Axon CPTO on How Law Enforcement Uses AI to Save Lives, Not Just Hit Metrics | Jeff Kunins | E303

Jeff Kunins 0:00 One example of that is we announced a couple years ago a moonshot goal over a 10-year period. We want to reduce the number of gun-related deaths between police and the public in the US by more than 50%. We were the very first tech company to declare a moratorium on using facial recognition technology. The risks and the capabilities of the technology were not yet at a place where you could get enough yield and value relative to the inherent biases and risks. I believe in not rebuilding anything that you don't have to, but building exactly what you have to to be differentiated in the market. If we all shy away from those issues, then progress never happens. It is our job to build products that make it easier and easier for police departments and everybody else to thoughtfully choose the policies and approaches that are right for them and their communities. If your org chart fits too perfectly on a PowerPoint slide, it's probably the wrong org structure.

Carlos González de Villaumbrosia 0:58 Hey, this is Carlos, CEO at Product School and your host on The Product Podcast. Today's guest is Jeff Kunins, Chief Product Officer and Chief Technology Officer at Axon. Axon ingests more video per year than YouTube does. They created the Taser and the body cameras that federal agencies wear, and today they are the leading public safety technology company in the world. With a market cap of approximately $32 billion and $2.78 billion in revenue growing 33% year over year, Axon is one of the highest-growth companies in the S&P 500. Jeff has spent more than 40 years building products used by hundreds of millions of people. Before Axon, he served as Vice President of Alexa Entertainment at Amazon, and prior to that, as Vice President of Kindle's Global Reading Experience. In our conversation, we cover how law enforcement agencies are using AI inside body cameras and Tasers to save lives, not just hit metrics. Why Axon declared a public moratorium on facial recognition AI for six years, and what finally changed. Embedding external activists and researchers directly into product manager squads as a design input, not a compliance process. Building first-party AI models for real-time license plate detection while using foundation LLMs for everything else. Let's get into it. Welcome to The Product Podcast, Jeff.

Jeff Kunins 2:13 Thank you so much, Carlos. Thanks for having me.

Carlos González de Villaumbrosia 2:16 I'm excited to have you. I think you bring some uniques, unique aspects to the table compared to some of the guests that I usually host, where one is that you guys are actually building AI for the real world. Like, literally, there is a connection between AI in the software world with, with hardware. And the other thing is that you guys are running a multi-billion dollar organization that not a lot of people know about.

Jeff Kunins 2:38 It-- That is so-- That is both fun and exciting and a little unique about who we are. We like to think of ourselves as a Goldilocks on many, many dimensions.

Carlos González de Villaumbrosia 2:46 So let's start there. You are the Chief Product and Technology Officer at Axon. What, what does Axon do?

Jeff Kunins 2:52 Yeah, that's right. So I'm the CPO and CTO at Axon. I've been here about six and a half years, uh, and if it were even possible for me to be more excited now about what we are, what we stand for in the world, and the growth ahead of us and our mission, uh, I would be. Uh, just before I hit Axon, you know, prior to Axon, which, as you say, not all your readers may, uh, be as familiar with, I spent about 25 years in both big tech, classic big tech and startups, including five and a half years as a VP at Amazon, and I've lived through more than 15 acquisitions at this point from both sides of the table as both acquirer and acquiree. Uh, so kinda three things about Axon. Number one, we're the leading public safety technology company in the world, uh, and we're actually one of the highest growth companies in the S&P 500 over the last handful of years. So you might be familiar with us, uh, as being the leader in things like body cameras for police. Anytime you see a body camera video on TV, uh, it is almost always ours. You might see our little logo in the upper right corner. Uh, and also Taser, uh, the company was actually founded originally as Taser over 30 years ago. I'll get to that in a moment. But we are massively more than body cameras, Taser, and the cloud-based digital evidence management software that powers both of those things. We are comprehensively helping to reinvent basically nearly every single hardware and software product category you can imagine that powers public safety around the world, both for law enforcement agencies of all sizes, uh, but also, uh, the federal government, retail, healthcare, and other adjacent markets. And just as one of many fun facts, we actually ingest more video per year than YouTube does, uh, um, uh, just to give a sense of scale. And then one last thing is that, uh, we are-- Like, I'm a big believer in founder-led companies, and Axon is still founder-led after more, uh, than 30 years, uh, which helps us relentlessly focus on our mission, uh, to protect life and obsolete the bullet.

