Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
The DARK SIDE of Code Migration | Apollo GraphQL CEO Matt DeBergalis
29 Plays3 months ago
Code migration is transforming how modern software is built, but few are talking about the risks that come with it.
In this episode, we explore how code migration is reshaping API security, infrastructure, and the very foundations of enterprise development.
From GraphQL to AI-driven architecture, we uncover what’s really happening behind the buzzwords, and why companies are struggling to keep their systems secure during massive migrations.
Whether you’re a developer, architect, or engineering leader, this conversation reveals the hidden trade-offs of code migration, the growing complexity it introduces, and how to protect your systems before it’s too late.
If you’re planning your next modernization effort or already knee-deep in legacy refactors, this is a must-watch. We break down practical lessons, real-world examples, and strategies to make code migration work for you, not against you.
Get Tern Stories in your inbox: https://tern.sh/youtube
Recommended
Transcript
AI's Role in Code and Enterprise Silos
00:00:00
Speaker
using AI to write code, it works to the degree that I have a systems level understanding of what I want the code to do and how I want it to work. If I had to bet on one thing that not going to change with AI, it's that enterprises will still have silos. Yeah, and a lot of enterprises are out there doing M&A and it's part of their strategy to go acquire another business unit. If you just think about what it means to arm every developer with a copilot or codex or an autonomous agent that it can supervise, right? We're going to get a bunch of microservices. We're going to get a bunch of new consumers of APIs. There's a giant integration problem
Introduction to Matt DeBurglas and Apollo's Journey
00:00:35
Speaker
because of that. Welcome back to Turn Stories. I'm your host, TR Jordan. And today i have an amazing guest, Matt DeBurglas, the CEO of Apollos. Matt, welcome to the show.
00:00:46
Speaker
Thanks for having me. It's great to be here. Awesome. So, Matt, um you and I go way back. We first met 20 years ago in the basement of what was my dorm at college at the time.
00:00:59
Speaker
And I've i been following your journey. I've seen you build ActBlue and go to a Meteor and then Apollo, which is nine years in at this point.
Transitioning from CTO to CEO at Apollo
00:01:09
Speaker
um Yeah, 10 in a month.
00:01:12
Speaker
Awesome. So here's what's interesting to me. Apollo came out of the problems that you saw at Meteor around APIs. um And you've been working on this problem sort of across those two companies since 2012 or so.
00:01:26
Speaker
um And you recently moved from the CTO seat to the CEO seat, which is really like says to me that you're doubling down on this problem. And it's a big commitment.
00:01:37
Speaker
um Tell me a little bit about what you're seeing at this stage that makes this problem continue to be worthwhile to work on 13 years in.
00:01:49
Speaker
Oh, gosh, there's so much there.
Apollo's Mission to Innovate API Platforms
00:01:51
Speaker
um Well, a lot it goes back to school for me because ah yeah we were both at MIT and that's where I really fell in love with open source and platforms and a certain kind of engineering ethos.
00:02:05
Speaker
And that's that's carried me forward through all of this. um mean, Apollo is an API platform and we build it because we we think it helps lots of people build great software.
00:02:20
Speaker
And that that's what gets me out of bed. That's why we do this. We we just want to see a world where there's better better tools, better ah experiences. i want that in my day-to-day life. I want my friends to be able to make better stuff. you know it's It's just so important to me. And the software that we use is is more and more um influential.
Challenges in Adapting APIs to New Technologies
00:02:43
Speaker
I spent time in politics. So much of politics has gone online now and it's become as much about the the technology and the software experience um as anything. And that's true in, you know, anything, art, science, commerce.
00:02:57
Speaker
So I think it's really important. I think it's worth doing. And i think the other, the other piece of it for us is about, there are, if you look at it,
00:03:09
Speaker
from a systems point of view, there's opportunities all across the stack to learn from maybe the patterns that we've used over the years and bake those into things that are reusable and battle-tested. And we've seen this rise of platforms ah you know in in cloud native with Kubernetes.
00:03:29
Speaker
ah React is a good example on the UI. And it's been interesting. we We're in the API space at Apollo. There aren't as many platforms there. And we think there's, you know, when you think about APIs, they they last a long time.
00:03:49
Speaker
APIs sit around for decades and decades. And just by their nature, because they're ah a rendezvous point between software systems, they're really hard to change. So you've got this interesting puzzle in the API world where companies have many of them.
00:04:06
Speaker
They're going to be around for a long time and they are just by their nature, fairly stable. And the question is, how do you, how do you adapt to a changing world around that?
00:04:16
Speaker
Like whether we're talking about AI today, or we're talking about the rise of mobile a decade ago, there are these big forces in the world where, Companies have to be able to move fast and and adapt and change.
Impact of AI on Microservices and API Consumption
00:04:28
Speaker
And the API tier is one of the things that I think has held people back. And we've we've been excited to make a big dent in that problem with what we do at Apollo. That makes a lot of sense. and I thought it was such an interesting decision to say, like, you know, there's ah there are...
00:04:47
Speaker
There's the compute layer, there's the API layer, there's front end. All these things need to work well together if you're going to build an awesome system. um But a lot of that, yeah, I was i was trying to think of other sort of competitors to GraphQL.
