AIMinds #010 | Matt Hammel, Co-Founder & COO at AirOps

About this episode

Matt Hammel is the Co-Founder and COO of AirOps, overseeing go-to-market strategies, customer support, professional services, and community engagement. With a decade of experience in startups and scaleups, he's dedicated to enhancing efficiency and fostering growth. He's passionate about unlocking the promise of AI for all.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of AIMinds, Demetrios is joined by Matt Hammel. He shared his journey from consulting to going startups, he has worn many hats. Together with his co-founders, Alex and Berna, they created AirOps.

AirOps makes it possible for low-code tools and AI models, focusing on empowering individuals and businesses to generate high-quality outputs tailored to their unique needs. Matt shared some of his views on the potential of AI, the challenges of building an AI-first company, and what the future might hold in this ever-evolving technological landscape.

Three key takeaways from their conversation:

1. Simplifying Complexity - The challenge of making new and complicated AI concepts easily accessible is substantial. Matt emphasized the importance of customization and flexibility in workflows while making them intuitive and user-friendly.

2. Harnessing AI Potential - Businesses are beginning to see the true potential of integrating AI into their products or services. Matt was particularly excited about the broad class of models and techniques that AirOps supports, and the way users have been mixing and matching to create fantastic results.

3. Future-Ready - Adopting a concept of forward compatibility, Matt highlighted the need to anticipate and prepare for future use cases whilst building solutions for current value creation. AI is moving fast, and only those adaptable enough will ride the wave effectively.

Fun Fact: Matt, Alex, and Berna, the co-founders of AirOps, all worked together in the real estate industry for four years before collectively deciding to venture into something new of their own.

Show Notes:

00:00 Lifelong passion for diverse interests, career impact.
05:28 Startup experience, personal growth, strategic career decisions.
08:41 Joined co-founder, led teams, faced challenging environment.
11:44 Founding team's experience and startup motivation.
15:41 Surprised by AI's potential, questioned current approach.
18:02 Customized AI studio for high-quality outputs.
23:13 Simplifying complex concepts for user-friendly applications.
26:05 Community creates diverse, high-quality content outputs.
31:16 Challenges with wide applications and unknown future.
34:08 Point solutions vs wide scope is a challenge.
35:54 AirOps offers products and consultancy support.

More Quotes from Matt:

Transcript:

Demetrios:

Welcome to the AI Minds podcast, a podcast where we explore the companies of tomorrow being built. AI first. I am your host, Demetrios, and this episode is brought to you by Deepgram, the number one speech to text and text to speech API on the Internet, trusted by the world's top startups, conversational AI leaders, and enterprises such as Spotify, Twilio, NASA, the one that puts those rockets into space, and, of course, Citibank. On today's episode, we've got Matt Hammel, the COO and co founder of AirOps, a company that I have been intrigued with for the past half year because of how easy they are making building AI and building workflows with AI for the common man like myself. So, Matt, it's great to have you on here. How you doing, dude?

Matt Hammel:

Great to be here. Thanks for having me. I'm great.

Demetrios:

Excellent. All right, I want to start with a little bit about you before we get into AirOps, because Airops is you've got my heart, and it is cool to see all the different ways that Deepgram is using AirOps. But I want to know, is this your first company? How'd you get into tech? All of that good stuff. Can you start us off at the beginning?

Matt Hammel:

Yeah. I will describe myself as a lifelong, since even I was a kid, jack of all trades. The pattern, sort of recurring pattern of my life. The theme that connects it all together is finding or stumbling upon hobbies or certain interests that I go probably too deep in, in some cases, from being a yoyo champion in the third grade out of nowhere, to later in life, getting super into triathlons, to, in the last few months, learning how to DJ. I tend to glom on to some sort of passion and then really go deep in it. And I think that's carried into my career as well. I started off in consulting. Like many folks, I was kind of millennial when the sort of great financial crisis or the great Recession happened.