Carlos González de Villaumbrosia 4:52 I, I've been taking notes because I want to unpack each of those, uh-

Jeff Kunins 4:55 Of course

Carlos González de Villaumbrosia 4:55 ... facts that you, that you shared. Um, maybe starting from the last, right? A founder-led business still after 30 years and a massive scale. Um, as a founder myself, I'm always very curious to know how has the founder been able to kind of scale with the company, and what are other, um, people or systems that have been put in place so, you know, the, the founder never becomes a bottleneck?

Jeff Kunins 5:20 I think when the history of the world over the, you know, is writ- written, uh, and looks back at these decades, I, even though you may or, may or may not have heard of him, I think unquestionably Rick Smith will go down as one of the most impactful, uh, tech founders of the last 50 years, which again, may be surprising if you don't know who he is, but I promise you that you will. Uh, and so Rick, uh, more than 30 years ago, in 1993, when he was a grad student, uh, uh, two of his, uh, high school football teammates were shot and killed in a road rage incident, and he was, uh, both horrified and inspired that the notion that human beings still kill each other by accelerating metal slugs at one another, AKA guns, uh, should be a solvable technology prog- uh, problem. And even though we're still far from solving that problem completely, he dedicated his life to building a company, uh, to obsolete the bullet and to protect life in the broadest sense of the word. And he actually went and found the literal retired NASA scientist that had first come up with the idea of a taser, uh, but had failed to commercialize it, and in a literal proverbial garage in Scottsdale, Arizona, founded Taser. And now 30-odd years later, Tasers, uh, uh, have saved hundreds of thousands of lives around the world. And, uh, to this day, they're the-- If you think about any use of force to try to deescalate a threat and compare its effectiveness at deescalating a situation with the frequency of severe injury or death, Tasers are the only use of force from bare fists up through an assault rifle, uh, that hacks the curve of more effectiveness equals more injury and death. And Tasers have extraordinarily high, uh, effectiveness rates with near zero, uh, serious injury and death, and that was the cornerstone of what started the company. And then as it evolved over time, it went to body cameras as a means at first to just sort of capture truth on both sides of the lens and help improve behavior, transparency, and accountability both for community members and for cops. And that necessitated cloud software to go along with this explosion of digital evidence, uh, uh, files and storage that went along with that. And like most, uh, non-linear, uh, startup to big tech company journeys, uh, uh, the company evolved to be able to do what it needed to do to scale to the mission and to scale

Jeff Kunins 7:43 to the moment. And now 30 years later, uh, Rick is still foundationally at the core two things. He is, uh, he spends a massive amount of his time out in the field with customers, uh, building and helping, uh, uh, uh, develop the vision, building and helping, uh, build customer trust in these extraordinarily, uh, trust-centric markets. And then on the other side, he is still, uh, deep, deep, deep in as an inventor and as a technologist in individual key specific pieces of our broad portfolio. That's my job to take his sort of infinite idea generation machine and translate it into what we can actually build and actually ship. Uh, and he polarizes his time between those, those poles and lets the rest of us run and operate, uh, the company at scale.

Carlos González de Villaumbrosia 8:29 I mean, being an inventor is very different than a business operator, right, at, at, at the scale of your company business. So for you as the CPTO, um, how is the, the kind of the C-level structure, who do you report to, just to make sure that he can continue focusing on his superpower while you can continue fo-focus on, on yours?

Jeff Kunins 8:47 That's right. And again, and the same thing I, I was told a long time ago by one of my former leaders at Amazon that, you know, if your org chart fits too perfectly on a PowerPoint slide, it's probably the wrong org structure because, you know, organizations are dynamic and fluid, and you wanna optimize them to the moment and the people and what's happening, you know, in, in what the organization needs to do. Uh, and so Axon, you know, basically, uh, we have, uh, Rick as the CEO, uh, and founder, uh, and then we have a president, a guy named Josh Isner, who ran sales for a very long time and then has stepped up to basically help run and operate, uh, the company as a whole, uh, so that Rick can focus his time in those two polar barbell areas that I was just talking about. And then all the rest of the C-level Section 16 officers, myself, our CFO and COO, our head of legal, people ops, et cetera, the three or four of us, uh, report into that president, uh, so that Rick, uh, can spend all of his time on those two core

Jeff Kunins 9:43 things.

Carlos González de Villaumbrosia 9:44 Well, well, now I'm going to digging into the product org specifically. I can imagine that first of all, as product and technology officer, there's like a deep integration between business and, and tech, but also software and hardware, right? So how do you actually design your org?