00:05:00
Speaker
It's not like REST is a single framework or a single spec. And all of the other things underneath and around it, you know gp GRPC or and you know Thrift, RIP, are more you know transport-y than... that's
GraphQL's Advantages in API Integration
00:05:16
Speaker
right.
00:05:16
Speaker
um than something that really like focuses on developer ergonomics. A lot of people see GraphQL as another kind of API. So you can put it next to REST or next to gRPC. And I mean, that's technically accurate.
00:05:30
Speaker
But it's really more ah query language that you can use with the APIs that you have. The idea of GraphQL really is i ought to be able to query my REST APIs, if that makes sense. Like I shouldn't be limited to exactly one API call at a time and getting back exactly all of the data from that API. That's a really limited way to think about an API.
00:05:53
Speaker
The big idea with GraphQL is It's like a database. i I might want a little bit out of this table and a little bit out of that table, and I want to join them together, and I want to make some transformation to what comes back. That's what makes databases so powerful.
00:06:07
Speaker
GraphQL gives us a ah language to do that with APIs. Because if you think about any piece of software, like think about ah ah you know ah an e-commerce application the stuff you're looking at on the screen came from a ton of different APIs.
Adapting API Strategies to AI-driven App Development
00:06:23
Speaker
And it's got to all be tied together in a very particular way. And every year or two, some new you know platform comes along, like maybe you have to build something for IoT or or you know now for the new AI platforms.
00:06:37
Speaker
Each of those means you have to go sort of rethink those combinations, rethink the the way you want to use your APIs. And the query language just makes that so much faster, so much better than handwriting a bunch of code to do it.
00:06:50
Speaker
That's where the the benefit of it comes from. It's the same idea as like, Kubernetes, we used to write code to deploy our software onto machines. We don't do that anymore, right? There's a piece of infrastructure that that does that for you. You just tell it what you want.
00:07:04
Speaker
and GraphQL is like that for your API calls. Yeah. And that makes that makes a ton of sense. I'm curious what you're seeing around... There is around AI in particular, there is, i think, a lot of information out there, especially, you know, the ah the Apollo customer list is pretty impressive. a lot of large companies on there.
00:07:23
Speaker
um Are you seeing a different sort of pull from them around how they're using and thinking about APIs with AI in the mix?
00:07:34
Speaker
Yeah, there's actually three. AI is such a big change. And i see it show up three different ways at at our customers. um One is you've got everybody's moving to agentic development.
00:07:52
Speaker
And we're learning how you do that well. We are learning that when you ask an agent to write code for you, the more that that code can sit on top of strong abstractions, the better it writes. It's better to write concise code. if you If you can use fewer tokens instead of more tokens, the LLMs get much, much better and people are learning about code quality and and all the the techniques that that drive that. So just when we think about If your options are writing a bunch of procedural code to call APIs in a certain order, but the alternative is is just writing a three-line or five-line GraphQL query, you get better results with the query.
00:08:36
Speaker
And um we've we've seen a lot of benefit as people move to that. Then you've got...
00:08:45
Speaker
So AI makes software really cheap to write because the engineers get so much more productive. right and It works better in smaller code bases.
00:08:57
Speaker
And so we're all about to be the proud owners of a lot of microservices. If you just think about what it means to arm every developer with you know a co-pilot or codex or or um an autonomous agent that it can supervise, right?
00:09:14
Speaker
We're going a bunch of microservices. We're going to get a bunch of new ah consumers of APIs. And there's a giant integration problem because of that, because every company is suddenly going to have to tie all that new stuff together with all the old stuff.
00:09:31
Speaker
And that's an API problem, right? And and the whatever strategy you had for integration probably isn't ready for a 10x jump in the amount of integration code you have to write if you just think about where things break and how this is all going to scale.
00:09:51
Speaker
And GraphQL you know goes right at this problem is our view, right? you you If you can do that integration work by just writing declarative descriptions of what those services offer and declarative queries of how you want to combine them together, that's a lot better than writing thousands and thousands and thousands of lines of of bespoke integration code for each new combination.
00:10:16
Speaker
And then the third thing they run into is ah everybody wants to ship agentic apps. and And this is where things get really interesting because, you know, to me, the story of app development over the last 10, 20 years has been apps get adopted when they meet the the user where the user wants to be met.
00:10:37
Speaker
And I think people are realizing just how much better the application experience is if you can put a little bit of agentic or freeform personalized attributes to it.
00:10:52
Speaker
And the thing about these AI apps is they they just, for APIs, it's totally different. They use MCP instead of REST, and they are inherently more freeform. So the world of, of I wrote an API and it has a very specific shape and it does a very specific thing.
00:11:10
Speaker
That worked pretty well for the old style of apps where you've got a certain set of screens that were carefully designed by people. But in a world where the app changes every minute or every session or between two different users,
00:11:24
Speaker
It doesn't seem like that model is really going to work. We're going to need a different way of thinking about what an API even is and how you think about them. You're going have a lot more of them. They're going to have a much faster pace of change.
00:11:38
Speaker
Just all kinds of interesting
Enterprise Integration Challenges with Silos
00:11:40
Speaker
implications come up. And ah that puts a ton of pressure on any sort of traditional API strategy or or or stack that that people are building around.
