster, and natural disasters are coming more and more frequently. How do you deal with it? This is where we see the rise of augmented reality again, where you can insert people into a difficult, dangerous, or far remote place without them having to be there physically. They can interact with the digital twin to see how fast the fire is spreading and what it detected early.
There's a company called Voxalis.AI on the third floor of the Venetian here, and they have a device that they put on helicopters. Now the helicopters can measure whether a fire spot is coming up or whether they're in the middle of fighting the fire. It's a startup of about twenty people, and what we're seeing is the benefit of having startups working with governments and universities, especially on things at the edge and AI. If you're in that space, you're in luck because it's growing fast. Now they will need access, and if you've heard of Kuiper, forget that it's called low Earth orbit, and that is to be able to provide internet access to underserved places in the world. They're also here at re:Invent, and I highly recommend having a conversation with them because it's much cheaper, much faster, and has lower latency than anything we've ever seen before. Countries can now get backup if the fiber was cut, and the devices at the edge can get internet access.
Number three here is robots and robots at scale. What on earth is a cobot? Collaborative robots. We're seeing people and robots working together, and it's a $165 billion market in just a few years. That's a big number. All of these we picked because they're very large dollar amounts and they have a very large cumulative annual growth rate. These robots are amazing. They're everywhere. They're on the moon, of course I care about that being from NASA. If you think about something unpleasant for a second, wars are being fought with drones, but what we're seeing is much faster adoption in the civilian space from things that grew up in the defense space. We're seeing people who fight fires use the same technology that people who fight wars. We're also seeing people actually form relationships with the robots. If you have a Roomba, raise your hand. Keep your hand up if you named it. Half of people name their Roomba. In Sweden, where I grew up, they have automatic lawnmowers, and 100 percent name them. It becomes a pet. We find that these robots in hospitals help kids interact with them and be much calmer. Aging people will have a companion and they need less medicine and less care. So it really becomes key.
For us at Amazon, we've been in a fulfillment center. If you haven't, I highly recommend it. It's a toy box for robotics. If you look at the person on the top left, she is stuffing things that you bought, and the red boxes say don't put it here because it's close to something that looks similar. So she doesn't have the cognitive load of having to decide where to pick it. It saves a lot of time. On the right is a robot doing the same thing, augmenting it, but going high and going low to save her back. In the lower left-hand corner, the picker is giving a right angle where to pick, picks it from there, picks it, scans it, done. Fewer errors, much faster. And then we have the SLAM process where in those few seconds that you see there, this is actually in real time. It is taking all the big data and putting it on the label with who it is and where they're going. The box is already sealed, so there's complete privacy. Now, how big is this? It's huge. The fulfillment centers, for instance, delivered 600 million packages the same day. And that is going up. So the speed is about three hours from start to finish in the fulfillment center. The average delivery is 1.9 days from when you bought it. And the average time to buy is about three minutes. So this is speeding up and going faster, and robotics is helping. But how do they deliver?
In the last mile, you have the driver now being able to use the same technology. You can see there it's going to shine a red light—don't pick that one—and then a green light, that's the one to deliver. That saves a minute per delivery, which is a lot. Now you add augmented reality glasses, and they can do the same thing hands-free. Everything builds on everything else here.
What we're seeing is medical impacts at scale as another one. In 2020, we never went to a remote doctor visit. Then came COVID. In one month, it increased 157 times. Now we're looking at it, and you can actually have a doctor's visit. We're wearing wearables, we're counting ourselves, asking can we sleep better, and so on. That is going towards a personal AI guardian. Genomics is an area that uses a lot of data and a lot of AI, and in the future you can even get a prescription before you even know that you have the disease because it looks at the genomic data and tells you what you need to do. So it becomes your guardian, and it's a big business.
This genomics is not just good for people, it's good for koalas. I want to tell this story about a researcher in Sydney, Australia, who had a lot of genomics data, but she didn't have the people and the skills to analyze it sufficiently. So she put it in Amazon's open data exchange, and a researcher in Sweden was able to solve it in just a few days. Do you know how you save the koala? You vaccinate them from chlamydia. Who would have known? Well, I'm happy to say that two months ago that vaccine was actually approved in Australia, so there are big hopes for the koala. The point of that is, if you share the data and you share how you can work faster, you can do miracles.