Jeff Kunins 9:59 That's right. So we are fundamentally, and especially in this world of the, you know, worries of the SaaS-pocalypse and hardware is cool again and all of these things, just to get a sense, again, Axon is fundamentally, uh, and has always been a hardware plus software

Jeff Kunins 10:13 company. We are both ultra-high scale SaaS software, and we go to market in a very interesting way that we like to call the sort of Amazon Prime-ification of how public safety buys its technology, and this will relate to the org very directly. Um, so like a typical police department, you know, it-- you know, Atlanta, New York, take your pick, uh, uh, or, or a small one, they will basically typically pa-- buy with us five or 10-year-long contracts where they, for one simple all-in price, purchase a sort of meta Prime-like subscription that covers all of the hardware and software and, and consumables that they need for the sort of tier of things that they want. So that might include their Tasers, their body cameras, uh, their dronest first responder docks, a ver-vir- uh, VR headsets, all of those hardware pieces, all of the consumables like cartridges for Tasers and things like that, and all of the SaaS software, uh, across all of our various things, from 911 call center things to a bunch of AI tools to the core digital evidence management and everything in between. And we, of course, have multiple tiers of those things, but it gives a predictable, easy Amazon Prime-like way for a city or an agency, uh, uh, to buy the full suite of hardware and software that helps them get their, uh, uh, community safety needs, uh, accomplished. And so then inside, the way that we're structured is, you know, has a lot in common with any other big tech company, uh, that spans both hardware and software, which is, you know, like, we are a, a mix. Uh, my organization looks like a mix of some matrixed and some functional. So essentially, my direct reports, about 13 of them, look like a mixture of, you know, SVP GMs of each of our core product lines of business. And for the most part, those people look like they grew up as product management leaders. They're very technical product management leaders, but they grew up more or less as product management people. And they, on average, tend to have product management design and some of those other functions hardline into them, and they act as the mini CEO of that line of business, um, while, uh, having dotted line into them the other core software, hardware engineering functions and AI and science. And then I have top-level functional leaders for hardware, software, and science. And then under each of them, there are dedicated single-threaded leaders that map one-to-one with those pillars or lines of business. So you get that good classic balance of single-threaded leadership and accountability with functional expertise and best practices for a large org. And of course, like I said before, uh, it's not all one size fits all. It's not that clean or cookie cutter. There are some places where we have a single leader that has all of the functions together under one roof, just depending on the nature of where we are in incubating or growing that particular line of business.

Carlos González de Villaumbrosia 13:04 It sounds to me similar to the way Apple designs the org with functional GMs that own P&L for specific products while still having some sort of, like, cross-functional leaders that support the different business units.

Jeff Kunins 13:16 That's correct.

Jeff Kunins 13:17 And then we also, another key thing is because security is so fundamental to everything that we do, you know, different companies put InfoSec in different places. I feel strongly, and we've been very successful having InfoSec as part of R&D. So our CISO is also part of my staff, uh, because the security lifecycle and everything about security is so foundational to actually what we engineer and

Jeff Kunins 13:38 build.

Carlos González de Villaumbrosia 13:38 So as, as we think about go-to-market, I mean, one of the unique things here as I hear you talk about is, is you-- not, not B2B. You are B2G. You sell to governments. I mean, public and safety departments within governments, right?

Jeff Kunins 13:50 That's right.

Carlos González de Villaumbrosia 13:50 So what does the go-to-market-

Jeff Kunins 13:51 And now, now, now it's enterprises as well, like large retailers and hospitals, but yes, predominantly to government.

Carlos González de Villaumbrosia 13:56 So how does go-to-market look like for those type of governments different than a more traditional B2B go-to-market motion?

Jeff Kunins 14:04 Yeah. And so I've, you know, over the course of my career, I've worked both in classic consumer businesses, classic enterprise businesses, you know, and now over the last six and a half years, government. You know, I mean, fundamentally, like, it has a lot in common with large whale enterprise sales. It's not the same, but it has a lot in common with that, meaning you tend to have, I, I like to call it a head-weighted market, meaning it's a relatively small number of customers, each of which are relatively big compared to the size of the market, which means they all think of themselves as a special snowflake. They all, uh, want very, you know, custom things. And a huge part of being successful is earning trust at scale and over a long period of time to build out, um, the, the breadth of your portfolio that they choose to take advantage of. And so the sales cycles tend to be, you know, on the longer side of things. They are, you know, complicated, you know, very, very intricate pr-procurement cycles, and it varies. You know, there's still a difference between, say, a small sheriff's department versus NYPD versus the FBI versus the London Met, you know, take your pick. But broadly, you have a lot of compliance, uh, things. You have a lot of complex procurement, where procurement also has to go to things like city council and, like, mayor approval and things like that. Uh, and all up, you're trying to, uh, make it as easy as possible for the customer to, to buy as it is for them to use. And that's why things like the go-to-market, uh, uh, packaging that I described, where we do this Prime-like set of bundles, has grown up over time as being super effective for our customers as well as good for our business because it makes their-- makes them have predictability of how they pay, right? Even, ev- It lets them take advantage of a very large basket of diverse products when and how they're ready while having predictability of their spending over a very significant period of time, which is very helpful when they're seeking approval from things like city council.

