#30 - AI agents are moving faster than you thought - Matt Yanchyshyn

AI agents in business aren't something that will happen in the future. They’re already here, and they're scaling a lot more rapidly than we expected.

In this episode of The Deep View: Conversations, Editor-in-Chief Jason Hiner talks to Matt Yanchyshyn, who leads AWS Marketplace at Amazon Web Services. Yanchyshyn's team helps organizations discover, buy, and deploy software on AWS, and one of the biggest shifts they’ve seen over the past six months is the explosion of AI agents in real-world use cases.

When AWS unveiled its agent marketplace in mid-2025, the internal goal was initially to launch with 50 agents. By early 2026, that number had surged past 2,600 agents, making it the fastest-growing category in the history of the world’s largest cloud platform.

So what’s driving that surge? Yanchyshyn breaks it down.

In this conversation, we cover:
+ Which types of AI agents are seeing the fastest enterprise adoption
+ The industries and use cases leading the charge
+ How companies are handling data security and sovereignty concerns
+ The role of multi-model orchestration in agent effectiveness
+ How AWS is using agents internally to drive lots of different wins

If you're trying to understand where AI agents are actually being deployed — not the hype, but the reality — then this conversation will reset your expectations. It will help you see where agentic AI is already delivering business value, and where it’s heading next.

Subscribe to The Deep View: Conversations in your favorite podcast player for more unique conversations with the brightest minds solving the biggest challenges in AI. You can also subscribe here on YouTube.

And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights that keep our audience ahead of the curve and help them put AI to work every day: subscribe.thedeepview.com

Thank you to our sponsor, Deel, an AI-native platform for HR, IT, and payroll. Hire, manage, pay, and equip anyone, anywhere. https://www.deel.com/deepview

Creators and Guests

Host

Jason Hiner

Editor-in-Chief of The Deep View

What is The Deep View: Conversations?

From frontier labs and enterprise platforms to emerging startups reshaping entire industries, The Deep View: Conversations podcast interviews the brightest minds and the most influential leaders in AI.

Jason Hiner (00:03.024)
All right, so Matt, when you and I talked last time, last month, just before AWS re-invent, we talked a lot about AI agents, about the work that has been going on in AWS. And there was this kind of incredible trajectory from when you launched last summer, summer of 2025.

to that point and how it far exceeded the expectations of you and your team. And I was hoping we could go back and talk a little bit about that because it was pretty exciting, especially during 2025, we saw this momentum around the narrative of AI agents, but there weren't a lot of places where people were talking about it the real world. And when you and I talked, was the first time I was like, okay, this is really happening. Like there's a lot going on and it's not just

experiments and things that people are downloading and trying. could you set the scene for us? Remind me when you launched your expectations were pretty minimal and I remember when we got to the end of the year it was pretty huge and then now it's I know it's already even just a month later you know far higher than that. So what were your expectations when it started and then where did you end up at the end of the year and then now?

Matt Yanchyshyn (01:25.39)
Yeah, I mean, you're right. Like, agentic AI has very quickly become the fastest growing category in AWS Marketplace ever. And, you know, to give you some numbers, I think when we spoke last time, we talked about how the team came to me last spring, in the spring of 2025, and said, you know, we're going to launch with 50 AI agentic solutions. And I was like, add a zero. And so I always tell my team to add a zero, but they actually beat me at my own game. And we launched a pretty high count with like 800 or 900 listings at launch.

as you said, in the summer of 2025. When we spoke, guess, in December, that had grown to about 2,100, I think. Is that right? And then I just checked this morning, and this morning it's at 2,660 listings. So I think from a breadth of what's available and enthusiasm, that's pretty incredible. Now, most of those listings are agentic solutions, i.e. they're SaaS solutions or other type of solutions that have agentic capabilities built within them. And we try really hard to vet the solutions so that they're

Jason Hiner (01:56.036)
Yeah.

Jason Hiner (02:00.932)
Yep, that's right.

Jason Hiner (02:07.684)
Wow. Wow.

Matt Yanchyshyn (02:24.898)
we call it agent washing, so they're not kind of fake. They're actually either agents themselves, like atomic agents, containers you can deploy and use on Agent Core or whatever, or they're truly agentic solutions that have either been created or changed. So yeah, about 2,600 agentic solutions, the bulk of which are AI agent enabled, typically SaaS solutions, but a ton of atomic agents too that you can actually deploy and use directly as agents themselves.

Jason Hiner (02:26.714)
Yeah.

Jason Hiner (02:49.328)
Wow, so you've also been doing some work to really make these enterprise ready, not just for a small experiment for one department or one part of the company or a startup or SMB. Those are all great and those things are happening, I'm sure. But.

When we talked, you also mentioned some things about some of the orchestration, about agent mode, about express private offers. Talk a little bit about some of the work that you all are doing to make these things really ready for enterprises to deploy. Because we have seen this move over the last couple of years, but especially in the past six to 12 months, that agents are one of the ways that enterprises want to move quickly beyond

and proof of concept to actually deploying these things and getting real value because the CFOs in these companies are pushing them. They're like, look, we're putting all this money into AI. We want to see some results. So talk about some of the things that you all are doing to make that a lot easier.

Matt Yanchyshyn (03:56.525)
Yeah, well, I first all, we're drinking our own champagne. You mentioned a couple of things, like we're using agents ourselves to help power the features and the services or products that we have in AWS Marketplace and elsewhere. So a big example of that that launched around this last time we talked around December is this agent mode. So agent powered search. that's example where we're taking a use case based approach where customers can use natural language searches and agents behind the scene.

Jason Hiner (04:00.708)
Yeah.

Matt Yanchyshyn (04:23.032)
go out and they look across the internet, they scour our own internal data sources and bring together product comparisons that you can then as a customer adjust in real time to help you sort of narrow down what you want to buy, whether that's an agent or otherwise. And those are agents, know, powering that. So it's a good example of a use case. A lot of people are looking for agentic research, agentic discovery, sort of next generation search, whether that's for internal or whatever. And we see that reflected in the searches actually we're seeing.

on marketplace, whether it's through agent mode. So like, to give you some numbers, like people used to search for like generative AI or agent. And we saw like, if you take generative AI, that specific search, that search term actually plummeted by 77%. Just in from, I think it was like the number two spot in 2024 down to like number 16 in 2025. And that's because people are looking for, to your point, like real things now. So

Jason Hiner (04:51.522)
Okay?

Jason Hiner (05:05.092)
Wow. Okay.

Matt Yanchyshyn (05:16.586)
how do I use agents for contract analysis? Like we have this Brazilian company called Matter Day and they use a bunch of different agents simultaneously to do scheduling and error detection and contract analysis. And that, so that customer, they would show up and they would look for a specific agent for a specific use case. so that's one big change is like what people are using or what they're looking for and what we are using to help them discover that. And then, yeah, you said orchestration. That, that's another big, and I would say more recent change.

Jason Hiner (05:38.682)
Sure.

Yeah.

Matt Yanchyshyn (05:47.247)
People forget that the whole term, like I think Anthropic coined the term MCP in about 2024. Google came out with A2A in mid 2025. Like this standardized way for agents to talk to each other in the first place is very new, very new. And people are just wrapping their heads around talking to agents with these protocols, let alone orchestrating multiple agents and governing multiple agents. And so that's kind of what we're starting to see more of. Like that Matter Day customer I mentioned, they have sort

Jason Hiner (05:52.271)
Yeah.