Matt Hammel:

Didn't really know what I wanted to do. Most of my friends kind of didn't get jobs right out of college. I was lucky enough to get one at a consulting firm, which is kind of the cliche way of people who don't know what they want to do with their lives, who kind of like a lot of variety get into. And that was early career. And I quickly realized, or relatively quickly realized, that, speaking to my penchant for going super deep, that I couldn't go quite as deep as I wanted in that career lane. And that's what ultimately brought me to startups and brought me to tech.

Demetrios:

Excellent. So you got the job as a consultant. You were like, you know what, this consultant thing is cool, but maybe not for me, because I want to go deeper. I've got questions and I need answers. Where'd you go from there?

Matt Hammel:

Yeah, I would say I always had a few friends go into startups after college. Was sort of, sort of fascinated by it, wanted to learn, I think I'm not an engineer by training, wanted to learn how to sell. Really was one big kind of, you get a bit of a taste of it in consulting, but I would sort of analogize it to what kind of account management or customer success is. You're trying to keep people happy versus closed deals, kind of getting your foot in the door is sort of done for you. So my first startup experience, I was living in New York, was in grad school at the time, needed both a way to pay my bills, but also really wanted to get a zero to one startup experience. And so I was the first business hire at a pre revenue, had a product, but fledgling product, medtech startup. One of those things where I went in wild eyed of, oh, working at a startup is going to be fun. And then they handed me a list of thousand people to call and said, get after it.

Matt Hammel:

So I would describe myself as first business hire, but I was first SDR, who had no idea what the hell he was doing, so to speak.

Demetrios:

Dude, that is so funny. I also got into tech through being an SDR at a company that had a product and it was a little bit wonky of a product. And so I'm trying to get calls with people and just spent 8 hours a day on the phone, and really every time I would book a meeting, it would be like victory celebratory dance around the house, go and give my wife, because I was working remotely, give my wife a hug and a kiss and be like, woohoo, we did it.

Matt Hammel:

I know the feeling. Yeah, it was an interesting mix where the company had a bunch of sort of problems, any sort of scaling problems that an early stage company has that I was able to bring some sort of expertise and structure and organization to, I would say, and then a sort of like big gap in my ability of being able to cold call, get on the phone. So I had to personally grow while helping that company grow. And I think it was a fair trade. I would say the advice I always give to anyone who's thinking about joining a startup, particularly if they're non technical or less technical, is to find a trade where you actually add value as quickly as you can, because ultimately, and if you're in it, you're going to capture just as much value from the startup and it becomes just an advantage to be able to add value on day one in whatever way you can, even if it's something simple. Hopefully did that and then ultimately realized that med tech and healthcare generally though, is a bit of a kind of passion of mine, is incredibly hard to one make money in. And the particular company I was with, I think I didn't see it having the long term potential, both from a growth and from a company perspective, as I thought, and so kind of sought out on going a little bit later, something, series A was sort of where I targeted for my next company.

Demetrios:

So then you jumped to a little bit bigger of a company, presumably more product market fit, and you were able to leverage the skills that you had learned. Because in those early, early startups, man, you can do like you said, you can do just about whatever you want because there's fires everywhere. You have to figure out if it's actually valuable to be doing that or if there's a bigger fire that you need to be putting out first.

Matt Hammel:

Yeah, exactly. I would say I was there for about a year and a half and most of the fires that were solvable we solved. And then the ones that were probably either going to take a big strategic shift that just wasn't kind of within my control to drive or just completely out of the realm of my capabilities, I said, hey, that's just not something that you're going to be able to move the needle on. And I think you're right. Having wanting to go somewhere with kind of real product fit, more scaling challenges, which again, just kind of given my background and interest in skill sets, was where I wanted to go. And as left would have it, went to a then series A. They just raised their series, a startup called Bungalow in the housing space. It's basically Airbnb for living with roommates is the easiest way to explain it.

Matt Hammel:

Nice. And met one of my current co founders, Alex, who was leading product there. We joined a week apart from each other and spent about four years total, three and a half years there together. I led, started off as bizops, but then over the course of three and a half years, led probably kind of five or six different teams in total. Everything from customer support to pricing to our sales function to compliance. So again, I would say at a larger scale, but the common theme of firefighting or getting thrown into the biggest bottleneck or hottest mess, to put it more indelicately of the business to try to help us up a scale. And across that time span, COVID happened and the way the business worked, it was a multi site, kind of predominantly, kind of single family housing co living product. So the way I would describe it, as soon as COVID hit, we were operating 1500 mini cruise ships all across the United States, where you had different roommates, each with their own lease with us, which you can imagine was a challenging macro environment mixed with a challenging kind of customer support and health and safety to a certain extent.