Carlos González de Villaumbrosia 16:07 You know, in B2B, it's common to think in terms of years, like when you, when you sign up for a product.

Jeff Kunins 16:12 Right.

Carlos González de Villaumbrosia 16:13 In, in your situation, what's kind of the average length for a product?

Jeff Kunins 16:18 Yeah, the overwhelming majority of our contracts tends to be about five years. Um, uh, we actually have very few that are less than five years, and we have a growing cohort that are, you know, more than five, can be 10 years or even more. But I would say the plurality of them are five years.

Carlos González de Villaumbrosia 16:34 Got it. Um, any military applications?

Jeff Kunins 16:37 Uh, interesting question. So we have a very substantial footprint in the federal government with more like fed civ, civilian agencies, so everything from the FBI and other parts of DHS to, you know, um, uh, inside the military, the, the non-true combat side of things, so a lot of, like, military police and base protection and things like that. In the raw, like, defense space, both US and, and elsewhere, we're relatively small there today. Uh, the closest thing or the, the beginnings of that is we acquired a couple years ago a very exciting company called Dedrone. You, you may be aware that, uh, of course, drones and then counter drone protection is an extremely hot space right now. Dedrone is the unequivocal global leader in, uh, um, uh, counter drone protection for the sort of f-federal and civilian, not, not battlefield case. Um, um, uh, and so for example, we are, uh, one of the primary people that are gonna be helping protect all the World Cup cities across the US, uh, over the next couple of weeks. Um, uh, but we have some deployments for Dedrone that are sort of true defense and military. In particular, in the early days of the Ukraine war, Dedrone was seminal in, uh, uh, helping Ukraine deploy thousands and thousands of early, uh, RF sensors for drone detection, uh, that are still in use today. RF is a complicated thing because, uh, in war zones, uh, RF is getting, you know, jammed and blocked more and more as there's drones that use fiber optics and other means to basically not have to emit or be suspect to any RF signaling. Uh, but that is one of the key places where it's a wedge into, uh, true military, um, applications, uh, for our portfolio. Uh, but it is a growth area, not a current large part of the business.

Carlos González de Villaumbrosia 18:24 I also want to explore KPIs or outcomes. I've heard you mention lives saved, right? That's a very ... In- that's an incredible outcome, first of all, and also pretty unique compared to maybe traditional B2B SaaS. W- they talk more in terms of, uh, net retention revenue or, uh-

Jeff Kunins 18:42 That's the number three-

Carlos González de Villaumbrosia 18:43 ... productivity gains. You, I'm sure you have some of those, right? I want to talk about, like, the bigger, the bigger outcomes.

Jeff Kunins 18:48 Yeah.

Carlos González de Villaumbrosia 18:48 Like, how do you actually go about measuring something such, as important as saving lives?

Jeff Kunins 18:53 Yeah, it's, as you say, it's very difficult, and, like, within, I'm a big, you know, spending a lot of time at Amazon, I am a, uh, we could do an entire podcast just on metrics and KPIs and setting them. It's actually one of my favorite interview questions for candidates of all levels, is how they set and think about, you know, the right numerator, denominator for the right thing to actually incentivize the right behavior, but we could put that aside for the moment. For Axon as a whole, I'd say one example of that is we announced a couple years ago a moonshot goal, uh, of that in over a 10-year period, starting a couple years ago, we want to reduce the number of gun-related deaths between police and the public in the US by more than 50%, uh, in 10 years. Uh, and that obviously is not a thing that we can directly entirely impact ourselves, but we believe we have a big role to play in that. So there's public, uh, records that sort of, you know, obviously, uh, record every time there is a gun-related death on either side, either a member of the community or a police officer. Uh, and we set a fundamental goal that we wanna be part of, be the change that we wanna be a part of, and to try to slash that number, uh, in half over 10 years. And there's a lot of our products that very directly influence that, like Taser, like VR training for police officers, like drones as first responder programs to help get eyes on scene more quickly, uh, and other things. And then there's, uh, many of our other products that help indirectly influence that by giving officers more time back to be able to actually do the real job of helping to save lives and, and not rush or anything like that and be able to be as effective as possible in the community, and that's things like our productivity tools, which we can talk about when we get to AI, like things like, uh, that we call DraftOne and some other tools. But that is a very clear specific KPI that's objectively measurable, even though it is, you know, not, uh, trivially easy to directly measure our specific contribution to it. But in the end, what matters is the outcome.