Jason Hiner (06:03.428)
Yes.

Matt Yanchyshyn (06:15.79)
12 autonomous agents. That's a good use case where it's not just one agent to do one thing, but how do I orchestrate multiple agents to do more complex tasks? And we're seeing that now reflected in the searches and the use cases, and that's pretty big change.

Jason Hiner (06:29.636)
Very good. are you seeing, do you have any data on the size of the companies that are accessing this? it a mix of SMB and enterprise? Is it mostly larger companies? Yeah, who are you seeing? What are the customers that you're seeing that are really getting into it?

Matt Yanchyshyn (06:46.958)
Yeah, I mean, it's the full spectrum. I was actually in Charlotte, North Carolina, just yesterday with a very large consulting company. And they have a new agent core based agent development and orchestration platform that they're rolling out. So we work with large companies all the time. On the partner side, I work with a lot of large ISVs and consulting companies who are turn also customers. They're broadly speaking, without exception, I can't think of any of them that aren't rolling out some type of

real, agentic, not just strategy, but like something in production. It varies in sort of sophistication and size. And then on the small end, of course, startups are leading the way as they always do. SMB, yeah, seeing some traction. Like SMB has actually always been the largest customer base of Marketplace. You maybe not making the largest purchases by revenue, but, and those customers are also exploring and buying agents. So it's actually kind of interesting. AI agents in particular is one that

Jason Hiner (07:34.106)
Sure.

Matt Yanchyshyn (07:45.743)
isn't pegged to a particular customer segment or size of customer. seems to be, from the data I've seen anyway, pretty much across the spectrum. Obviously, the use cases differ. A tiny company is not going to be looking at contract analysis at scale for healthcare. But yeah, you know what I mean. It's across the board.

Jason Hiner (07:59.473)
Sure.

Yeah, yeah. How about industry-wise? Are you seeing some industries lead the way? Are you seeing special momentum in certain industries? What are some of the industries that you're seeing doing some of the kind fastest uptake and maybe leading the way?

Matt Yanchyshyn (08:18.605)
Yeah, again, from the data I've seen, there doesn't seem to be any outlier. What is interesting is some of the most regulated industries like financial services, healthcare, we are seeing strong interest in adoption, like the healthcare example from Brazil I just gave you earlier. And that's interesting because that requires often, with that type of data, more careful governance and regulation and security. so I think even those industries, financial services that tend to be...

more careful with their technology choices are jumping headlong into into agentics. So yeah, there's nothing that comes to mind that jumps out. Obviously, in terms of drinking our own champagne, like Amazon itself is investing heavily in, you know, from a retail e commerce perspective and logistics perspective. And we're using our own services, whether it's, bedrock or agent core and agents built on top of those to orchestrate our own services. And my AWS marketplace, I mentioned agent mode. So that

Jason Hiner (09:01.584)
Okay.

Matt Yanchyshyn (09:14.669)
Definitely in software, obviously. Retail e-commerce, logistics, manufacturing, the use cases are super obvious. But yeah, even regulated industries. And interestingly, even government. We have a very healthy public sector business as well.

Jason Hiner (09:28.292)
glad you went to government because I was going to go to that one. was going to say, know, are you seeing you have always Amazon has had, you know, long public sector relationship, which is great. And, you know, there are places in which governments, the US government and many others are trying to get more efficient, right? They sometimes are very concerned with waste, with shrinking length of deployment. We've heard all the horror stories of, you know, projects that took

two, three years to deploy and by the time they were deployed, they weren't sort of the solution wasn't needed any longer. Are you seeing governments take a really active role in using AI agents to solve some of those problems?

Matt Yanchyshyn (10:14.031)
I don't have sort of actual stats like I do for some of the others on my fingertips for some of the government. What I can tell you though is, you know, we just launched, for example, like the European Sovereign Cloud in Europe and Marketplace was like a day one service. It launched at day one with European Sovereign Cloud. And I think that is indicative of our investment in supporting. That's not just for government, obviously. There's going to be a lot of regulated, say, financial services companies in Germany, for example, who are going to leverage that.

Jason Hiner (10:19.226)
Sir.

Jason Hiner (10:28.142)
Okay.

Matt Yanchyshyn (10:41.805)
But certainly government will too. It's targeting those use cases. We've had a long standing investment in the what we call ICMP or intelligence community marketplace that's popular. And I definitely believe, like you said, that as a taxpayer and just as an AWS employee that I want to help government be more efficient and help government agencies, government employees find agents and other solutions more quickly. So we're definitely invested in that and enable them to use it.

Jason Hiner (10:41.86)
Yes.

Jason Hiner (11:10.564)
Yes, yeah.

Matt Yanchyshyn (11:10.807)
If they have sovereignty requirements in Europe, for example, they should be able to procure and use solutions. So that's why we're investing in those areas for efficiency and for access. Whether it's in an air-gapped region for top-secret information or for sovereign reasons like in Europe, every government should be able to take advantage of the same amazing technologies and things like agents that we're seeing that's available to commercial. And that's definitely part of my strategy.

Jason Hiner (11:38.137)
I'm so glad that you brought up

sovereignty because that's one of the areas where we're seeing the world move more and more to data sovereignty, cloud sovereignty, where they want assurance that all of the data that happens in whatever systems in these highly regulated industries, especially like healthcare and finance and government that you mentioned, but AWS is very much a global organization.

to agents in a world that becoming more and more focused on data sovereignty when it not just agents but you know other AI solutions are

you able to offer that? mean, does Amazon, it's funny because we're going to talk about data centers now because you actually have to have physical locations in many of these places in order to do it, right? Like does Amazon have the wherewithal to do these data sovereignty platforms in essentially all of these platforms, Asia, North America, mean, even North America, Canada and US are having separate requirements and all of

that? What does that look like? How is that complexity? How are you able to manage that complexity?

Matt Yanchyshyn (12:56.579)
Yeah, from the AWS Marketplace perspective, that's actually always been a strength and it's something we're continuing to invest in. So like today, AWS Marketplace is available in 37 AWS regions and customers can pay in 14 different currencies. have multiple, like the European sovereign cloud, or I mentioned the ICMP, which is the fully air gapped region. We have multiple types of regions that meet different regulatory or customer needs across the globe. So I think,

Jason Hiner (13:08.291)
Okay.

Jason Hiner (13:17.221)
Yeah.

Matt Yanchyshyn (13:27.243)
Sovereignty or local needs comes in a lot of flavors. Like for example, if you're in India where we just launched the AWS marketplace in India late last year, there's certain types of companies because of whatever position they're in, whether they're selling to government or local regulations, they must use the Indian rupee for transactions. So we allow them to use local currency with a local bank and a local entity. So we actually invest in local entities in places like India, in Europe, in Australia, Japan, South Korea we just launched last year.

Jason Hiner (13:45.314)
Okay.

Matt Yanchyshyn (13:56.431)
Obviously, in America, we have a bunch of different actual local offices that we can do like local invoicing, local currency, local tax treatment. That's very often important for, you when you talk about regional differences. And so if you're a business, let's take an example, like you're in Japan, and if you're a Japanese business, you must buy through like a specific reseller and you must have, there's something called Japanese consumption tax that must be handled in a certain way, withheld in a certain way. The invoices have to look in certain way.