Matt Hammel:

Challenge. So no shortage of challenging times across a challenging time for everyone, so to speak.

Demetrios:

Yeah, I can imagine.

Matt Hammel:

And then somehow, I guess through grit, that team, I would say, I say learned a lot of the intangible elements of operating an early stage startup, of navigating through uncertainty, of hopefully what the ingredients of being a good founder are in that time frame. We weathered the COVID storm the business came out of. It was able to raise a series C at that point. And that was about three and a half years in. And Alex and I had always joked we were kind of work husbands and work spouses, so to speak. He had always had been a serial entrepreneur. Previously, I had always had a hankering to be a founder. And kind of long story short, both were ready to move on around the same time and then stumbled into the space where the sort of origins of AirOps begin from there.

Demetrios:

Well, that's perfect segue. Talk to me about AirOps and what was the inspiration behind it.

Matt Hammel:

So both just to set the stage of the founding team for a bit and how we all sort of know each other and then how the company evolved from there, I can go into. But so Alex and I worked together for four years at Bungalow. As I mentioned prior to that, he led product at Masterclass, which I'm sure you've heard of. And our third co founder, Berna, led engineering, led an engineering team there. And so Alex and Berna had worked together for several years, had been also kind of wanting to start something over the course of their career. And then Alex and I had the same kind of same intent. We started off in a space that was very near and dear to our hearts, both at Masterclass, before and then at bungalow, we had used a wide variety of tools, typically kind of what you would describe as low code tools, everything from Webflow to Zapier to retool, this kind of emerging class of tools designed to help teams scale, often with less engineering resources than building everything from scratch. And so the initial idea and kind of where we started off was building for that same audience as many startups do of our kind of former selves and former teams.

Matt Hammel:

How could we make them? How could we ten x the impact that typically the person who's owning an OKR in a business might have, whether that's a product manager, kind of a biz ops lead, the person who is the sort of named owner of sort of changing the arc of the curve in the business, if you will. And we always saw kind of ourselves early on, and then the people that we were hiring as being those individuals. And when empowered with the right resources, with the right data, with the right tooling, they can have impact outsized of what a normal individual can do. And so that was sort of the who. And again, somewhat of a general problem. But I would say we were typically sort of running up against a ceiling of what was possible using the tool set that I described to you, without either needing engineering to rebuild something totally from scratch once you added another zero to the business, or needing to throw bodies at something you've built or something that you had operating. And neither of those were particularly great options where we're quickly pulled. The direction we were quickly pulled as we started building, talking to customers, was originally towards the data space.

Matt Hammel:

One of the key bottlenecks to, again, if the goal is to have this operator be able to ten x their impact on the business individually, getting the right data was often an impediment to that. You're sort of driving blind of the levers you're pulling if you can't see clearly how they're impacting the business. Made some headway on a product we had there. Ultimately realized about, call it like six to seven months into that product being in beta, that it was just too complicated or a tad too technical for the original audience that we set out for. And so this was the middle to second half of 2022 we started using and playing around with. Then it was GPT 3. This is before chat GPT came out to make our original product easier for everyone to use.

Matt Hammel:

So things like writing SQL, analyzing certain data sets, being able to do some of the kind of like final mile technical pieces of our product, we were kind of blown away with what you could do with the right prompt, with the right data fed into a large language model. And we were like, should this be the product? It was the first time we thought, have we been focusing on the wrong, so to speak, the sort of like, wrong solution for the right audience? Right. I don't think we changed who necessarily we were focusing on. But we instantly had this glimpse of, wow, there are a whole lot of problems that we've previously experienced that our user base is currently experienced that could be solved with this technology. And right now. And even after chat GPT came out, there was still this big gulf between what you could do with Chat GPT and what you needed to build against the APIs of any, be it large language model or other AI provider. There's this wide Gulf of opportunity that became more and more clear to us as we started building, as the model started to improve. And ultimately we were faced with the crossroads in.