00:11:50
Speaker
Yeah, that makes sense. So here's here's the thing that... i I thought a lot about this probably about 12 months ago, and I still think a lot about this, is if you think about like all the things that are going to change, um if I had to bet on one thing that was not going to change with AI, it's that enterprises will still have silos.
00:12:08
Speaker
And yeah, it it's hard to imagine a world without that. um How... I think and think it makes a lot of sense, like in theory, to to add like more connectivity and to push, you know, push the the power of the tooling such that you can integrate more data together. Like that's, I absolutely agree. We're all going to own, you know, 10, 100 times as much code.
00:12:33
Speaker
But, How do you think about the idea of like, it it almost, it feels like a paradox to me that enterprises are just never going to be flat with, in on any dimension. They're not going to be flat with their organization. They're not going to be flat with their PNL and they're not going to be flat with their data access.
00:12:51
Speaker
So what is how does that tension like run into using something like a GraphQL or even just like influencing the API strategy overall? Yeah, and a lot of enterprises are out there doing M&A, and it's part of their strategy to go acquire ah another business unit. And just by definition, none of that stuff has been built in a coordinated way.
00:13:11
Speaker
But probably the goal of the acquisition was to tie it all together somehow. I'll give you an example. we We do a lot of work in the hospitality space. And if you...
00:13:23
Speaker
a If you go to a ah a conference for you know hotel operators, one of the big topics is moving to a model where you you want to think about not just renting hotel rooms for the night, you have a certain price for those, but a model that's more about the total revenue per guest.
00:13:47
Speaker
And it's just a fancy way of saying like, you also want to offer that guest a restaurant reservation. You want to get them, you know, a spa appointment. You want to sell them maybe for an extra fee, um the ability to check in early.
00:14:03
Speaker
And it's interesting because some of this is is coming. It's not just coming from the CFO that wants margins. It's There are travelers that want to know they can check into their hotel room early and they'll pay for that.
00:14:17
Speaker
And they will pick the hotel that offers them that. So there's this urgency to create this new kind of, you know, checkout experience.
00:14:29
Speaker
Sorry, like reservation experience online that that gives you this this this stuff. And if you go look inside the hotel, what you find is, The room reservation system goes back to the Mesozoic era.
00:14:42
Speaker
It was never built for the web, right? that's It's the thing that the the person at the desk is using when you when you show up to the desk to to like assign a room. And so you have to adapt that system.
00:14:53
Speaker
It isn't going anywhere. They're not going to rip that out. But you have to adapt it to this whole new kind of customer experience that they're looking for. So you can call it a silo. I just think it's the world we live in it's It's a consequence of building a ah large company that's durable.
00:15:14
Speaker
The world's going to change around it and you're going to accumulate all of these systems and capabilities. like the The ability to assign hotel rooms is sort of a core business capability of any hotel.
00:15:26
Speaker
And it's to the degree that they can wire that into these apps that they want to build or um you know partner experiences, because a lot of this stuff gets booked through credit cards now or or other travel partners. like you Just think about the modern way that people go make these travel plans.
00:15:45
Speaker
That's the heart of, I think, every large company's challenge.
Media Companies and New Business Models
00:15:51
Speaker
And
00:15:54
Speaker
I just think the the companies that are at best at putting those things together in new ways are the ones that they keep winning. They're the ones that you know just feel modern and they meet the customer the way they want to be met. And and there's a there's a big advantage to that.
00:16:10
Speaker
Yeah, I buy that. And there's a certain amount of like, you don't need like those silos to the extent that exist will will continue to exist. You don't need all of the data out of any particular silo. You just need the ability to book a room.
00:16:23
Speaker
How it gets done, it's like, that's all behind, yeah? Yeah. And I think it just, in practice, it's a lot faster to do what you just said and and make use of that capability in a new way than it is to go rethink the core idea of a room reservation and all of the systems and technology and everything that's baked into that. right like um or you know Media is another example. Media has changed so much.
00:16:55
Speaker
It was a content business And it was fueled by advertising. There was ah a certain business model for a newspaper or or any you know content publisher, but media is so different now. It's really a ah subscription ah ah model for a lot of companies.
00:17:15
Speaker
It's also about m and a And if you think about a property like New York Times, they've got product reviews with Wirecutter. They've got cooking. They've got the Athletic.
00:17:26
Speaker
like that That's a media experience in 2025. And but it it would be a mistake to think about rebuilding all of those things to eliminate the silo. That's not the point at all. The point is you want to, as a company, be really good at having separate capabilities it's how you scale it's how you it's how you grow um so i just think about it in terms of you know if those things are your lego bricks and the people that are fastest at making cool lego kits are are going to have the the the models that are are what people want yeah yeah that makes a lot of sense do you do you think that the world do you how much of this this knowledge like there's this same implicit like um
00:18:12
Speaker
I was going to say model, but model means a very specific like thing right now. It means an LLM. But there's like a there's a mental model. There's a yeah a um like a schema of what you want baked into any given API or or more precisely the usage of any given API.