Jason Hiner (14:06.206)
wow, yeah. For sure.

Matt Yanchyshyn (14:26.095)
invested in those features in Marketplace so that whether you're a company buying for the Japanese government or you're just commercial company that needs to operate in a certain way from a tax compliance perspective, which in a way is a form of sovereignty, you can do that. And so it's actually been a huge part, this whole what we call international expansion for Marketplace has been a huge part of our investment that extends to things like the European sovereign cloud and other types of separate partitions in separate regions. So the customer demand is there. It definitely takes...

different flavors, I would say, per country and per region. And we respond to that by sort of launching in different ways in different regions. How South Korea and Australia or European sovereign cloud and Japan respectively work couldn't be more different. And so we work hard to manage those differences and sort of meet the customer needs. So yeah, I think it's a growing need. And it also just comes with meeting customers where they are. It just makes their lives easier. Like in some cases,

Jason Hiner (15:01.7)
Okay.

Jason Hiner (15:10.224)
Sure.

Matt Yanchyshyn (15:24.579)
you can pay in US dollars and make it work, it's just harder to do. You kind of need to do the tax handling and withholding and all that stuff. And we want to make it easier for customers by doing that on their behalf. And same with European sovereign cloud. You you could meet a lot of the regulations through a combination of, you know, complex data flow and encryption, et cetera, or you could just deploy the European sovereign cloud and have us do it for them. So a lot of it comes down to making it easier for customers and more convenient and oftentimes, you know, less expensive from an implementation perspective. So we're invested in like,

Jason Hiner (15:27.92)
Okay.

Matt Yanchyshyn (15:54.425)
I know I threw a lot of stuff out there, but like a ton of axes around what I would just generally call like international expansion, localized access.

Jason Hiner (16:01.048)
Yeah, so you made that sound all pretty easy and simple, but that is incredibly complex. So what you're doing, just to recap, and obviously this is AWS sort of broadly, as I understand it, and correct me if I'm wrong, but is that you're making it so that a company, an international company, could come and bring their product to market.

in the US, in this case, an AI product or an AI agent. And likewise, like a US company could quite rapidly decide they want to bring this to Japan, to Asia broadly, APAC, to Europe, to maybe the Middle East. And you're handling a lot of the transaction, you're handling the currency exchange between those things, the local laws and taxes, as well as the data sovereignty requirements that you can

you have to make sure you're doing or else you could end up in hot water and end up paying fines and run into a number of obstacles and sort of things that could really be penalizing to your business if you do them wrong.

Am I characterizing that correctly? Because that's a lot of complexity that you're sort of building into the system, going both ways. So North America to international and international wanting to sell into North America.

Matt Yanchyshyn (17:26.359)
Yeah, I mean, it's really about reducing the complexity, like to give you some examples. And it works both ways. I think where we've seen a lot of success is actually enabling companies outside of the US to sell into the lucrative US market, like a good example. And I really like the company and like the CEO. So I'm always happy to talk about them. It's this company called Data Mask from New Zealand. And there's this relatively small security software company that has landed some massive logos in the US and done that exclusively through the marketplace. So they can sell to...

Jason Hiner (17:31.194)
Yeah.

Jason Hiner (17:36.74)
Okay.

Matt Yanchyshyn (17:54.927)
to huge companies in the US through the marketplace motion where they don't even have really salespeople on the ground in the US. And not only that, but to your point, they can sell in the local currency through a local entity and from like a US currency billing, et cetera, it all kind of works. that would be very hard for them to do. They wouldn't invest in that themselves. Similarly, the other way could be on the small end or high end, but just to pick one like high end like Salesforce. They do a lot of business through the marketplace in Europe, in Euro.

And so we're able to do end to end Euro transactions and Salesforce deals tend to be big. You know, they're selling to big companies. These are big deals and currency fluctuation can therefore be a big deal. So if you can issue the offer in Euro and accepted in Euro, it takes the FX risk right out of it. Right. So it not only does it reduce complexity, but it creates consistency and from an FX, like you know what to expect in terms of how much that deal is going to be worth. So we have a lot of companies now using our end to end local currency support to reduce FX like

Jason Hiner (18:29.253)
Yeah.

Matt Yanchyshyn (18:54.103)
Wow. Local currency conversion variability to reduce that and tax handling. To your point, yeah, that can get complex. Where do I withhold the AT or withholding tax? It varies per country. And as much as we can do that on behalf of the customer, it makes it more consistent and easier, cheaper. yeah, definitely something that a lot of companies, both big and smaller, are leveraging in both directions. We're seeing that in India too with the region. We want to enable both large global companies.

Jason Hiner (18:57.316)
Yeah? Yeah.

Matt Yanchyshyn (19:21.411)
lot of them based in the US to sell in the Indian market, but also India has one of most exciting software markets in the world and to empower those companies to sell to a global market as well. And that is to meet really the role of Marketplace to create that access and to reduce the complexity of making it

Jason Hiner (19:27.706)
Sure.

Jason Hiner (19:39.481)
Okay, let's double click now a little bit more on some of the sort of trends you're seeing in AI in AWS Marketplace, especially some of the AI agent stuff. But really there's a number of things. Again, when we talked before AWS, we sort of went into this a little bit when you and I caught up and I wrote a little bit about what was happening because it had happened so quickly, so rapidly since last summer, so summer of 2025 when you all launched the AI agent marketplace. And it was a little...

It was a little shocking to me that like that was far along because in most of the places I look and that was one of the reasons I wanted to talk to you most of the places I look AI agents are still very proof of concept. They're very early a lot of a lot of things feel very nascent But on but what you all are seeing is different than that like you're seeing this maturing, know, pretty rapidly So that's why I was I just I would love to to hear a little bit about that. There's all kinds of things

to kind of unpack around costs, around what companies are seeing. So just I want to get into a few of those real quickly.

you know, this pricing and discovery is key. If companies are, like, hey, we, want to see what we could do with the agents for whatever reason. They're like, we want to invest in this, whether they've tried some proof of concepts themselves, or they're seeing sort of case studies of others doing it. And as they're, you know, as they're getting into it, you know, how is that sort of discovery and management sort of changing in the AI era? You mentioned what you saw, which is that you're using AI and agents to help

companies do some of that discovery but you know what what kind of data have you seen about like what people are buying how are they describing it you know what are what are sellers doing why don't you unpack a little bit of that because certainly you've got some intelligence on on what what's getting transacted in this because there's a lot of transactions happening around the agents in AWS.

Matt Yanchyshyn (21:46.724)
Yeah, so let's unpack that. There's a lot there. So first of all, at a high level, I'll try my best to root this in numbers where I have them at my fingertips. So even just the term AI agent, it went from number 64 in late 2024 to number three of overall AWS Marketplace searches in just about 18 months. So people are...

Jason Hiner (21:54.34)
Okay.

Jason Hiner (22:06.51)
Wow. Not AI searches overall, all searches in AWS marketplace. Okay. Okay.