Carlos González de Villaumbrosia 20:52 I've had a couple of, uh, CPOs on the pod that were also talking about those type of outcomes. Uh, CPO of Waymo was talking about-

Jeff Kunins 20:59 Yes

Carlos González de Villaumbrosia 20:59 ... literally prevention of accidents, car accidents.

Jeff Kunins 21:01 Yes.

Carlos González de Villaumbrosia 21:01 And CPO of, uh, Samsara-

Jeff Kunins 21:04 Yes

Carlos González de Villaumbrosia 21:04 ... was also talking about, uh, y- I think you, you guys partnered some-somehow with-

Jeff Kunins 21:07 That's correct

Carlos González de Villaumbrosia 21:07 ... with some of these companies, right?

Jeff Kunins 21:09 That's correct. Companies like Samsara are awesome because they, um, obviously I l- I, I have been a customer of Waymo. We don't work directly with Waymo, but, like, Samsara is interesting because, you know, they, in this, uh, uh, very workforce centric set of businesses where you have large fleets of drivers doing interesting things, they are working to make those drivers more safe. They are trying to make therefore the roads for everybody else more safe. But it also, by consequence, is they are, you know, in a different context, but similar to us, they are putting sensors on vehicles that are out on the road, and there's data that those, uh, sensors are collecting that are, is very interesting and useful to help with public safety, and that, when done in an ethical, responsible, you know, opt-in and transparent all sides sort of way, we are c- mutually confident are, you know, helpful for driving positive outcomes over time.

Carlos González de Villaumbrosia 22:00 The thing that comes to mind as, as we, we talk about hardware is speed, right? Like, obviously in software, we can make changes much faster. I hear you talk-

Jeff Kunins 22:08 Yes

Carlos González de Villaumbrosia 22:08 ... about Tasers, drones, body cameras, VR headsets. I'm thinking, man, like, how can you actually upgrade those systems in ways that you can, you know, test, go to market faster without having to wait for a proper hardware upgrade?

Jeff Kunins 22:23 Yeah, yeah. I talk a lot about the need for an organization, and we think of ourselves as the hustle and grit of a startup with the scale and impact of a elite, large public tech comp- you know, large tech company. And doing that requires, uh, operating at multiple clock speeds, right? It, it, it involves being able to have some projects and some products that work at blinding s- you know, st- uh, garage startup level, uh, ingenuity and speed. Also, where the volumes or the nature of the program is such that you can afford to go to market super quickly with extremely, uh, early, uh, low volume versions of things that you can be a lot scrappier about than when you're talking about, you know, millions of Taser cartridges that have to have extremely ultra-tight, uh, uh, tolerances and all of those sorts of things. And so we have to be able to walk and chew gum at the same time. But one fun example of what you were saying, and again, this is the beauty of hardware and software, but also being aggressive at both scrappy and scale. So w- a new product that, new hardware product that we shipped last year, turns out was actually the single fastest major hardware product we've ever shipped in the history of the company, and it was, we went after, uh, doing, uh, you know, a thing that you might see in the news a bunch these days, is about license plate detection cameras on poles. We're already the leader in license plate detection technology built into police cars, and we, from the ground up, have taken a fundamentally ethics-first, responsible innovation-first way to how we bring this technology to communities and to police agencies. And we felt that it was important, given some of what's going on in that market segment and some of the concerns that are out there, for us to bring what we thought could be a better, uh, way that would give better crime fighting results for agencies with better transparency and auditability for communities. And so we decided to build our own, uh, uh, pole-based, you know, uh, ALPR camera system. And in less than one year, from deciding to say go to actually shipping GA with that set of cameras that are performing better, uh, than, than any other one that's in the market, uh, we were able to make that happen through, uh, you know, taking everything we've learned before, uh, incorporating some new, uh, aggressive ideas on how to do so, and ha- starting to have poles in the ground in less than one year. But then on top of that, to your point about software along with the hardware, of course, uh, we can both do, you know, and we have long, long, long experience with doing safe, secure OTA firmware updates to physical devices, but they're also all IoT cloud connected devices, and a large part of their functionality is a mix of what happens on the edge of the device and what's happening in the cloud. And so we can dynamically add fun- f- uh, features, capabilities, performance, battery life enhancement, et cetera, on an ongoing basis for the life of those devices without touching the physical hardware again at all through a combination of safe, effective, you know, firmware and application, client application software updates, uh, to the devi- devices in the field, and to continual cloud style updates to the cloud software.