Legacy Systems in API Strategies
00:18:28
Speaker
It feels like there is there's both a um there's like a chicken and egg problem here of you need to understand what is happening in You need to understand what what exists out there, which, you know, for ancient systems is like really challenging to understand.
00:18:45
Speaker
um And you have to go look at the usage of those things in order to understand both what is capable, what a system is capable of and what is valuable about a system.
00:18:56
Speaker
And i think i I think what I'm having trouble squaring is like you said that APIs live forever and like even new APIs live for quite a long time. How do you think about the kind of like accumulated cruft, not just of like the code that does stuff, but of, you know, the APIs and the ways to do things?
00:19:14
Speaker
um and especially how that interacts with like what is the right way to do things. It feels like building endless APIs on top of of internal systems in order to build new capabilities is is just another like several layers of
GraphQL Insights for API Optimization
00:19:28
Speaker
stuff to do archaeology through as we look five or ten years out.
00:19:35
Speaker
I'll give you an example. Tell me if I'm on the right track here. i in a traditional api like a REST API, you get back a pile of stuff, like a JSON response, right?
00:19:50
Speaker
And it might be really big. you you You could have a hundred, maybe a thousand different you know properties or attributes, pieces of data inside that response. and the whole thing at some level is opaque.
00:20:02
Speaker
Like you you call the API, you get back the stuff. and And the whole idea of an API is there's this abstraction layer, this boundary. So the provider of this stuff has no idea and does not care how this stuff's being used.
00:20:16
Speaker
that's That's like the idea of an abstraction. GraphQL plays with that a little bit um because with GraphQL, you you write a query and you you get really fine grain. You say, I want this property, this property, this property, and this property, right? And I kind of want to shape them this way.
00:20:32
Speaker
So one of the things we've learned is that if you own a microservice with an API, if people are calling that through a REST or a thrift or you know a traditional API interface, you don't really know what they're doing with that.
00:20:55
Speaker
Oh, that's cool. You don't know which fields are actually being used and which ones got dropped on the floor. You don't have any real way to know where to focus maybe your optimization work. you you're You're flying blind. like You just have no idea.
00:21:10
Speaker
But with something like GraphQL, you do. you You can see they only wanted these parts. they wanted you know You can also see they went off and called this other API next.
00:21:23
Speaker
to get these other parts. There's all kinds of valuable for a technology you know person who owns one of these APIs, there's all kinds of valuable operational diagnostic you know performance data that lives there.
00:21:37
Speaker
If you're a business person, there's business intelligence to understand how your customers are interacting with your capabilities. that that's always been hidden before.
00:21:49
Speaker
is that what you're getting at? like i'm Yeah, absolutely. I think that's we that's ah that's a perfect insight is that you actually know what people are consuming because they have asked for it and they don't have to ask for the whole thing.
00:22:00
Speaker
Yeah, it's ah it's an explicit sort of intent model instead of ah procedural approach where I just follow the instructions and I don't i don't know why.
00:22:11
Speaker
And this turns out to be ah delightfully good fit with AI because... when we think about what it means for LLMs to talk to APIs and and and that's what an agent is, right? It's just a model that can do stuff.
Security in AI-API Interactions
00:22:24
Speaker
The models are really good at describing intent. that's that's a fancy way of just saying a query, right? Like it's they're they're really good at saying, i i want the following thing.
00:22:37
Speaker
And that that I think turns out to be a good architecture to separate the non-deterministic part of the equation where the model can be creative and play around with whatever's going on in there.
00:22:47
Speaker
And then there's a deterministic part that says, okay, you want this, I'm going to go do it in this in a you know a way that that like we can observe and understand and and put some controls around. And that's the job in the API tier.
00:23:00
Speaker
Yeah. Yeah, that makes a ton of sense that you really see sort of like the right way to use this is that you declare your intent. I need these things. I am going to use them for this purpose. AI or not, like that's the right way to use GraphQL and that's the right way to interact with an API.
00:23:14
Speaker
And I don't know, we're really early in all of this. So um take all of it with a grain of salt. It's just what I think. But it it stands to reason... Here's another way of looking at it. every every Everyone I talk to is really between a rock and a hard place. They have some kind of a mandate to build AI-first experiences.
00:23:41
Speaker
Makes sense. On the other hand, i don't know if you've been following, but like the MCP... security track record is pretty bad, right? Like I think the batting average is is maybe one in three of, of these MCP services that get announced that don't have some gaping security hole that comes up a week later.
00:24:01
Speaker
yeah, It's just a tough place to be for anybody because you you've got to move fast. You have a mandate all the way from you know the top of the company to do this. But all the usual rules and requirements still apply.
00:24:15
Speaker
like You have to run a responsible business. There's here all kinds of stuff you have to get right. And I think one thing we've learned over and again is that good platforms are or part of the way you square the circle
Rapid API Integration for Agentic Applications
00:24:27
Speaker
there.
00:24:27
Speaker
Like it it lets you build fast, but on top of some principled stuff, it lets you build in a world where a lot of the underlying technology may change. Like MCP itself is really new.
00:24:40
Speaker
it's It's still changing every month. And yeah I don't know. if there's
00:24:46
Speaker
it It just goes to the idea you were getting at around how you tie these things together and what that looks like. Yeah, absolutely. um and i want to pick at this a little bit because um you know MCPs, the security um notion within MCPs, a lot of the conversation these days is around the lethal trifecta.