Matt Yanchyshyn (22:09.921)
Yeah, AI agents. So people are looking for agents, right? And of that, the number one type of agent that people are looking for these days is customer service agents. Probably not surprising, but you know, that's a data point right there. That's what we're seeing on the marketplace. And from, you know, like I said, 2,600 plus agents, millions of customers, so decent base to work off of. If you actually go to the marketplace public website, again, I just checked to try and give you some real time, we live in a real time world, right? So the...

Jason Hiner (22:19.716)
No guy. Yep. Sure.

Jason Hiner (22:36.858)
Yeah.

Matt Yanchyshyn (22:38.691)
The number one AI agents and tools, or number, I guess, top four today are Brave Search API Pro, Tavoli, which is for AI assistance, Asana, which is interesting, so business applications, and Stripe, MCP server, which is really interesting, because now we're getting into payments. So e-commerce, obviously things that facilitate discovery and search, like Brave and Tavoli, but also Asana, like business productivity.

Jason Hiner (22:50.8)
Hmm.

Okay.

Matt Yanchyshyn (23:08.367)
When I look at the a little bit broader and take a step back, know, companies like crew.ai doing a great job with agent orchestration and a lot of momentum there. Agent force from Salesforce, which a lot of people are buying through the marketplace, good traction in terms of, again, of agent platforms that are enabling agent, not just prompts, but using prompts to sort of orchestrate agent workflows. That's definitely an area that we're seeing grow and I'm seeing in the market more broadly grow pretty quickly.

Agent security and governance is rising. And if you look also at some of our stats, have successful, actually, speaking of international, like a successful startup on Marketplace, OnePassword, a lot of people know them for their consumer-based password manager, but they have a growing and pretty successful enterprise base as well, and they have agents. They're a startup, they're on the, and so that's another good example of of agentic security at that particular pace in the sort credential and security space.

And if you look at our security offerings, which have historically been the best selling SaaS solutions on the marketplace. So people like CrowdStrike, they have an MCP server. Databricks, they have an MCP server. that, think MCP servers as helpful add-ons to SaaS solutions is definitely so that you can interact with agents via MCP primarily with your existing solutions in data and analytics, Snowflake, Databricks, security like CrowdStrike, for example.

Jason Hiner (24:15.631)
Yeah.

Matt Yanchyshyn (24:35.28)
All of those companies are launching or have already launched MCP servers and are listing them on marketplace for customer discovery. And then some of the use cases I talked about. Just one other data point that I think is interesting, because it came up just yesterday in a meeting. Even though MCP came out, I said, I think late 2024, A2A was mid 2025. We are still seeing a dominance of MCP. A2A is useful. Like I mentioned, drinking our own champagne just yesterday, someone on my team.

Jason Hiner (24:53.797)
Yep.

Jason Hiner (24:58.798)
Okay.

Matt Yanchyshyn (25:03.376)
used the A2A interface on the AWS Marketplace agent mode to build a voice interface. So you could use your own voice to search through the AWS Marketplace. So A2A is making it easy for people to build, but we're still predominantly seeing MCP servers and the overwhelming majority of MCP servers are add-ons to existing primarily SaaS solutions, largely concentrated in the big infrastructure categories like security, data analytics. But you know, is maturing. Again, like we talked about healthcare earlier, we are seeing more and more sort of...

Jason Hiner (25:08.996)
Hmm. Okay.

Matt Yanchyshyn (25:32.108)
narrower industry use cases and sort of more multi-agent use cases, but that is still new and, you know, new is moving so fast right now. I expect if we talk again in six months, it'll have changed rapidly again, but those are off the top of my head some observations I've seen in terms of trends that we're seeing reflected in the data.

Jason Hiner (25:50.713)
Okay? How about...

pricing, you know, Amazon AWS, Amazon and AWS have always been really well known for the fact that you pay as you go. You pay for what you use. You don't pay for a lot of infrastructure that you're not using. AI agents pricing is a really interesting thing, of a, you know, agent as a service sort of model. I have to imagine that the pricing has got to be challenging

And I'm guessing that it's pretty experimental at this point, that people are pricing in different ways. And then you also have the fact that really freemium offerings had come as the standard for a while, and in many ways in SaaS still are. How do you do that in agents? How are people and companies pricing the agents? What are you seeing there? And is it as dynamic as I'm thinking it must be?

Matt Yanchyshyn (26:49.956)
Yeah, well, I'm seeing two things. You mentioned premium. There's definitely a lot of that. When you look at the AI agents and tools category in AWS Marketplace, most of those, like I mentioned earlier, are either SaaS solutions with actual sort agentic capabilities included, or like I mentioned, like MCP servers that attach onto an existing CrowdStrike or Databricks implementation. Those MCP servers that attach to an existing implementation tend to be free. So you pay and so...

Jason Hiner (26:53.913)
Okay.

Matt Yanchyshyn (27:19.384)
They provide agentic capabilities as a free add-on to a paid, either pay as you go or contract-based solution. So that would be, if we're being realistic, the common pattern. However, the AI agents and tools category has surprisingly been great from a financial perspective too. So not all those agents are free and not all those agentic solutions are free and it's quite a healthy business. We don't break out our business by individual categories, but I was just reviewing on a monthly financial review this week.

Jason Hiner (27:29.475)
Okay.

Matt Yanchyshyn (27:48.707)
and it is healthy. are seeing large private offers, which is interesting for, we've even had large by any measure private offers for individual agents, not just agentic SaaS solutions. So people are buying contract-based agents, which is honestly not something I really expected. So we're seeing that. We are seeing people to your point.

Jason Hiner (28:06.228)
Interesting. Because that's pretty mature, right? Like, if you're doing it at that level, those are pretty mature contracts. Yeah, and you're making a commitment. Yeah.

Matt Yanchyshyn (28:11.184)
what you're committing. Like, yeah, by definition, you're buying a one, two, three year contract to like an actual agent, right? And so, you know, while it's not like SaaS still kind of dominates it, but it's a growing thing. People are paying and engaging in private pricing, custom terms, what we call private offers. Pay as you go, yes, that is nascent, but yeah, you're spot on. I think people are still figuring out how to meter these things. There's a lot of talk in, you know, forums that I'm in, where we're all sort of, you know, between different industries and different companies shooting the breeze about

How do you charge for value when it comes for agents? does that look like? And I think there was some early experiments on charging per query or question answered. And for the most part, from my observations, that was a mixed bag. Customers, because how do you predict how much your employee base or customer base is, how many questions they're going to ask? It's a really hard thing to do from a business predictability perspective. So people are starting to try and pivot to other types of metering, other types of measure to value.

Jason Hiner (28:44.154)
Sure. Yeah.

Jason Hiner (28:54.959)
Okay.

Matt Yanchyshyn (29:08.622)
And that is still, I think I'm seeing multiple experiments. I haven't seen a single pattern emerge. The other thing that I see kind of on the horizon, this definitely coming up in industry forums or hallway talk is, are things like stable coins gonna play a role? And we're hearing a lot of murmurs there, but that comes with lot of complexity and it's still very, very early days. But I wouldn't be surprised if that's something that comes later. So how you pay, what you pay for.

Jason Hiner (29:23.919)
Mmm.

Matt Yanchyshyn (29:36.856)
up in the air, but we are seeing people paying in the form of work being you go and in the meantime, the business isn't waiting. So it's exciting, honestly.