Carlos González de Villaumbrosia 25:19 I think Tesla is one of those prime examples on how they are able to upgrade the software on a car-

Jeff Kunins 25:24 Yes

Carlos González de Villaumbrosia 25:24 ... and suddenly give it superpowers, right? Like, a lot of those cars now self-drive, and when people-

Jeff Kunins 25:29 That's exactly it. Agreed

Carlos González de Villaumbrosia 25:29 ... purchase them, they have that capability. So in your case, kind of connecting this with the, with the poles that you install, like, give me an example on how-- What does it take to actually put some of those, uh, pilots in place to actually test what works before you do a broader rollout?

Jeff Kunins 25:46 Sure. I'm, I think, you know, there's both not rocket science to this, and there's, you know, art and practice, right, and responsibility to it. So I think, you know, like, some of these things will, they will, they will sound like the same as anyone else would say that they do, but I think it's a matter of having the rigor and the practice to do them well. So number one, of course, you have to make sure that when you're building, uh, connected devices, that those devices pass all of the right security reviews and all of those things so that you can update them, but someone else can't update them, even if they, uh, or, or get, uh, uh, inappropriate access to them, even if they get physical access to the physical devices in the field. Uh, number two, you have to make sure that you have an extremely reliable OTA update stack so you can safely simply just deliver packages of new firmware or whatever to those devices in the field without bricking them. Uh, and then number three, uh, you need to be able to, uh, have a flighting system where it's straightforward for you to roll out new firmware, new application software, whatever, to portions of that fleet, validate and soak in things in, in, in the field with appropriate samples, and make sure that things are performing, uh, nominally compared to what you're expecting and tested in the lab. Uh, and then of course, lastly, for the pieces that are in the cloud, the same thing. You need to make sure that you're able to, uh, safely, uh, through, you know, uh, to safely bring new versions of cloud capabilities, uh, and be able to, uh, field soak those with portions of your, you know, device footprint in the field, get comfortable before you then roll out. And then in government and in, you know, some enterprise cases, you have the extra thing of there are some, uh, kinds of releases and some kinds of updates you can do that we can make, um, unilaterally on our own and just say, "These are going to everywhere," especially if they're things that are effectively exposing new settings or new capabilities that customers can choose to take advantage of but don't preemptively move their cheese underneath them. And then there are cases where, uh, we need to actually have, you know, customers make choices about when to adopt a given thing. And again, you need to get really, really good at that sort of version and release management.

Carlos González de Villaumbrosia 27:55 I think this is time to talk about AI.

Jeff Kunins 27:58 Yes.

Carlos González de Villaumbrosia 27:58 And, uh, very curious about real impact on these new LLMs, on the way you've been building and upgrading your products?

Jeff Kunins 28:07 Yeah. So first I'd like to b- I, you know, uh, while I'm always open, I would, and I would love a challenge to go head-to-head, I think, uh, I, I think that we are proud and believe that we're pretty much on the tip of the spear and would go toe-to-toe against almost anyone, uh, other than the foundation, you know, model companies themselves, at basically building every, building AI into everything we build and using AI to build everything that we build. Uh, so that's both how we are, you know, updating and upgrading our engineering processes and our people and the like, in, like, h- what tools we're using to write code and the like, and then there's actually, as you were saying, real world impact of using AI in the products themselves. Uh, we were-

Carlos González de Villaumbrosia 28:46 Jeff, sorry, just, just to clarify, is this, are you building your own models or partnering with some of those foundational models?

Jeff Kunins 28:51 We actually ... So across, um, we don't do foundation LLM models ourselves, but we've, I was gonna say, we've been doing AI in our products for a very long time, since even, frankly, just before I got here. Uh, and in the nature of our world, there's, I, I believe in not building, rebuilding anything that you don't have to, but building exactly what you have to to be differentiated in the market. So we, uh, there are a handful of key areas where our individual q- ins- internal science team, we build first-party models, more traditional, you know, traditional models. For example, you know, our, take license plate detection. Foundation models are phenomenal for a lot of things, but both performance, but especially price performance, they're actually not nearly as effective yet at doing ultra real-time, 140 miles an hour clo- cross closing distance, five lanes of traffic real-time license plate detection and vehicle attribute detection. We have a extremely high-performing, efficient, first-party set of traditional CV and other models for that that live a combination of resonant on our devices and then in the cloud, and that's a thing we do ourselves. Uh, by contrast, when we're, and I'll talk about a product called DraftOne in a second, of course, where we're using big, open LLMs, we are using, uh, best in class, continually experimenting the same LLMs that everybody else has access to. Uh, and so we are not precious. We're basically what is the most bang that we can get for the buck.