Digital twins is not just for things, it's for humans. So if I have a disease, they can have a digital twin of me because everything is being measured, and they can see where they should operate before they do it. Now they can bring in robotics and have somebody from outside actually do the surgery. The medical impacts are fantastic, and we're going to need them. We lived until about 47 years old in 1900, 78 in 2000. In 2100, I don't have a clue. But we're going to live longer. We will have implants, and AI is going to be what keeps us healthy. So it's not just a new thing, it's very personal and very important.
Data, Crypto, Quantum Computing, and Energy: Completing the Technology Landscape
The fifth area here is data. There's no surprise to any of you that data is going faster. Does anybody know what the biggest number we measure is? Yottabyte. Yottabyte was impossibly big. Well, we're going to get there in 2030. So now you need to learn a new word: quettabyte. A quettabyte is 10 with 30 zeros behind it. So you learn so much here. But how are you going to analyze all this data? We're seeing agentic AI as being an ability here to analyze the data that's coming and to deal with it differently. The data is not just structured; 90 percent of it is unstructured. How do you get your arms around all of this data in your company? The answer, as uncomfortable as it is, is you won't. You won't get your arms around all the data, but that's okay.
Just think backwards on the problem and think about the problem you're trying to solve, and then use generative AI to tie into partner's data. What we're thinking here is flipping the script. That's the easiest thing we have seen successful companies do. Instead of you having to fill out a form to use my data, I have to fill out a form to protect the data, and it changes everything because now you can experiment using AI and Internet of Things. It creates a lot of data, and now you can start analyzing it. You can also use synthetic data. Is anybody using synthetic data yet? Okay, very few.
So this is the hand scanner called Amazon One. I'm not on a run; I go into an Amazon Fresh, pick up a bottle of water, just hold my hand over and run out. That is because they trained it with synthetic data, and it's 99.9999 percent accurate. It's more accurate than an iris or a fingerprint, but it was the synthetic data that made it happen. So that's the opportunity here with data. The sixth out of eight is crypto, and the crypto revolution—
we may or may not be right, but if you had invested in Bitcoin ten years ago, it would have been the single biggest investment you could have made. However, it's volatile, so it's worrisome. But now we have regulations coming and there is growth, and blockchain keeps advancing. So the token economy means that you will have digital tokens for physical things. That changes everything. Have you ever thought about how a software agent will pay another agent? They'll probably use stablecoin and micropayments—a fraction of a fraction of a fraction of a penny. So all of a sudden, it creates a new economy, and decentralized finance may be one of them.
Then we have quantum computing. We've been worried about building the hardware, and in the left-hand corner is Amazon's quantum lab at Caltech, located in Los Angeles. Now we've actually made error-correcting qubits. A qubit is not a qubit if it's error-correcting—it's 90 percent faster and better than a non-error-correcting qubit. It's a large market. Where will it go? You can see it's going to have a big impact and create a lot of jobs. Where will it go in terms of business applications? That's going to be the question, and when. We think this is one of the key trends because of the growth rate and the amount of money being invested, but what applications will we see? Which quantum computer will win?
These are four completely different quantum computers. We don't know which one will win the ring that rules them all, but Amazon provides choice. You can use Braket to try and see which one makes the biggest impact. The important part here is to get your programmers to try using quantum computing to see what you can do. Then we have Airbus and BMW looking at being able to do logistics and soundproofing, with a lot of near-term opportunities. The biggest one of all is quantum-safe encryption. That's where you worry about future quantum computing decrypting your data. That's being worked on and is the single biggest opportunity right now, and you don't have to wait.
Use AWS quantum-safe encryption algorithms. Others do too. The key here is to use it. The last one is: do we have enough energy for this? We use fossil fuels, then we use renewables, and Amazon is the biggest producer of renewables and also a major purchaser of renewables. Now we're looking at nuclear power. It doesn't produce fossil fuels, but we're going to need all of this energy. It isn't just about energy to power things like windmills; it's about having the energy to handle this and having this dual mandate.
So if you pull out your smartphones, which of these emerging technologies will matter the most to you? There's the QR code. If you do that, you'll be able to see which ones of these you care about. Click on all of them, click on none of them, and we're going to see the results in just a few minutes. So it looks like genetic AI and data and energy—that's a surprise. I didn't expect that. Crypto, not so much. Quantum, somewhat. Otherwise, it's an even split. I think it's surprising that not more people are focused on edge computing, but this is why what matters to your company and having to do this on your own is hard work. So isn't there a better way?
Introducing TechRecon: An Agentic System for Automated Technology Reconnaissance
Absolutely. In the last twenty minutes, we've done a whirlwind tour of eight technology trends that we felt were important, but they may or may not be important for you. This thing is moving so fast that you've got to build your own capability to stay in touch with technology trends. What you need is to upgrade your cockpit with a radar system that can do this reconnaissance for you. What you need is a radar that can sweep the landscape on a periodic basis and identify what technologies are starting to appear on the landscape.