00:25:06
Speaker
If an MCP has access to private data, the ability to communicate externally, and exposure to untrusted content, it's very easy for somebody, an attacker, to get that agent to do something that the user didn't want it to do.
00:25:20
Speaker
Yeah. It seems like GraphQL would make this problem worse because it's so flexible. how How do you see that evolving, especially as we try to give more power to the tools and and let especially you know these sort of enterprises move faster when they do have these these individual silos?
00:25:38
Speaker
Well, what GraphQL gives you is a declarative language that explains what's going on. i guess the way I would say it is, GraphQL lets us analyze all of this stuff at build time instead of runtime.
00:25:53
Speaker
So the model, if if you if you imagine you've got a big pile of APIs, and and let me take a step back and just... The way we use GraphQL at Apollo is You've got a bunch of underlying APIs. Those are the REST APIs that you've had for a long time.
00:26:11
Speaker
You use Apollo to define, we call it a schema. It's it's basically just a catalog of what's in those APIs and how those objects connect each other. That's your graph. And then you write a query, ah which says, I want this, this, and this out of the graph in this order.
00:26:28
Speaker
And so Apollo, if you want an MCP tool, you define that MCP tool as a query. you You say, here's here's the here's the tool, here's what I want it to return to the agent.
00:26:40
Speaker
And it's it's defined not as code that you run when the tool's called, but as a query that gets executed. So the cool thing about that approach, you're right, it's very flexible. um ah That's what makes it so much faster to build an MCP tool this way. It's just like SQL is flexible. It's faster than writing a bunch of handwritten B-trees, right?
00:27:00
Speaker
moving But the the security story is good because I can look at a query at the time it's defined and I can reason about what data is it accessing? Is any of that data marked as personal information?
00:27:17
Speaker
I can write rules that say certain things can only be used in certain combinations. And we get all kinds of this stuff. Like, like in, um, We have bunch of finance customers and, you know, there's there's like legal requirements. This is downstream of an actual law in certain countries that says customer service reps can only look at a end user's billing record if there's an open support ticket from that user.
00:27:46
Speaker
Like you you just legally can't have a human support person looking at somebody's like financial transactions. How do you implement rules like that has been a big challenge for, you know, companies over the years.
00:28:01
Speaker
And now we want to have an agentic version of that, something where it's a LLM that's poking around your transaction records. And it just, it starts to get pretty complicated. And the more we look at that, the more we see,
00:28:15
Speaker
ah a solid sort of declarative model, I think is really the only way through because the alternative where you're just going to have piles and piles and piles of what code you wrote in your favorite Python MCP framework this week that like does what exactly? You know, it's just so hard to build that way if you're trying to go fast. And if you don't know which of those projects you're going to want to take to production and which ones won't make sense. And when there's, when there's so much at stake from a security point of view.
00:28:51
Speaker
That makes sense. How much do you see the I got to imagine that there's like the best customers in terms of like best practices and adopting, you know, not not just like the security of of but the this like security first approach to writing GraphQL, but just broadly making more data available.
00:29:11
Speaker
Yeah. What do you see as the most productive customers when they're trying to adopt a new technology at the API level layer? and And conversely, where do you see people making mistakes and not getting the adoption they want?
00:29:28
Speaker
With MCP? I was thinking with GraphQL. Oh. But if you've got information on MCP, like what, three minutes old? Yeah. There's a lot more of it happening than than I think people realize.
00:29:40
Speaker
what One thing I've noticed with MCP or or with agents in general is that the companies that have gone furthest with this stuff and are finding value are finding a lot of value.
00:29:54
Speaker
Really? And they don't always want to talk about it. I certainly don't care about it. It's a huge competitive advantage. Interesting. and And AI is interesting because it's coming fast.
00:30:06
Speaker
Like if we think about some of these big technology trends, really big stuff like cloud, cloud took its time. If you really think about like how many years was that? And and Amazon, I think to this day is still very fond of starting their earnings reports with what a small fraction of the total enterprise spend is on cloud, right?
00:30:28
Speaker
Like, like, it's just it's it's It's a gradual rollout. um I don't know. The SAI thing could just happen in a couple years. It's crazy. Interesting. and so if you're in a vertical and you've got some kind of an agentic experience that's better than your competitors...
00:30:52
Speaker
You may not want anybody to know that initially, right? You certainly don't want them to know how they're doing it. I don't know what my point was. How are people doing this? um but what What are those companies doing better than than others? like
00:31:08
Speaker
Well, I think with you know one model that emerged pretty early for agents is support. like customer support. And ah you can look at it as a margin improvement because agents are all in cheaper than the, you know, the human support people that you would have used before. But I think the bigger story is it's what customers prefer. Like you can use the agent 24 hours a day. it it speaks whatever language you speak. Like there's there's all these
00:31:44
Speaker
and And we're getting all kinds of examples yeah ah that that are you know published now around better support quality. So that's a model that's, that's I think, got legs.
00:31:56
Speaker
And it really boils down to how much of your... how much of your footprint can you wire up to that agent? Cause the agent's not very good. If it, if it helps you for a second and says, Oh, I can't actually do that part.