Matt Hammel:

It was like Q one of 2023, so q one of last year on the direction that we would ultimately want to take the business. And that was kind of the sort of persevere pivot, so to speak, moment that we faced.

Demetrios:

And so it was never losing sight of your end user, but realizing that things had changed, because you now have a whole new technology at your fingertips that you can leverage, and your end user can leverage more than you. You can just build that into your product, and then the end user can get immense amount of value. So I think here's a great point to explain to myself and people who are listening, all the ins and outs of AirOps, what is it as a product? What does it do?

Matt Hammel:

Yeah, so we quickly chose door B, which was, okay, we've made some really powerful applications of large language models, but AI models that we see the power being in allowing others to build sort of custom, very unique, or very, call it tailored applications of AI for their business need. And so the product evolved quickly into what it is today, which is effectively an application studio, a workflow builder that allows you to take and use really any of the best AI models out there. So all of Deepgram's transcription models, all of the GPT models, anthropic models, we are very model provider agnostic, be able to easily either use one of our templates or build from scratch workflows that typically take a couple of different form factors or building blocks that then can generate really high quality outputs. So typically those are one or more AI model that you'd be using, the prompts that you would feed into that, the examples that you would feed into that. So using kshot prompting and different prompting techniques, two retrieval. So the ability to leverage existing data sets that you have, long form text data, your documentation, your sitemap in a memory store, is the way that we describe our product. And then the third piece would be connecting to internal or external data sources that are able to help feed context into the workflow, into the calls to the LLM, and then being able to push those outputs into any number of systems, whether that be a CMS, a CRM, any number of tools that we provide both direct and kind of one click integrations, as well as a fairly straightforward process for setting up any API connection. So you can think of it as a workflow builder where the focus is really, and the differentiation is really on being able to get really high quality outputs that are a function of all the really granular and high quality inputs that we make easy to deliver.

Matt Hammel:

And so that in and of itself poses a problem. One, the use cases, the way I describe that are almost infinite, right? So kind of any number of business function, kind of a wide breadth of technicality, just the permutations of models and data sources kind of can pull you as a company or as a user in a lot of different directions, which I think to your question is one of the challenges we faced early on and one of the reasons that we really decided to focus on narrowing the subset of customers and use cases that we can get really ten x results from relative to just asking chat GPT something or a basic call to any AI model. So I'll pause there, but that's sort of what the platform does, and I can talk about some of the use cases and how we're thinking about that as well.

Demetrios:

Well, I think it's worth noting too, that when it comes to these workflows and the way that you're plugging in each piece, what I've found is it's very intuitive. And so it feels like you've done a lot of work to make sure that even though you're dealing with potentially, quote unquote complicated data sources or like a sitemap, for example, you don't really make it that hard on me to have to go and figure out, like, okay, this sitemap, how am I going to do this? I got to Google around what is a sitemap or what am I doing? I imagine most people that are using this and they are plugging in their sitemap, they have a bit of technical chops. But I was surprised when using AirOps at how seamless you made. All of these different connectors and the templates also help you get, they spark ideas, but then it's like, oh, well, yeah, in this template, I don't necessarily want to go with this option. I want to try something new, and then you can plug in whatever that new thing is. So that's one thing that I've noticed has been really cool to see and really like the empathy for the end user to start from zero almost, and have that ability to still take them and do really complicated things with them is appreciated from my side.

Matt Hammel:

Well, that's great to hear. Yeah, we really are focusing on taking what, to be honest, are new and complicated concepts that if you had asked someone what a vector store was, sort of even the 80th percentile engineer two years ago, they might have known, but kind of knowing what the applications would be and how to use that is a tall order. And so there's all these new techniques and concepts that are almost emerging faster than you can build features or documentation or even film content on how to. And so what we're trying to do is take the best of those, really simplify down how a user can apply them to their workflow, to their business problems, and to do it in a way that is relatively straightforward, but at the same time allows for the rich customization, which time and time again we found is the most important ingredient to getting, if it's content, really high content that ranks for Google, for example, or if you're creating a chat experience, getting responses that aren't hallucinations, that are really narrowly tailored to the inputs, to the data, to the request that the user is given, for example. And so that's a really thin line to tread between customization and flexibility and kind of the quality that you need while making something simple and easy to get started. So glad it was helpful for you. But that, I think, is a continual struggle and important one for us to have as we continue to build for our users.