More importantly, you need a systemic way that can then identify those technologies, understand how far they are from impact with you, how fast they're moving, what the business impact potential is, and how mature they are to make an impact for your business. That's something you need to continuously do and be able to share with your business leadership to stay ahead in this dual mandate.
Based on this data, you also need to be able to tell if this is a technology that you should be deploying in your business right now because it's making an impact in your industry and among your peers and competitors. Or is this something that is close to making business impact, and therefore you should be starting pilots or experiments? Or if it's something that is less mature but moving fast with potentially big impact, this may be something that requires monitoring. This is a lot of work, and Tom and I in our past have done this manually using that framework to help us in this journey, but things are much easier now. This is 2025, the year where agents can help you solve this problem.
We are not just suggesting that agents can solve this problem for you. We actually have a solution for you. Here is the backstory. When we were doing the research for this talk, Tom and I, as often happens at Amazon, we wrote a paper on this topic saying what CIOs need today is a systemic, automated way to scan the technology landscape and do these assessments and understand what is important. As again happens at Amazon, this paper gets distributed and people add to it, provide inputs, and some of them say, "Hey, I'll build this for you." I want to introduce Jiyun Park, who is part of our solution architects team. She raised her hand and volunteered, saying, "I'll build this for you." What we are going to show you today and give you today is an agentic system that can automate this entire reconnaissance work.
Thanks, Arvin. Hi everyone, I'm Jiyun. Today I am going to walk you through what is TechRecon powered by Agent AI. Before I show you the result, I am going to briefly explain the architecture of TechRecon behind the Agent AI. TechRecon is a multi-agent system that researches public sources like newspapers, consulting companies, and IT vendors and generates a series of documents to IT leaders so they can understand the fast-changing emerging technologies and deep dives where they need them.
We have a supervisor agent here which orchestrates the entire workflow across the sub-agents. We define TechRecon by a prompt so it is going to support the CIO of the XYZ company in researching emerging technologies. We iterated with many different prompts and finalized this with the best one. Inside the system, each agent has a specialized role which orchestrates the work on their respective tasks. The planner agents create tasks for the other agents. Researcher agents gather the information and actually research on behalf of humans. Coder agents read, save, and create files, and reporter agents synthesize everything and make it as a final output. These agents work in a sequential way.
Based on the user request, there are going to be two major outputs. Part one is emerging tech landscape analysis, which is going to cover the broad landscape of technologies. Part two is position papers on specific technologies, which dive deep on one specific technology and create a report of your company's position papers. This is the strength of an agentic AI system with multiple agents that have specialized roles. They work together using their tools and make the final outputs autonomously.
As we were working together on this, what we realized is that not only can agents do research for you, gathering all of these insights and converting that into summarized documents, but they could even do the analysis we talked about earlier. If you build a slide, what the agents were able to do using coding capabilities is create three measures, three scores for every technology that it found. We calculated an impact score, and this is based on the work that was actually done at NASA on technology readiness scoring.
The way this happens is very specific to your company. June mentioned earlier that it's designed for your company. So let's say you are the CIO of a pharmaceutical company—it will do the analysis and the impact analysis for your industry. Similarly, maturity scoring measures how ready this technology is for use today, and finally momentum measures how fast it is moving—whether it's accelerating or slowing down. For every technology, we were able to create these three scores.
Once you have those scores, we were able to build one more assessment. Based on those scores, you could actually reach some sort of judgment on whether this is a technology that has high impact and high maturity and is already showing examples of use cases making an impact in your industry. If so, that's something you should deploy right now. But if it's earlier in maturity, then that may be something you want to pilot or experiment with. Whereas if there's something else that is in very early stages, very immature but moving fast with potentially big impact, that's something you want to monitor. These were some parameters we chose as we were experimenting with this, but you can pick your own parameters and create how you want to assess and decide what to do with each of these technologies.
From Center of Excellence to Center of Engagement: Building the Human Side of Innovation
Now June will show an actual demo of how the system works. Before I show you the demo, I'm going to briefly explain the front end of the tech record. As you can see in the middle, you can see the real-time logs as they are being streamed. These will be the outputs of each agent displayed on the middle of the screen. On the right-hand side, you can see the control panel where you can see the status of each agent. Below that, there's a button so that you can execute the run to create the reports—part one landscape analysis report and part two technology position papers. Once all those tasks are completed, you can download all the files in the bottom right corner.