Carlos González de Villaumbrosia 30:15 For the use case, uh, uh, you mentioned before about y- video ingestion.

Jeff Kunins 30:20 Yes.

Carlos González de Villaumbrosia 30:20 I think you mentioned that you ingest more video footage than YouTube.

Jeff Kunins 30:24 That's-

Carlos González de Villaumbrosia 30:24 That is in- insane. I mean, the amount of data that you're collecting, like, how do you treat that data? What type of models do you use?

Jeff Kunins 30:32 Yeah, so traditionally the, the, that, i- in, uh, traditionally AI didn't ... Right? If you think, you, you fa- go back 10 years, um, it was pretty straightforward ingestion, storage, secure storage, and then core workflows around sharing, editing, redacting, you know, um, uh, uh, uh, examining, annotating, et cetera. So not the, the ... Even to this day, the, like, fundamental core workflow about ingesting body camera video, allowing the off- all the way sort of capture to courtroom, all the way through to, to prosecutors and defense attorneys and the courts and to the public for FOIA requests, all of those things is relatively straightforward, but just very high scale, and not a lot of AI is involved with that. Where AI starts to get involved is the interesting workloads on top of that. So for example, um, redaction is a key thing. Even long before LLMs, right, a huge need with a lot of this video, when it's involved in an important case or when it has to be sent out to a member of the public for a FOIA request or with the press or with defense or whatever, you need to redact, um, the un- i- inappropriate, you know, uh, uh, personally identifiable information. You wanna focus on the victims or the suspects, not on, you know, random passerbys on the street or computer screens that are involved and things like that. And there are huge staffs of human beings at police departments that traditionally do all of this by hand, painstakingly, like in Adobe Premiere, editing videos to make sure this stuff is perfect, and you can imagine how time-consuming that is. And we, for a long time, have had AI-powered redaction assistant tooling that helps speed up that process for that very specialized workflow. The raw technology is very similar to what you might see in other video editing and other tools, but we have it built in for our specific context, and it is only getting better as we start to apply other, more sophisticated models today.

Carlos González de Villaumbrosia 32:28 Maybe the counter, uh, case would also be interesting, like when not to deploy AI and kind of resist that temptation-

Jeff Kunins 32:35 Yeah

Carlos González de Villaumbrosia 32:35 ... to, to sprinkle it.

Jeff Kunins 32:36 A great thing about that is, you know, again, fundamentally, you know, we, uh, we take ethical, uh, and responsible innovation deeply seriously. We're one of the only tech companies that, uh, has had a continuously, uh, operating, uh, set of one or another independent third-party equity and ethics, uh, you know, advisory boards of one form or another. Uh, it's changed a bit over time, but we have continuously had one, uh, for, for many years, and they play a huge role in being independent advisors for us about how we think about straddling this very high stakes dev- uh, uh, dynamic of public safety and first responders, communities, and all, and, and everyone else in between. And so one of the things that Axon did, there was actually a New York Times article about this a few months before I joined back in 2019, we were the very first tech company to, say, to declare a moratorium on using facial recognition technology, because at the time we said, "There is no way that the juice is worth the squeeze," right? The risks and the, the capabilities of the technology were not yet at a place where you could get enough yield and value relative to the inherent biases and risks that were built in to what performance could be at the time. And we said, "Eventually w- it will be time to reevaluate that, and there will be facial matching or other use cases that are very relevant and very helpful, but it has to be right at the right time and the right way." And so for six years, we, uh, held to that moratorium. And then actually just about six months ago, we announced publicly and, uh, together with, uh, uh, some agencies in Canada, to start doing very careful, very rigorous, uh, in conjunction with our ethics, uh, council and things like that, uh, some careful evaluation of appropriately narrow and specific use cases to do limited amounts of facial matching, uh, technology to help again, we, what ... Our mantra in ethical innovation is every day we're trying to make the right things easier and the wrong things harder. Uh, none of these things are binary. These things are all very complicated. How do we make sure that we are on the right side of the arc of history in helping everybody get home safe more often?

Carlos González de Villaumbrosia 34:41 How is the new technology for facial recognition different than what casinos and organizations have been using in the past?