Demetrios:

Yeah. Even something that is very much on the forefront and people are talking about quite a bit these days, like using an LLM as a judge and letting it decide if what the LLM before it spit out was actually factually correct or it related to the question in any way. It's super easy to set that up in AirOps and it makes it like, oh, wow, that is some really advanced stuff that you can just plug in and get working like in five minutes, man. And so knowing that you can do it like that in AirOps, it makes it really hard to be like, all right, let's set up our whole system and make it overcomplicate things and over engineer things. So I wanted to ask, though, kind of changing gears, what are some of your favorite template workflows or maybe more obscure templates that you've seen out there?

Matt Hammel:

I'm always amazed by what our community, what our users are building that we would never have ever thought of. I. I think it's interesting because content was kind of like the most basic and first use case, if you call it, of LLMs generally. And I think that it was kind of swept with a brush where you had what for the most part was not high quality outputs. And what I've been really blown away by and where I think we've done a good job in terms of templates and can obviously do better, is templates that work for generating content, but are very responsive to the individual data sets of an organization. Being able to take your data warehouse or live web research or kind of internal research you've done that you might have somewhere and blend those together in a way that is coherent and it actually generates really high quality outputs going through. Oftentimes these workflows that people will come and show us have 45, 50 different steps where they've broken the structure down into sort of a very meticulous outline and then have used the LLM and their data sources to craft each component of it. I think some of the use cases we've seen Deepgram build follow this, but being able to combine different, call them primitives for lack of a better term, into really complex but really high quality outputs around content I think is probably the sort of meta answer of the thing that has been just most mind blowing.

Matt Hammel:

And then the permutations of that that people have thought of that we never would have been able, if we were trying to list 1000 templates, this wouldn't be in the top 10,000 ones that we could have come up with. But people finding it and building it and then having it work is really special. I think thinking about where things evolve to with some of the new models, I think there are opportunities to use AirOps and this combination of different techniques and tools to create a lot more. Call it responsive AI. So you can imagine kind of a landing page that molds itself based on search intent. And so being able to have responsive experiences live for customers, for marketing purposes, the tools are just getting there and we're excited to have that be kind of one of the core use cases over time that we are investing in. And really, if you think about it, can really move the needle for numbers like customer acquisition, particularly in the world where you don't have tracking the same way you did. And CACs have just been kind of going through the roof for certain marketing and direct acquisition.

Matt Hammel:

So excited to continue to invest in our growth and marketing users who are really excited to get an edge and use AI in really thoughtful ways for their user experiences.

Demetrios:

That's pretty wild stuff. Thinking about a landing page that will morph to the search query and make sure that it is answering your questions based on what you ask. Google is fascinating. Think about that as like there's some futuristic stuff, right?

Matt Hammel:

Yeah, we have some early results from a version of that and it's extremely promising. And just see so many applications and tributaries of where that can evolve for teams and users. And really the core platform doing something that really requires specificity and precision to your own brand, to your own sitemap, to your own set of assets that you want to use. And so there's a lot of solutions out there that are very rigid in terms of what and how you can do it. And so it's an area where us being able to unlock flexibility for teams who really want to create thoughtful experiences is important, where I think we can be really helpful.

Demetrios:

So talk to me about some of the challenges of building a company that is not only just the normal company building challenges that you may have, but also now you are building a company that has AI at the forefront of everything that you're doing and it is dealing with AI in every way, shape and form.