After the demo is completed, you can see the result, which is the emerging technologies reconnaissance report, which is part one. This is a real result that you can utilize in your real world. This is an example of part one report for the CIO of a pharmaceutical company. If there's anyone here from a pharmaceutical industry company, congratulations—you can get this report today. Otherwise, you can run this code in your local system and then replace the variables with your industry and your company name. Part one is going to cover the overview of the technology landscape, strategy recommendations based on the score assessment that I explained before, and the industry-specific context as well.
Because TechRecon follows a strict, fixed template, you can run it monthly, quarterly, or whenever you want to catch up with new emerging technology trends. After creating the landscape analysis report, you can create the technology paper, which dives deeper into a specific technology. Here you can see generative AI. As you can see, it includes the key findings, strategy recommendations, technology deep dives, and also the real-world use cases you can utilize. TechRecon not only covers the technical aspects but also the business side. So you can see why this matters for your organization.
TechRecon is not a new service—it's a blueprint so that your teams can deploy it within a few weeks. On average, it's going to cost them less than two thousand dollars per year, but it depends on how much you're going to run. So it can cost less or it can cost more.
Many companies don't know where to start. They actually want to build agentic AI, but they don't know where to start. This is why the Tech Recon is so powerful, and you can start it today. We're going to give you the code architecture and the implementation guide today so you can run this and build your own strategic advisor with the Tech Recon. So this is the technical thing. Then what about the human side? Tom, I'm going to introduce it. So I'm not sure it was clear, but the system actually writes and scans the technology for you and it writes those reports. That's what you're going to be asked for. That's what the dual mandate is. So you get a lot of help from what we've built, but it's not about technology alone. It's about people. So I want you on three to yell out what COE means. But it shouldn't. So I created a center of excellence for cloud at NASA, and we've seen many companies do that. What do you do? You create a center of people who think they're excellent, and everybody else hates them. And then they create their own center of excellence. You now don't have just shadow IT, you have shadow COEs. So isn't there a better way? There is with all of these new technologies that are coming.
What you really want is to work backwards on the business problem, and you want a center of engagement. You want people to try it, experiment with it, and be able to very scrappily test it. And if it pays out business dividends, you double down. So you think big, start small with these experiments, then scale fast. But they're business experiments, not technology experiments. And trust me, the companies we talked to, this works, and the people who do it get promoted. So this is the way to actually test and try this out. Now the other thing that it does is it engages teams and people beyond your organization. You're all sitting here and hopefully you're talking to each other because that's the future collaborators that you're going to have. So the Tech Recon center of engagement can now reach out to business units as many as you can, and then to external partners and the entire ecosystem intelligence, and you will start feeding each other information that you're looking for and perhaps even create new partnerships with these agents. And it all starts with the Tech Recon that you've built. But how do you fund it? How do you prove that this is worthwhile?
Arvin, how do you fund it? Yeah, so I would say that the human side of this is probably the most important investment needed, but you do need some dollars as well. And in my experience as we got this thing going, one of the most important things was to create the fuel for innovation. We all hope we'll go to our CFO and CEO and say, hey, we need to stay in touch with technology, emerging technologies, and please fund it. It's very hard to get funding for that. The way I've always done that is to start the innovation process, the new technology application process in areas that generate some savings first. And then I have the agreement with the CEO and the CFO that instead of taking that saving to the bottom line, we'll reinvest this into more innovation. And that's the reason I call it fuel for innovation.
And once you have the technical capability, the human side of it, and this little bit of funding, then the way you want to think about this is to get your flywheel of emerging technology and innovation going. The way Amazon likes to look at flywheels is you think of this as a flywheel. You need to give it a start by getting some business leadership engagement. Attracting talent into it, you get the right talent, the right ideas for innovation come up, and you make them successful. You invest in making the early ones successful, and that gets your flywheel going. And as you gather momentum, more and more ideas are generated and more business impact happens. And as someone who's trying to get this Tech Recon capability going, unless you want to do this all by yourself, you've got to think about how the flywheel is running and how you can accelerate it further. So that was one of the areas that we focused on.
And a common learning that both Tom and I had was that oftentimes the question that gets asked is ROI. And what I have done, sometimes successfully and sometimes not very successfully, is kind of shift that conversation to ROA, which is return on attention. What truly matters when you're looking at emerging technologies and early stages is whether there are ideas that actually business cares about.