00:32:11
Speaker
Like that's the worst agent in experience, right? Like, let me talk to your supervisor and you're, you're pounding on, on like the zero key over and over again. um And it's ah at least what we hear is,
00:32:27
Speaker
because we're coming at this from the API side, is people just tell us like, look, my priority is to get as much of my API footprint agent ready as I can. I don't quite know what I want to do with it yet, but I know that an API that I have today, like ah a system or a capability that I have today that can't be used by an agent is probably not that valuable.
00:32:49
Speaker
And but i think sometimes you really need a really clear use case to to know how to move forward. But sometimes there are sensible sort of first principles or foundation things that you can get a jump on.
00:33:04
Speaker
that make a ton of sense. and And I think we're in one of those moments because there's just so many open questions about AI. Like, i think everybody reasonably looks at their phone and they you you know count the icons on your home screen and you think, how many of these are still going to be with me? Like, AI is going to change these a lot, probably.
00:33:23
Speaker
And so there's just, it's so hard to predict, at least I don't feel like I can say what AI really means for the apps of the future or the, you know, agentic experiences that you might want.
00:33:40
Speaker
I can imagine some things, but who knows? But what I do think we can say is that there are certain things in terms of how you're going to go build them or what the stack is going to look like that you're going to need in place and that we can start to get a clearer view of.
Platform Strategies for AI Disruptions
00:33:54
Speaker
And those are just timey or timeless things that you know don't change like we were talking about before. So we we try to come at it from that point of view. And it's a long way of of saying the ones that have approached this the best way, I think, are are taking a principled sort of platform view of it because they know that they're going to have to be really quick and agile.
00:34:18
Speaker
And the best way to do that is to is to start on some terra firma that you can trust. Yeah. I think that makes a lot of sense. I i remember i was an AI skeptic for a little while, which in retrospect, huge mistake.
00:34:29
Speaker
Totally wrong on that one. My bad. But my first my first theory was like, if we get this right, I could just talk to a company. like I can talk to Walmart in a way that is... Not possible today. And like we're starting to get there in in bits and pieces. And it turns out that Walmart doesn't actually want me to talk to it as a company with like full autonomy.
00:34:48
Speaker
But I understand that. um I think there is an interesting... ah I think there is an interesting um growth path here, though, where, yeah, the best companies are are exposing as much of the company as possible through user-friendly, AI-driven areas.
00:35:10
Speaker
um One of the things that I hear a lot, because I talk to a lot of people about like... let's modernize your software and let's let's upgrade it. And frankly, where we're focused right now tends to be people who own their software and like understand it. They want to do React upgrades or whatever. And those that's a well-trodden um bit of the world. But there's this whole other part of the world, which is like,
00:35:32
Speaker
um COBOL and mainframes and the you know the the transaction processing cores of the financial services, those seem like some of the most valuable things to expose, but you literally don't have teams that like know how they work anymore.
00:35:49
Speaker
So even if you have a platform team that takes a principled approach, starts on that terra firma, what do they do if nobody knows how the system works? Well, we we actually have a lot of customers that have um ah graph on top of mainframes.
00:36:02
Speaker
yeah And the the the story here is... but its Quick side note. My first internship when I was in school for a summer, I guess it was 1999.
00:36:15
Speaker
And i went to a startup that was... ah decompiling um IBM mainframe software because they didn't have the original source code in COBOL and patching it for Y2K compliance.
00:36:33
Speaker
like So we'd add a bunch of like little trampolines and it would jump to this little bit of code that we inserted into the... Anyway, I'd forgotten about that. love I was a compiler kid growing up. i ah I love that stuff.
00:36:48
Speaker
Yeah, so the... um It's exactly what you said. They've they've got a mainframe. It's running business transactions. It's not going anywhere. And the obviously, it was all designed long before we had the web. And i mean, it was designed before we had ASCII. Like, it's it's the stuff's old.
00:37:08
Speaker
Yeah. and um the point is, you've got to adapt. You've got to you gotta to build some sort of a translation layer or an adaptation layer so that you can use it in the modern world.
00:37:21
Speaker
And so ah i see this pattern a lot. We see it with ah legacy databases too. there's There's a ton of stuff out there that's on you know old Oracle stuff, and they want to move to a modern database.
00:37:39
Speaker
And it's really hard to do that if you don't first... create some sort of new abstraction that sits between the legacy database and the the rest of the software that talks to it.
00:37:53
Speaker
Once you've got the abstraction in place, now you now you can go over to just the database and you can make whatever big migration you want from Oracle to Postgres or you know Oracle to MongoDB, and you're not wrestling with 100 different users of the database all at the same time.
00:38:12
Speaker
So there's this pattern. It's it's sort of like a strangle pattern or a distraction pattern that lets you do that. I think with mainframes, it's just i don't I don't see as much of the appetite for retiring them. I mean, I i just think there's...
00:38:27
Speaker
It's not my area of expertise, but if you're if you're running financial transactions or ticketing systems i you know for an airline or or ah you know in a company like that,
00:38:42
Speaker
That mainframe is an asset and there's 50 years of business logic that have been baked into that and it's valuable, I think is the short way of saying it. um And so the question is just how do I use it in more places and how do I make it web ready or AI ready?