Jason Hiner (29:39.683)
Okay.

Jason Hiner (29:45.049)
Yeah, so there's still a lot to figure out in terms of the pricing, not surprisingly. AI agents are a little bit of a paradigm breaker in some ways. It's just not the way we're used to buying software or that kind of thing. I'm glad you brought up, though, private offers. Let's talk about that for a second, because that gets to larger companies, ISVs.

people that are selling these agents, because it's one thing to just have a pay as you go, Amazon's, sorry, AWS, Amazon Web Services is very well known for being a place where you can go, you can figure out what you want to buy, and you can buy it right there. And it can really shrink the sales process. We know that in the enterprise, sales processes, know, tend to drag on, they can, you know, six months more, right? Because it's a lot of, there's a lot of negotiation, there's a lot of sort of discovery of,

what you have and all that. But in AWS's strength is the opposite, like letting you come see what you want, buy it right there, be ready to start deploying it. But with what I understand about this private offers thing is you're letting companies, the companies that are selling the software, allow there to be some...

sort of negotiation and even some of the work that's done in contracts. Not all of it, they're not gonna do a one size fits all sort of thing when it comes to buying it. They're gonna put some potential offers out there. And as I understand it, that's a little bit, sounds like itself is a little bit AI agent driven, the private offers. I overthinking that or what does that look like? Talk a little bit about what you're doing. Okay, okay.

Matt Yanchyshyn (31:23.077)
No, I don't think you're overthinking at all. I think you're remembering maybe a feature we launched at re-event called Express Private Offers. And we talked about drinking our own champagne and what that, okay, so just stepping back, Marketplace started like 13 plus years ago as a self-service where you buy like server-based products to deploy an EC2, kind of like more like an app store. And that was a very healthy business. we pivoted, you know, we...

Jason Hiner (31:31.022)
Yeah.

Jason Hiner (31:40.836)
Yes.

Matt Yanchyshyn (31:47.974)
We're not afraid to change, just like with this agents thing. And we pivoted to Sass and kind of help grow that business or that market. And then we added private offers, reseller led private offers, all these sort of more traditional enterprise buying motions. And that has actually become the most successful part of our business. So we still have a super healthy self-service product like Growth Business where people come in, maybe do a free trial, do a pay as you go. There's like no human involved. That is one of the strengths of Marketplace. And even for our largest sellers that, you know, in sales speak.

Jason Hiner (31:51.386)
Okay?

Jason Hiner (32:11.663)
Yeah.

Matt Yanchyshyn (32:17.049)
is the top of the funnel where they get leads and opportunities that either stay self-service for low cost of sales or they convert to bigger contracts like custom pricing. But there's this thick middle. There's the people who want a discount, effectively a private price and maybe more often not custom legal terms. And then there's people who are cool with the standard terms and pay as you go. But then there's this massive piece in the middle, like often SMB, maybe smaller customers, smaller deals, depending on the type of...

Jason Hiner (32:19.866)
Okay?

Matt Yanchyshyn (32:46.587)
customer and region you're in could be like $10,000, $100,000 typically in a deal. it's not like the ROI and putting like an enterprise salesperson on that is not the best. They're not gonna get much commission and you know it's ultimately you're not getting much margin on it. And for the customer too, just typically do those deals, they want them quickly. So we launched this thing, it's actually agent powered called Express Private Offers where we qualify the customer using agents. Like is this person who they say they are, do they work for who they say they are?

Jason Hiner (32:50.97)
Sure.

Jason Hiner (33:08.464)
Okay.

Matt Yanchyshyn (33:14.533)
does this customer meet the criteria that the seller has set for getting a discount? And they have their rate card and they control the rules and say, okay, if this is like an enterprise customer from the Northeast United States who's done X volume on the AWS marketplace, or is a bona fide enterprise buyer who's in the healthcare space, I'm getting really granular in particular, give them a 15 % discount automatically, don't talk to me, I'm good. And so we can issue custom pricing on demand, custom pricing as a service where we qualify the customer.

Jason Hiner (33:33.996)
No, that's great.

Matt Yanchyshyn (33:43.73)
And that's huge because it lowers cost of sale for the seller and the buyer gets it immediately. to me, this is exciting because it's bringing together the OG world of kind of the beginning of marketplace self-service with the enterprise buying contract, but reducing the cost of sale and time to value for the customer. And that's kind of the future is, hey, I'm a big buyer. I deserve a discount. Sure. Fine. We'll give you one. But that doesn't mean you need to like have a lengthy negotiation with the salesperson. You know what? That's not good use of our time. We got agents to build. So, yeah.

Jason Hiner (33:44.846)
That's, yeah.

Jason Hiner (34:12.462)
That's pretty great that you're using AI agents to sort of streamline the AI agent marketplace and allow especially SMBs to be able to buy things quickly, but also vendors to sell to them quickly and remove some of the friction and that long sales cycle that can happen in that. that's great. That's amazing.

Matt Yanchyshyn (34:35.314)
Yeah, because you know, there's certain like sales lunches and you both show up to that sales lunch and you're like, okay, this is a renewal. It's a three year contract. You know, I need to do a mandatory 5 % uplift to make you know, my my sales leader happy. You know, the drill at sign and you you sign immediately like that's that's most sales. And so, you know, that's kind thing we can automate and it saves time for the salespeople so they can focus on the big gnarly new contracts or the ones with conditions or, you know, expanding with additional SKUs or whatever, you know, because nobody wants to invest their

Jason Hiner (34:41.667)
Yeah.

Jason Hiner (34:51.45)
Sure.

Matt Yanchyshyn (35:04.452)
smart and expensive sales force on those kind of middle deals that should just kind of happen and most importantly, customers just want it quickly. You know, there's this whole generation I always say of people who are now like CIOs and CEOs who grew up on their phones on the app store. You know, they're getting like, you know, they used to buy whatever like bubble burst instantly and now they want to buy Salesforce instantly. So, you know, it's wild. Yeah.

Jason Hiner (35:17.209)
Right?

Jason Hiner (35:21.616)
Yeah.

No, that's really interesting too. Yeah, the workforce that's sort of more digital native doing these things. But, you know, it's funny, we talked about AWS selling agents, but that's maybe one of the best, most practical uses that I've heard of about agents is what you're doing to actually the process of selling them. So that's pretty cool. How about in terms of the ways people are buying AI, beyond just agents, because people are buying AI solutions, you know, on AWS Marketplace.

I'm sure in large numbers, in large volume. What are you seeing in terms of this move toward ROI? We saw a lot the last couple years around experimentation, proof of concept, but we really started to see in the last six months of the year of 2025 of these companies starting to really get antsy about ROI.

and about wanting to, and scaling solutions as well. And so I'm sure that you all must be seeing some of that, some of this AI deployment, these trends start to shift in some different directions. What can you tell me there about what you're seeing in terms of the way AI is getting deployed by companies?

Matt Yanchyshyn (36:43.29)
Yeah, well, I guess two things come to mind are like the beginning and the end of the buying cycle. One thing that we see a lot of is buyer's regret. was Gartner who came up with a study, I think last year or two, that said that like 60 % of what people end up buying in an enterprise software perspective sits on the shelf. Like it gets used once or never, which seems like a huge number until you think about it.

Jason Hiner (36:52.675)
Okay.