Matt Hammel:

Yeah, I would say there's two challenges there. One we talked about a bit, the horizontalness, the wide range of applications of what you can do with this technologies is a challenge for us. I think we're getting to a good space on that one in terms of letting there's this cohort of users who they're going to build, what they want to build, and we're going to empower them to do that. And then there are cohorts of users where like, hey, I want to move this number in my business and here's how we're going to support you. So I think we're making a ton of progress on that. The second big challenge of being an AI first company or really being this connected to these core technologies is really, you don't really know what the next set of use cases to be unlocked will be. Frankly, we're still unlocking new use cases out of the current cohort of models and being to deliver real value, allow users to deliver real value with the current set of use cases and models while building a platform that as soon as the next wave come along, we are able to quickly allow those use cases, whether they be multimodal or very kind of hyper specific models. There's any number of vectors that value can be created in.

Matt Hammel:

And so keeping an aperture towards what will be possible and making sure that our platform is building both for the use cases that are really valuable today, but then giving our users a clear path on how they can then adopt the next wave of primitives, the next wave of use cases, and ultimately the next wave of meaningful impact to their business, I think is sometimes the most challenging, but it's also sometimes the most rewarding and creative aspect of the job as founders and thinking about the user's problem. And okay, we're going to solve that this way now. But you can imagine it being solved in this way in six months from now, just if you extrapolate the pace of progress. So a challenge for sure, but kind of the why you sign up for it challenge, if I had to describe it.

Demetrios:

Yeah, I've heard that explained as forward compatibility. You have a lot of people talking about backwards compatibility. Right. But you are trying to look around the corner and think, okay, this topic is getting talked about a lot and I think that's going to be the next wave. We need to start planning for when people are asking us for it because the new models may have that ability before we can blink.

Matt Hammel:

Totally. I think it's also a challenge, to be quite honest, for point solutions, if you think about it. I think the advice in SaaS in particular was always to start extremely narrowly, don't go multi product too early. Horizontal is challenging, which it is, but at the same time, in this world where just the model capabilities can render a previous technique that in the old world of SaaS could have been, would have been a company. It's a challenge if you're only trying to do one very narrow part of the stack, which is why I think our point of view is that the winners will be those who can kind of enable the sort of widest set of really valuable use cases or companies that are focused super specifically on a vertical. I think the legal, Harvey and legal have had a ton of success, but Deepgram, from what we've seen, has really been able to kind of cross just being fans of your product. That first threshold where the number of use cases really can sort of trickle them their way into the company. And we see that for ourselves as well.

Matt Hammel:

Right. Being able to have multiple workflows that then really drive great value for a company is one of our key objectives for any user that adopts us.

Demetrios:

Have you seen companies being built on.

Matt Hammel:

Top of had we. There's sort of two flavors of it. We have a self serve product that you can come in and just start building with AirOps and have started to see several features kind of go live for companies that want to adopt AI into their core product or their core service, which has been just super thrilling. The sort of second flavor of that we started to see there's some organizations who need some help getting a workflow in AirOps built. And so we've started to see kind of either individual or small consultancies kind of crop up building with AirOps, which is really cool. And we've started a program to foster that and create content and helpful how to's if you're looking to build and build a business with AirOps. And then the third bucket we've been working how we started the business. Kind of when we made that pivot was to really try to kind of sell into organizations to find the most powerful applications of AI that were being underlooked or weren't being served by the kind of Chat GPTs of the world.

Matt Hammel:

I keep coming back to that. And in that process of call it like sales led discovery or sales led customer development, we've worked with several companies who were really frankly changing the way their entire business works using AI and we've learned a lot. And now they're starting to go live with some of these solutions in their product that are, some of them are just mind blowing, to be honest, and going to have a huge impact on their business or already are having a big impact on their business. So we want to be supportive to all three of those buckets in the way that we can best. But I think we've only really scratched the surface for the really good use cases, so to speak. There's a lot of gold still to be mined with this class of models and techniques that we're excited to support.

Demetrios:

It's very cool to see what you're doing. Matt, thank you for coming on here. I've learned a ton. I really appreciate hearing your story and what you all are doing at AirOps, so I'm very appreciative of that.

Matt Hammel:

Thank you, Demetrios. It's great to spend a little time and hopefully this is helpful. I had a blast.