00:38:58
Speaker
isn't Yeah, absolutely. i think that that feels like the right approach for so many of these things. I think there's always appetite for getting rid of stuff that feels old or crufty, but it is so valuable.
00:39:09
Speaker
So, you know, ah what are you willing to pay to make that change? Like, sure, if it's free and snap your fingers, go do it. um But you got to look at the actual cost of both moving and the actual cost of of keeping it around. And most of the time the calculus says like, well, you know, but put a layer around it just keep using it.
00:39:27
Speaker
Yeah, I mean, I think the other thing is with, ah as with many things in the world, interest rates matter. And ah one thing I've seen is the appetite for Big Bang rewrites, re-architectures, transformations really waned as interest rates went from very low to where they are now.
00:39:52
Speaker
And the, you know, in, in, 2020, when capital's cheap, when you're happy to have a five-year return on your investment, and when you're really trying your best to hold onto your technical talent and give them an interesting, you know, technical problems, there's a lot more of this. And you saw it all across the stack and all kinds of places in 2025, where,
00:40:18
Speaker
um The talent market's different, where capital is more expensive, where the street wants to see ah better you know earnings per share number.
00:40:31
Speaker
It's just different. and And that lends itself, I think, much more toward a, what can we get done this quarter? How can we make incremental progress where we're we're we're reaping the benefits of our work you know immediately?
00:40:44
Speaker
That's the the mindset I keep finding. And it just leads to a different kind of transformation, a different way of approaching this kind of stuff. Yeah. ah There's a question that's on the back of a lot of people's minds.
Importance of Developers in AI-Driven Systems
00:40:58
Speaker
If we've got a world where the infrastructure is more flexible, better Legos allows for more combinations of what exists already, AI is great at writing glue code between all those things.
00:41:09
Speaker
What is ah developer's role in the next five years?
00:41:17
Speaker
Well, think everybody's... So starting to see, well, I'll say my own experience using AI to write code, um I have found it.
00:41:29
Speaker
I mean, I'm the architect, I'm the tech lead, call it what you want, right? But in particular, what I keep finding is that it works to the degree that I have a sort of systems level understanding of what I want the code to do and how I want it to work.
00:41:46
Speaker
And maybe that's just me and i've I've got a systems background. So that's how I think about writing software. But I i just find if if if I can be very prescriptive about the, you know, where state lives in different parts of the app or what the API interface or maybe the database schema looks like, and I'll get very in the weeds on that with the AI, the rest of the code between those interface layers, I don't have to worry about so much. Like I don't i don't need to know as much about the,
00:42:16
Speaker
you know, the the the TSX code that's like drawing the screen, as long as I understand sort of the data model underneath it. So i think, you know, models will get better at that level of reasoning.
00:42:31
Speaker
i don't know how quickly, I don't think anybody does. And, um And yet I think that same idea of you can't really operate at one layer of the stack unless you understand the abstractions below it is still going to hold true.
00:42:49
Speaker
and I just think as AI gets better, the developer moves up and up and up in the stack. um The world needs a lot of software is the thing.
00:43:00
Speaker
i I don't see any desire to, um I haven't met a company that doesn't have a backlog of software they really want to write. You know? yeah So I think we're a long way from any sort of more fundamental change in, you know, not having people that are responsible for software development. I don't know, maybe play this out for future and,
00:43:28
Speaker
The AIs run the roadmap and they figure out all the software and they write everything. But but that's that's not today. No, and I don't think it's five years away either. I i agree with you. i've I've had the same experience of if you're if you're clear about where is the data and where what what should it look like to operate this system? like there is There is an engineering way of thinking about systems that you're not going to get from requirement stock. and You have to know what you want.
00:43:55
Speaker
you yeah you got to decide. Yeah. You have to know what you want. I don't think AI is, is all that close to knowing what you want. Like it's, yeah it's not magic. It's just a predictive, you know, next token thing.
00:44:08
Speaker
Uh, and so there's, there's, there's a lot still for the savvy developer. actually, I mean, what I find is that it really rewards, um, that kind of skillset in developers, right? It's, it's, uh,
00:44:21
Speaker
it's a powerful tool in the hands of a developer that sees it that way. at least that's the way we're trying to use it, um, at Apollo. And that's been my experience. Makes sense. Um, we, we could talk all day. Let me move into a couple of quick fire questions.
00:44:39
Speaker
Um, and then we'll wrap up the, okay. I believe that every startup on some level is always bet. You're trying a theory, and if it works, it's going to be great. And if it doesn't, you're going have to come up with another bet.
00:44:54
Speaker
What is the bet that Apollo is making over the next three to five years?
Apollo's Vision on AI's Impact on APIs
00:45:00
Speaker
The bet is that ai and I walked through the three things it's doing, right? We're going to move to a different way of writing code. We're going to have a giant integration problem, and we're going to have new kind of application experience or new user experience.
00:45:14
Speaker
ai for all three of those reasons, ah really raises fundamental questions about what APIs are and how they're going to be used.