Jason Hiner (37:05.752)
Wow, 60%. Wow.

Matt Yanchyshyn (37:11.312)
difficult for me specifically to measure, but I think they're right, you know? And so that's for a couple reasons. It's maybe you made the wrong buying decision or you re-upped on a renewal without wanting to or because you had to, or it didn't get set up properly. And so that's like two ends of the buying cycle, like the research upfront or after you bought it. And so, and this is not specific to AI, but I think AI is not immune. It's the same thing. And in fact, even more so because people are still learning.

Jason Hiner (37:14.158)
Mmm. Okay.

Jason Hiner (37:27.31)
Okay.

Matt Yanchyshyn (37:36.403)
and you buy this AI solution, it's like, okay, well then your governance security team comes in and say, okay, well, to use this, you need to do X, Y, Z, and you don't have a person to do X, Y, Z. So to address that, goes back to what we talked to the beginning around buyer regret with agent mode. We're giving customers a lot more tools to find the right solution for their needs, whether it's an AI agent or otherwise. So we launched AI powered product comparisons, and now we have a more sophisticated version of that with agent mode. And that's important because we want to help you

compare and make the right decisions so that you don't have that buyer's regret. You have all the information and with agent mode you can say, show me this specific granular detail that may not even be in the product listing page and it'll give you that detail about the product and compare it to comparative products so you can make the right agent decision or whatever software decision upfront. And then once you buy that thing, that's why we're investing in things like we launched IAM temporary delegation at reInvent, which is a fancy way of saying it allows the software you bought to take actions.

on your behalf with your permission and your AWS account to help with the setup steps. And we launched a more involved version of that with CrowdStrike this year and Databricks last year. So after you buy it, it automates the setup steps. So it takes the clicks from like 60 to six. And that's not just a buzzword. it's, if anyone has ever set up a data lake solution or a SIM or some type of a sophisticated security or a multi-agent workflow, it's hard, you know, if you do it yourself. And so we want to automate those steps so that it doesn't sit on the shelf. So.

Jason Hiner (38:44.526)
Hmm.

Wow.

Jason Hiner (38:57.998)
Yeah. Yeah.

Matt Yanchyshyn (39:03.014)
We're investing in the upfront stuff and the post purchase procurement or post purchase setup experience pretty heavily because we want, know, sellers they like it because it increases customer retention, renewal probability, but more importantly the buyers like it because it increases the immediate time to value and you know, it doesn't sit on the shelf and they don't have that regret, did they buy the right thing? you know, we're no longer relying on customers to do their own research, we want to help facilitate that.

Jason Hiner (39:16.282)
course.

Jason Hiner (39:31.929)
Okay.

Matt Yanchyshyn (39:32.006)
because it's in everyone's interest for them to buy the right thing.

Jason Hiner (39:35.736)
And then once they buy it, you're doing some work, some after sale work on helping them deploy it. It's almost like the customer success. that your sort of, have like a customer success team or something that's helping do that?

Matt Yanchyshyn (39:47.4)
Yeah, actually funny you say that. My peer, Bargs, we have these customer success managers, a whole global team, but now we have a focused customer success team under my boss, Ruba Voronok, run by someone named Bargs. So yeah, focused on customer success from a business perspective, but even from tech, I have a whole organization called Deployments that is dedicated to what happens after you buy. My team is actually divided into the buyer journey.

And so each we have organizations for each part of the buyer journey. That's how we're organized as a software engineering organization. And the deployments team focuses, like they do a ton of research, like how do we help the customers be successful post subscription? That could be a free trial or it could be a paid subscription with any type of motion where they get the product. And that, so they're the ones who launched the CrowdStrike feature that worked with our identity team on that delegation feature I mentioned that launched the Databricks feature. And it's worked like we see the...

Jason Hiner (40:14.51)
Okay.

Matt Yanchyshyn (40:42.367)
retention go up, like we have a year's worth of data now from Databricks to show that what we launched with them in terms of helping customers discover in the Databricks product and then set it up effectively means happier customers and they stick around and they make better use of the product.

Jason Hiner (40:57.36)
Very cool, very cool. So.

I want to ask a little bit about something you mentioned earlier as well, which is you talked about the top four AI agents in your marketplace. Could you speak to the rest of the top 10? What are the rest of the ones that are in there? Because it's really interesting to see what people are buying. And I know some of this information is out there for people that are already Amazon AWS customers. They can go and see this, right? They can view this. It's out there in the marketplace itself. Yeah.

Matt Yanchyshyn (41:28.275)
Yeah, mean, for the actual numbers, I would have come with them if I had known, but we only list the top four. You can see for yourself, it's public, it's not behind, we only list the top four on the website. And those are top four self-service, and then separately there's top four by private offer. And I'm sorry, I just don't have those, but I'll get back to you with the data. But the trends are consistent. They correspond a lot of times to the...

Jason Hiner (41:36.198)
you do? Okay, okay.

Jason Hiner (41:42.34)
Okay.

Matt Yanchyshyn (41:53.364)
top SaaS solutions we have. mentioned CrowdStrike with an MCP server, Splunk has an MCP server, Databricks like the data and analytics and security categories where we've seen such success over the years with Marketplace are also areas where people are investing in agentic capabilities. yeah, I think maybe the one I would flag that's linked to one of the top four.

Jason Hiner (41:58.705)
Yeah.

Matt Yanchyshyn (42:17.651)
I mentioned there was an Asana MCB server and I also talked about how Salesforce has seen great success for the marketplace. That whole category of business applications has been one of our, like AI agents is our fastest growing category all up. from business applications is our fastest growing, I don't know if that's a vertical, it's not really a vertical, but we'll say business category that has grown over last two years. Salesforce, ServiceNow, Workday, Adobe, I mentioned Asana, they're selling

Jason Hiner (42:20.079)
Yeah.

Jason Hiner (42:30.672)
Okay.

Jason Hiner (42:37.466)
Mm-hmm.

Matt Yanchyshyn (42:47.707)
of stuff through the marketplace and they're bringing a whole new set of buyers who in turn are looking for a whole new set of agents to do different things. Like the buyer looking for an Asana agent for sort of tasks and workflow management is very different from the buyer, you looking for the Stripe agent, but certainly different from the CrowdStrike buyer. That's literally a different human with a different title at the company. So we've actually had to adapt, you know, how we go to market and the features and how we appeal to those types of buyers looking for agents in new categories. So business applications is huge.

Jason Hiner (43:09.06)
Hmm.

Matt Yanchyshyn (43:17.585)
And again, another area where we are drinking our own champagne, mean, literally not a day goes by that I don't, or I should say not an hour goes by that I don't use our own Amazon quick suite product internally. And that's, you know, Anthropic has seen a lot of early success with their cowork product, for example, and this rise of agentic business applications that you see reflected in the top four on AWS marketplace with Asana is rapid and I think will be sustained. It was really focused on developers as you know.

Jason Hiner (43:34.704)
Sure.

Matt Yanchyshyn (43:47.653)
both AI and agents for a lot of the last 18 to 24 months, but there has been a strong and heavy pivot toward business users that I'm seeing reflected in the top four and in just my day-to-day life.

Jason Hiner (44:00.474)
What about, what's quicksweet? Amazon quicksweet, yeah.