The way we measure that is through how much engagement we are getting. Are people showing up? Are they volunteering to be part of this ecosystem? Is the velocity of the ideas moving from early stage to actual experimentation growing? And finally, most importantly, is this leading to strategic conversations in the executive room saying, "Look, here are the 10 technologies which are coming up. These have potential. We want to do experiments with these 3, watch these 5," and so on. That is the sign that this Center of Engagement is working for us or not.
Conquering the Dual Mandate: A 90-Day Roadmap and Live Demo
Now, there are many challenges to making it work. This is not straightforward. The common issue in the earlier stages is that you launch this thing and nobody shows up. No one has time. Everyone is very busy. This is a classic issue. Investing in the early few projects in a way that saves dollars as well as capacity is very useful. I found that always very helpful. But more importantly, it is about creating that culture. Leadership has to convey the message that learning is important, not just for this job and this company today, but for your careers in the future.
The second thing that often comes in the way is that once you get people excited, everyone wants to work on the same hot technology. If you truly want to look at a portfolio of emerging technologies, you want more people to be interested in different things. There are many ways to do that by making sure the portfolio is very clear and you move people from one idea to the other, and not let everyone get focused on the same thing. And finally, very importantly, your business has to feel like it is their problem to solve. If that is not happening, you have to work backwards from where the business opportunities are.
With that, hopefully we have given you a lot of ideas on how to do this. We have seen that it is possible in a 90-day period to get this whole engine and this flywheel going. In the first month or so, using what is now possible with TechRecon, you can actually set the foundation in place and have the conversation with your CEO and CFO that this is something as an organization you want to put more attention on. In the second month, you can start looking at some of these position papers, get people to own it, build more detail into it, and by month 3, you should be going into your executive boardrooms and having a conversation about what are the technologies you want to keep an eye on and potentially do experiments with.
Now, I make it sound very simple. This is not a silver bullet. Of course, there are challenges. But this is becoming so important for us in our roles as CIOs that this is not something that can be ignored. We have to figure out a way to manage our operational delivery responsibilities while also building the muscle to do technology reconnaissance. So hopefully with that, what you should be able to do is this: "What is our position on the new AI regulations?" Already briefed the board last week. Position papers sent to your tablet. "How will quantum computing affect our encryption?" We have been monitoring quantum developments. Recommend moving to quantum-resistant encryption in quarter 2. Full analysis ready for review.
All right, so with that, we want you to go out and conquer the dual mandate. You have a lot of resources here. You will have access to the source code for TechRecon, which is posted. You can access that. We will be hanging around for a few minutes. We will actually redo the demo as well since that did not work earlier. But very importantly, this week is an amazing set of folks who are around here. Everyone who is in this gathering hopefully are folks who want to build this capability, so talk to each other, connect with them. Go to the expo floor. Go to the replay. There is a lot of new technology on demo. Get to see that and learn how that works.
So with that, one more point. When I asked you how many had an agentic something running, like two people raised their hands. Here is your opportunity to have an agentic something running. That is very low risk and very low cost, but gets you started.
While the demo is being set up, you'll have 2 minutes to see what you would actually get. It's worth watching because people will get interested. As we started doing this, we realized it was not at all what we thought. In the beginning, you might think an agent is like an intern where you can just tell them to go do something. You can't do that with an agent, so the prompt that was shown is very telling. It came from a lot of iteration, so this gives you a chance to get hands on with it.
When you run the Glue Cryptal run button, the supervisor agent takes control of the entire workflow, and the planner agent begins planning for the researcher, coder, and research report agent to run. You can see the real-time logs as they're being streamed from each agent, specifically the planner agent right now. The planner identifies what needs to be done. Once the planner finishes generating the task list, the researcher, coder, and reporter agents begin executing their respective tasks and start filling in their information.
When all the tasks are completed, you can see on the right-hand side the control plane where all the agent statuses change to complete, and the generated files become available for download in the bottom right corner. After you create part one, you can proceed to part two, which generates specific technology position papers based on the part one assessment scores and the higher ranked assessment scores.
This is an open source blueprint that gives you a chance to have this dual mandate. It's hard, but once you've set it up, let's say you have a board meeting tomorrow and they ask you to come in and talk to the board of directors about what will happen with Agentic AI. You can just run this and have the latest report created for you. That's the key benefit—it saves you time.
We'll hang around here if anyone wants to actually implement this. Jiyun is here, and through the GitHub link, you can get access to the code and instructions. You can also reach out to us for any support or help needed to make it work.
; This article is entirely auto-generated using Amazon Bedrock.
Top comments (0)