00:45:27
Speaker
And because of the urgency around AI, every company is going to need to make those APIs that it has a i ready. And we think Apollo can help with that. We think the technology is a good fit, the declarative architecture, the standards-based approach.
00:45:42
Speaker
Um, and we're excited to go see what we can do to, um, help, help companies get to a place where they're, they're able to make the best use of, of that, that giant force, that giant disruption, um, that's coming.
00:45:56
Speaker
Cool. I love that. Um, talked a lot about what technologies will sort of win and and work over in this new grand AI-centric world.
00:46:11
Speaker
What do you think just dies in 10 years? What technology do we look back at and say, I can't believe we wasted time on that? Well, that's an interesting question. Gosh. Well, one interesting thing I've noticed is The models are trained on what's on the internet, and they have gotten really good at the things that are most popular.
00:46:35
Speaker
And there's been this interesting effect I've noticed where it's a virtuous cycle for those technologies. Like the models are really, really good at writing React code.
00:46:47
Speaker
And I think that has interesting implications for the stuff that isn't as popular in any given domain. Like there are other view frameworks, right?
00:46:59
Speaker
And they have their own advantages, but it's, it just seems like AI drives a certain winner take all dynamic. And i suspect that that means certain frameworks, certain languages, even,
00:47:17
Speaker
may just find that um they fall further and further behind on
AI's Influence on Technology Adoption
00:47:26
Speaker
adoption. And again, it's a i mean vicious cycle, virtuous cycle, depending on on what you what you're aiming for. But i i think that's an interesting consequence of of the model companies focusing so much on the code authorship problem.
00:47:46
Speaker
and And I've learned over my career, I'm i'm a bit of a I've always gravitated to the oddball stuff in some ways. Yeah.
00:47:57
Speaker
And I've learned the value of making safe and conservative choices. I've learned the value of um you go with the open standard over the more you know feature laden experimental idea.
00:48:13
Speaker
You go with the vendor that's got the stronger business model rather than the superior technology in some respect, because the vendor with the strong business model is the one that can make the greater investment.
00:48:26
Speaker
And therefore over time, you know, and we saw this play out and in the database world with with so many companies. I mentioned MongoDB before. I mean, there's an example. MongoDB was was sort of known for a long time as as like,
00:48:41
Speaker
Not the best database in some technical respects. Especially of that crop of databases, that it was really obvious that Mongo was the worst at that point in time. And yet they had the strongest business because they understood the adoption journey. They understood ah you sell into a developer. They understood what really mattered to a certain audience and how to grow that audience.
00:49:06
Speaker
And ultimately that meant Mongo kept investing in the database and now it's really good database. yeah And if you picked one of the Mongo alternatives 15 years ago, you're the proud owner of a migration project now, right? like yeah so We've got several of them on this show.
00:49:22
Speaker
I've learned that there's there's value in the in in attributes that aren't always the sort of which product is best. And i I think AI seems to be another example of driving that and driving us toward a certain set of things that have...
00:49:38
Speaker
you know rich training data and just are are a better fit for the models in whatever way. Yeah, that tracks that. That absolutely tracks. um
00:49:48
Speaker
So many, so many different thoughts. But yeah, we've had a couple of people the don't know what that you know, don't know, the F sharp community, but there you go. Sorry, friends. um Cool. And last, last question, um which is really two questions.
Community Engagement and Open-Source Collaboration
00:50:04
Speaker
um Where can people find you on the internet? And what would you like folks to get in touch with you about?
00:50:11
Speaker
We're at Apollo.dev and we're interested in all things GraphQL and APIs. ah We're particularly interested in how can we help?
00:50:24
Speaker
What's hard about building agentic software um and how can we help? Where where can um some of the things that that you know we've talked about today and that that live inside the API tier um be of some value?
00:50:39
Speaker
And the whole company is built on open source and taking pull requests. So it's it's the only way we've we've made forward progress and and I think the best way to do it. So we we we welcome we welcome the input and help.
00:50:53
Speaker
Awesome. Well, hopefully a couple more pull requests will show up in the and your projects as a result. Matt, thank you so much. I mean, there's been a ton of interest in the MCP tools we've built so far. I've never seen anything like this. The energy level around MCP is bananas.
00:51:13
Speaker
I'm here in San Francisco. We we have you know multiple meetups a week. it's It's just incredible. and there's so there's something about MCP that harkens back to the early days of the web. it's it's It's sort of an open, there's almost like a mashup angle to it, right? Like I can make an MCP server for anything and then I can use the model to glue that together in interesting new ways. And I think some combination of that and the the you know the excitement in the investment community, um it it just feels like the highest energy moment I've seen.
00:51:47
Speaker
and um it's just a fun time to be doing this stuff. So yeah, the, the pull requests um are the least of it. We just, there's, this is one special time to go learn a lot together and build some cool stuff together. And we're, we're excited to be a part of that.
00:52:05
Speaker
Absolutely. i I, I've seen, I've seen so much of, so much of what inspires me right now is like watching people use AI weird because there is so much upside that is not already like,
00:52:17
Speaker
crystallized into a company and the one way to do it like yeah they're just people exploring stuff people doing stuff and figuring stuff out and it's it's a super exciting time to work around this pile of technologies yeah all right well thank you so much matt this is a great conversation