Matt Yanchyshyn (44:04.179)
Yeah, Amazon QuickSuite is a publicly available service that we sell alongside Kiro. Kiro is like the ID for developers. And then you have Bedrock for inference, where you can use like Anthropic and Nova and do inference. We have SageMaker for training. And then we have QuickSuite, which is for business applications. So you can connect like your Office 365, your Box, your Google Drive, all of your different repositories. And then...

Jason Hiner (44:12.184)
Yeah.

Matt Yanchyshyn (44:30.823)
you can run customized chat bots and flows on top of it. So every morning I get to work, just this morning actually, and it automatically scanned all of my slacks, all of my emails and said, you should pay attention to these ones first because they're your highest priority ones based on you, based on your business. Because it knows about my business, what my business priorities are and me. And it does, that's an autonomous agent workflow that's running and goes through my emails, my slacks, my documents, and then surfaces what to focus on first. And that I kind of take for granted now. But a year ago that like,

didn't exist. So that's just, I have multiple flows running during the day that help me manage my business. And even little things, like in preparation for this interview, where we used to have to go to multiple dashboards and multiple documents, I can do a natural language query in a quick suite. It goes off and queries both external and internal data sources and we'll pull together structured examples with data to back it about how our AI agents and tool category is working on AWS Marketplace. And that's something that would have taken me a ton of time.

Jason Hiner (45:00.676)
Wow. Sure.

Jason Hiner (45:29.392)
Sure, sure. That's like an hour or two of work and it's doing it for you.

Matt Yanchyshyn (45:29.997)
and that I can do in a few minutes.

Yeah, and that shift also from, I still do some sort of ad hoc, what I would call queries, real time type of queries, but what makes I think Quicksuite really interesting is this whole notion of flows and scheduled actions and orchestrated where it does actions on your behalf in the background. And that is relatively new in sort of the agentic space in terms of mainstream use and something that I know my team and I benefit from, like even on my developer team now, if you're on call.

when you turn up to your on-call rotation to do ops and maintenance and watch the health of the service, it says, hey, welcome to on-call, here's an auto summary of everything that's going on. And it doesn't wait, it tells you that proactively. And that, in the whole agentic space, is a shift. This proactive, multi-agent orchestrated schedule of actions is relatively new.

Jason Hiner (46:21.55)
I love that you all are not only using AI agents in your AI agent marketplace to help customers find things, to help deals get done, but you all are using it internally. You're using agents internally to be more efficient, to be smarter at your job, to do some of the work that...

that increasingly would have been, you before would have been very, very manual and is now becoming automated. So this product, QuickSuite, is similar to cowork, to Anthropic Cowork.

Matt Yanchyshyn (46:55.826)
Yeah, it's this whole idea of empowering. Listen, developers can use it too, but it's primarily you don't need to necessarily write code or anything like that. It provides a layer that essentially allows you to build agents. You can make custom chat bots that can operate in custom agentic flows. They're called quick, sweet flows that can run on your behalf and you can design them using natural language. So you can literally write like a prompt that says every morning.

Jason Hiner (47:05.722)
Okay.

Matt Yanchyshyn (47:22.748)
run a flow that goes and checks all of my Outlook emails and all my slacks and based on everything you know about my business, prioritize what's important. That's literally what you would write. And it can go ahead and build a workflow that goes and runs on your behalf. And that's kind of wild. That's taking prompt engineering plus sort of generative AI plus multi-agent and making it accessible to an everyday user, to someone who doesn't necessarily know how to string together a bunch of MCP servers.

Jason Hiner (47:41.53)
Yeah.

Jason Hiner (47:51.44)
It's a matte agent.

Matt Yanchyshyn (47:52.821)
Well, it's funny to say that there's literally it's my team now has Matt agents when they bring me a you know We have a doc culture we do a lot of writing at Amazon and before they bring me a doc to review They'll run it by a very skeptical Matt agent to anticipate what I'm gonna criticize and it's amazing how the docs have gotten better And I'm like hey guys. This is great So yeah

Jason Hiner (48:01.178)
Yeah.

Jason Hiner (48:09.807)
Wow. That's very cool. So Matt, can you talk a little bit about your, mention your role at AWS? Talk a little bit about that, how long you've been doing it, how you ended up, how your sort of role evolved into what you're doing now.

Matt Yanchyshyn (48:27.54)
Yeah, I've been at AWS for 13 and a half years now. I'm based here in New York City. I actually started as a Solutions Architect. So I left, was with the Associated Press for almost a decade doing technology with them. And I was hired by the AP as a Solutions Architect. And for those who don't know, that is basically work. It's a form of technical pre-sales and post-sales where you go into engagements and help customers put things together. It was a lot of fun.

Jason Hiner (48:53.914)
Yeah.

Matt Yanchyshyn (48:54.968)
And then, but I could always code and I did product management as well. So I slowly over my career, latter half of my career shifted back into engineering and product management. So I lead AWS Marketplace and partner services. And the easiest way to think about that is all of the ways that customers do business with third parties, what we call partners across AWS. There's the marketplace itself, but if you're buying a private offer for Anthropic through Bedrock, that also uses marketplace under the hood. If you're buying Oracle DB through our RDS database service, that also uses some marketplace services. So.

Any way we expose third parties across our services that uses sort of my stuff under the hood. And then if you're a partner and you're doing business with us, you need to do, funding requests or you want to get trained and all these, have, you know, I think over 170,000 or more than that partners now in the partner network and they use my service partner central to go and log opportunities, get trained, get funding. And we try to automate doing business with AWS with agents as much as possible.

So those are the two halves, the customers with partners and partners with AWS, and I run a suite of services that runs that. And it's been an amazing ride. run my actual team. I do a lot of things at Amazon. My actual team is primarily software developers and product managers. And so that's why I'm heavily focused on AI power, developer productivity, and just productivity in general.

Jason Hiner (50:07.152)
Okay.

Jason Hiner (50:13.902)
What has the AI boom, how has it changed your job over the past couple years? Clearly, AWF is one of the places that companies are going to move quickly into AI. There's a lot of sense of urgency, in some cases FOMO about it, but how has it changed your job and your company, your team, should say, within AWS that's really focused on these things, also, so like how's it changed your job itself, but also,

How's it changed what your team is focused on and the ways that you're trying to be at the leading edge of this and help companies be at the leading edge of it?

Matt Yanchyshyn (50:53.961)
Yeah, well, mean, it's actually consistent, you know, with our discussion today and previously, I really believe in and try hard to drink our own champagne, to use the same technologies that we're trying to sell or co-sell to our customers. And, that came up a few times today, but that definitely extends to my engineering team and the products that we launch. You know, if you follow my LinkedIn feed, and by the way, I hate hyperbole. You know, I live in fact, I live in, and so if you look at my LinkedIn feed, it's not hyperbole, it's not fake.

Jason Hiner (51:17.122)
Okay? Okay?

Matt Yanchyshyn (51:23.283)
The velocity of features that we have launched over the last year has increased or last two years really has increased dramatically. And I can say unequivocally that that's primarily because of AI and we measure it very carefully. Last year, our productivity sort of normalized across a few different domains, went up by over 30%. So with, you know, the equivalent resources we were able to, and that's reflected in features. Now it's not a one-to-one ratio between AI power productivity and the number of features, because features come in different shapes and sizes, but no matter how you measure it, like,

Jason Hiner (51:49.626)
Okay?