Jeff Kunins 34:48 So like a lot of the things that we do, the core elements are no different than anywhere else. The key is about context and, uh, uh, context and, uh, uh, expectations and consequences. So if you're in a ca- If you're in a casino or if you're in the airport going, you know, uh, through TSA or, like, any one of these things, and by the way, some of those reasons are reasons why things are starting to change. Even now versus, right, six months ago versus a year ago versus two years ago is very different than five or six years ago in the fundamental capability of the technology to deliver better results, right? The models keep getting better and better. But also, uh, then thinking about, about context. So again, if you're in a casino and, uh, they're, you know, trying to determine whether or not someone is cheating or they're trying to ... or someone who's there who shouldn't be there, the combination of, right, the stakes of the situation and the, the risk of irrevocable, immediate, bad consequences are much lower than when you're talking about, you know, a police officer's at, you know, out in the field, right? The things that could go wrong if you get it wrong and there's misplaced trust in the results that come from the technology are v- of a very different order than in some other use cases. But those other use cases have been huge because they've pumped more and more and more data into these models. They've also pumped experience into community members to just generally get more used to. We're all using it on our iPhones, you know, every single day. And the combination of those two things are starting to make it at the point where we're actually quite confident that there are, uh, appropriate, highly useful, and, uh, uh, appropriately safe means, uh, to begin to apply this technology, but we're in very active and careful testing.

Carlos González de Villaumbrosia 36:40 You also talked about the ethics committee. This is the first time I actually hear it from a company. I mean, I know there is public regulation, but, like, what do you mean by having your own ethics committee?

Jeff Kunins 36:51 Sure. Um, uh, so, uh, uh, you can look it up if you search for Axon Responsible Innovation. Uh, but basically we have an independent, uh, body that's made up of a, uh ... First, I'll say we are extraordinarily close to our customers, right? So we have nonstop continual engagement and very, very close relationships with, with law enforcement officers, with public safety, and other first responders, so our inputs from that audience is extremely high bandwidth and very rich. And we wanted to make sure that we have, uh ... and, and we also have a constant means of getting direct feedback and voice, uh, and influence from, uh, uh, from other constituents that our technology, you know, uh, cohabitates with, which is the, uh, the public as a whole. And so our Equity and Ethics, uh, uh, Advisory Coalition is a mixture of academic researchers and activists and other folk from, uh, uh, from communities, uh, that have, uh, a very interesting perspective, lived experience, research, roles to play in understanding how, uh, police technology, uh, together with police interacts with, is perceived by, has concerns by, uh, members of the public and, and certain communities within the public. And what we basically do is, uh, we actively seek their advice and counsel. Like, they're not a binding judgment group. We're not beholden to take any particular piece of advice. But it's a very highly transparent and proactive part of our product development lifecycle, where every single thing we work on, we have transparency, uh, with that group. And then we actually embed members of that advisory council with our product managers on specific, uh, products, especially ones where we are m- have an even more heightened concern. Take anything related to facial matching, et cetera, and they actually help us, uh, upfront see around corners and design better products.

Carlos González de Villaumbrosia 38:47 And what, what I've seen is, um, maybe this is a different name, customer advisory boards or, like, some sort of, like, groups of people who are helping, not just with feedback, but more involved in the actual product development. Um, but I w- the, the piece that I find unique in what you are saying here is, is the, the ethics and ... the ethics piece. Uh, when you mention you, you involve activist, like, the, the ... I, I kind of get nervous, right? As a founder, it's like, okay, why would you want to have activists inside your own organization? Like, what is the trade-off of that? Are they going to slow down your, your own development cycle? Like, I, I see the benefits, but I also, uh, wonder, like, where do you define the, the middle ground here?

Jeff Kunins 39:26 Yeah, no, it's ... that's, that is the magic question. In fact, everything about who we are is about ... And it's one of the things back to being founder led. You know, Rick, in my opinion, is, is one of the most inspiring and, uh, wonderful examples of a, uh, brilliant innovator who is also relentlessly pragmatic about just trying to drive forward progress and, uh, is always trying to embody, how do we not shy away, uh, from hard, divisive, complicated issues? Because if we all shy away from those issues, then progress never happens. Uh, on the flip side, we also know that we don't have the luxury of any one of us saying, "This is my view on this issue, and we are gonna set policy," because that backfires and doesn't work. It's not our job to set policy or enforce policy. It is our job to build products that, that make it easier and easier for, uh, for police departments and everybody else to, uh, thoughtfully choose the policies and approaches that are right for them and their communities, uh, and then to be able to have, you know, auditability, transparency, uh, uh, uh, and the right knobs and dials that make sense for them to make their jobs easier, uh, uh, in ultimately helping everyone get home safe.

Carlos González de Villaumbrosia 40:46 Jeff, thank you for going deep with me on this session around products that combine software and hardware, are being used by the government, and they have a multibillion-dollar scale. I think this was, like, the magic trifecta here.

Jeff Kunins 41:01 Absolutely. Thank you so much for the time. [outro music]