Matt Yanchyshyn (51:52.886)
Codes checked into repository, code reviews that we do. Code reviews went up by over 80 % year over year. And the quality of the code actually, it doesn't just mean that they're shipping junk. They're shipping higher quality code because we're using it for better automated testing and unit testing and their ops is going up. The number of people we need to support our software, we're able to support a higher ticket volume with the same or fewer people. so that, end of the day, we're delivering more features to customers than we ever have before because...

Jason Hiner (52:02.564)
Hmm.

Jason Hiner (52:06.445)
Okay.

Matt Yanchyshyn (52:21.119)
we're drinking our champagne. My team uses Kiro, the AI powered, it's a spec driven development tool, almost exclusively now. They use agents to do, I mentioned ops tasks. We use the QCLI now, Kiro CLI. We use Bedrock. We use Bedrock with Coheer and Anthropic to power our AI powered searches that you and I were talking about internally. We are a customer, like anyone else. I use QuickSuite, like I said, every hour of every day. And so our, I would say that,

The changes I've seen, especially in the last year, but over last two years roughly powered by AI, are unprecedented in my 13 years. I mean, there was a big shift, if you're old enough to remember, from agile software development years ago, or even the switch to Java, that had some improvements, but it is nothing compared to what we have seen with the improvements brought by things like Bedrock and Kero and QuickSuite. It's wild, and we're customer number one. We're using the same tools as anyone else.

Jason Hiner (53:03.023)
yeah.

Jason Hiner (53:18.296)
Matt, do you have time for me to ask one more question about Bedrock? Okay, okay. Okay, so, you know, one of the things that we've seen at the DeepView is some real interest in open source models, some growing interest in open source models, in smaller models, domain specific models. On Amazon Bedrock, you also allow AWS Marketplace, you know, in AWS, allow...

Matt Yanchyshyn (53:20.457)
Yeah, yeah, go for it.

Jason Hiner (53:42.925)
allow companies to choose the models they want to use. And that can have a big impact on the cost of your project, on the things that you're building. And can you talk a little bit about that? you also seeing that? You offer a lot of different models, including many of the open source models, Chinese models, open source models, even OpenAI's open source models, which are not very well known, GPT OSS, but are actually pretty good, at least from the data that we've seen. Can you talk a little bit about that? What are some of the trends that you

that you're seeing there.

Matt Yanchyshyn (54:15.145)
Yeah, so you're right, by the way, those GPT-OSS models are awesome. I have a lot of customers who use them and we've evaluated them for a things. Just as is Llama, for example, from Meta, another open source model that is supported by Bedrock. Yeah, this whole model choice thing on Bedrock is amazing. like we've discussed, I'm customer number one of these services. And so we, as an AWS marketplace, use model choice. And I actually think we're a good example of a customer. So for example, in Gen.ai search,

Jason Hiner (54:22.81)
Okay.

Matt Yanchyshyn (54:44.341)
We use a mix of Amazon's novel model, Anthropic, Claude specifically. We use the Claude four, now four or five series models. And we use a great AI company from Canada called Cohere. They have a rerank model that we use for search rankings. for like the search bar at the top of the marketplace site and agent mode, those are different types of use cases. We use different models. We have leaned heavily on Anthropics models, again, served through Bedrock. What allows us to do this and what allows our customers to do this in this quarter of the strategy is

Jason Hiner (54:56.196)
Yeah.

Matt Yanchyshyn (55:14.613)
is bedrock because then you only have to sort of code to one service and one framework, but you can access a bunch of different inferencing engines or models under the hood. And that's been key. So you don't need to go kind of rip out and replace all of your API calls. You can use bedrock and then use those models. We also train models. And I think you talked about open source models. It's not just because they're fast or good and often they are or they have good cost performance. Like man, this space is changing so fast. You have to be able to switch to

Jason Hiner (55:43.056)
Hmm.

Matt Yanchyshyn (55:44.17)
new models and try new ones all the time. But the open source ones and solutions that are new that I love like Nova Forge also allow for customization. And model customization is still not something I see widespread across my customer base. It's something that we do. And we do see a lot of rag usage, if you're familiar with rag, to sort of real time augmented retrieval. But actual sort of model compensation and frontier models with Nova Forge, that's definitely emerging in an exciting space where you can have

Jason Hiner (56:04.407)
yeah.

Matt Yanchyshyn (56:14.325)
heavily tuned, heavily customized models and people are doing that with open source models for a while as has my team. So the benefit of having access to all these models is each of them have different strengths and weaknesses and each of your features may need different things. So I'm again a great example of using different models for different use cases and open source models allow easier customization. So that plus the sort of emergence of NovaForge and Frontier models means that you can have much faster and often more cost effective

and better trained models to answer those specific domain specific questions or use cases, et cetera. So yeah, as it matures, mean, agent mode, like we spoke about a lot, is a really good example. Like customers don't like to wait around. It's every millisecond matters. So we're working hard to find not only the right model, but tune and refine that model so that it can kind of get to those product comparisons as quickly as possible. With every millisecond, you lose a customer in churn, right? So we want to kind of remove those milliseconds. So speed matters.

Accuracy matters and it's a fast-changing space so choice matters. You need to have access to the best tool for the job and I'm a customer like no other in that respect. I'll use Nova, I'll use Anthropic, I'll use Llama, I'll use GPT, whatever is best for my customers and I'll use it all through Bedrock.

Jason Hiner (57:28.336)
So that really does confirm this trend that we're seeing, which is that there was this sense that it was going to be one model to kind of rule them all, that you were going to use one model. And there was this race to create the biggest, best, most performant model. But really, we're seeing a trend in the opposite direction, which is that models have different capabilities. Different models are better for different things. And even sometimes these smaller domain-specific models might be cheaper, faster, and more accurate for

specific things and so the orchestration of multiple models is where a lot of this is moving and one of the things that know Bedrock is sort of acknowledging that by letting you use a lot of different models.

Matt Yanchyshyn (58:12.723)
Yeah, and you know, we didn't get a chance to talk too much about it, but it's similar in the agent space with Agent Core. You know, you're going to have a ton of different agents built in different ways, potentially from a mix of third party agents, first party agents, but you need all those agents on a common gateway and being governed importantly, consistently. So it's a similar type of approach, you know, model choice, and you don't have to rewrite workflows and code effectively for every single agent. So whether it's, you know, bedrocks with multiple agents or Agent Core, or sorry, bedrock with multiple models or Agent Core.

with multiple agents, the same sort of design pattern applies. And I talked about velocity and efficiency before. Every line of code, every hour of my developer team matters because we got a lot to deliver to customers. And so the less we can kind of worry about the orchestration and kind of the engine that powers the models and the agents and more on just, hey, what's the best tool for the job that we can implement quickly, the better. The faster we can ship to customers.

Jason Hiner (59:09.722)
Very good. Well, Matt, thank you so much for your time being on the show. Great catching up with you. This space is changing so fast. Even from when we talked, you know, a month ago before AWS, it's clear it's already adapting really quickly. So grateful for your time and appreciate you being on this on the show.

Matt Yanchyshyn (59:26.613)
Thanks, Jason.