← Invest Like the Best with Patrick O'Shaughnessy![Ray Ozzie - The Future of Intelligent Machines - [Invest Like the Best, EP.391] — Invest Like the Best with Patrick O'Shaughnessy cover](https://megaphone.imgix.net/podcasts/a7f4e30c-7f67-11ef-8f08-973682cfa508/image/89402462b242e238152be57d37d17013.png?ixlib=rails-4.3.1&max-w=3000&max-h=3000&fit=crop&auto=format,compress)
Invest Like the Best with Patrick O'Shaughnessy
Ray Ozzie - The Future of Intelligent Machines - [Invest Like the Best, EP.391]
01:34:28
Loading summary
Transcript47 lines
- [00:00]
Patrick O'Shaughnessy
Two fun facts about our newest sponsorship partner, Ramp. First, they are the fastest growing fintech company in history, reaching a level of revenue in five years that I can't quote exactly but is eyebrow raising. Second, they are backed by more of my favorite past guests, at least 16 of them when I counted, than probably any other company that I'm aware of. A list that includes Ravi Gupta at Sequoia, Josh Kushner at Thrive, Keith Raboy at Founders Fund and Coastal Ventures, Patrick and John Collison, Michael Ovitz, Brad Gerstner. The list goes on and on. These facts demand the question why? Having been personally obsessed with the great businesses through history, one clear lesson is that the best of them are run by disciplined operators. These operators manage costs with incredible detail and they are constantly thinking about how they can reinvest every dollar and every hour back into their business. This is RAMP's mission to help companies manage their spend in a way that reduces expenses and frees up time for teams to work on more valuable projects. First on expenses. The average American business has a profit margin of 7.7%. This means saving 1% on costs is the equivalent of making 13% more revenue. The average ramp customer is able to save 5% on their expenses each year. Of course, every entrepreneur is looking for ways to grow revenue by 50%. They should just as seriously seek to save 5% on their expenses. Second, on time. Unnecessary complexity is why most finance teams spend 80% of their time doing operational work and only about 20% of their time on strategic work. Ramp makes spend management very simple by handling your company's expenses, travel, bill payments, vendor relationships, and even accounting. It's notable that some of the best in class businesses today, companies like Airbnb, Anduril and Shopify and investors like Sequoia Capital and Vista Equity are all using Ramp to manage their spend. They use it to spend less. They use it to automate tedious financial processes, and they use it to reinvest save dollars and hours into growth at both Colossus and Positive Sum. My businesses, We've used Ramp for years now for these exact reasons. Go to ramp.cominvest to sign up for free and get a $250 welcome bonus. That's ramp.com invest. This episode is brought to you by Tigis, where we're changing the game in investment research. Step away from outdated, inefficient methods and into the future. With our platform proudly hosting over 100,000 transcripts, with over 25,000 transcripts added just this year alone, what sets Tigis apart. It's not just the sheer volume, it's the unmatched speed at which our library expands consistently outstripping competitors. Our platform grows eight times faster and adds twice as much monthly content as our competitors, putting us at the forefront of the industry. Our collection is investor led, ensuring unparalleled quality and giving you access to questions and topics investors care most about. Plus, with 75% of private market transcripts available exclusively on Tigis, we offer insights you can't find elsewhere. Forget the traditional way of doing things with tigis, you have the most comprehensive, insightful and rapidly growing transcript library at your fingertips. See the difference that a vast quality driven transcript library makes? Unlock your free trial@ticus.com Patrick hello and welcome everyone. I'm Patrick O'Shaughnessy and this is Invest like the Best. This show is an open ended exploration of markets, ideas, stories and strategies that will help you better invest both your time and your money. Invest like the Best is part of the Colossus family of podcasts and you can access all our podcasts including edited transcripts, show notes and other resources to keep learning@joincolossus.com Patrick O'Shaughnessy is the CEO of Positive Sum. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of Positive Sum. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of Positive Sum may maintain positions in the securities discussed in this podcast. To learn more, visit PSC UM VC My guest today is Ray Ozzie, one of the great technologists, software developers and entrepreneurs of our time. Ray is perhaps best known as the creator of Lotus Notes, a collaboration tool that revolutionized business communication in the 1990s. He later succeeded Bill Gates as Chief Software Architect at Microsoft, where he played a key part in the development of Azure, Microsoft's cloud computing platform. Ray's work has earned him numerous accolades, including induction in the Computer History Museum hall of Fellows and the National Academy of Engineering. Throughout his career, Ray has been at the forefront of technology innovation and paradigm shifts, founding multiple companies including Iris Associates, Groove Networks, and most recently, Blue's Wireless, which focuses on connectivity in the physical world. His insights on cloud computing, collaboration tools, and the future of technology have shaped the industry for decades. In our conversation, we explore Ray's journey through the evolving landscape of software development, his perspectives on the current state of technology, and his vision for the future of connectivity and collaboration. Please enjoy this fascinating discussion with Ray Ozzie okay Ray, we were just about to Go down a story rabbit hole. And I stopped you. But I'm going to reignite the story. Bring us to the time you were in an air raid shelter, texting with your wife.
- [05:11]
Ray Ozzie
Hello, Patrick. Yeah, this was just a few months ago. I was in Ukraine, and actually probably on a relative basis, the safest place I could be in Ukraine, which is Kiev. And I was there with a few volunteers from Safecast, which is a nonprofit I'm involved in. And I was actually also there with my son, who works for my company, Blues. And it was the middle of the night. We're sitting there, I'm alternating back and forth between telegram and an air alert app and messages on my phone. And my wife is like, I don't understand why you had to be there. Why did you have to go, this isn't going to end well. And it was relatively peaceful, but there were air raid sirens on the outside. From what I could see, there was nothing happening in my immediate vicinity, but I was just so immersed in what was going on. We read about what's happening in Ukraine from afar, but it's a bit different when you're there, when these things are happening around you, when the infrastructure is being attacked, when the power is going out and you're hearing stories from everyone that you meet about their families, about their extended families, about their friends, it's just intense.
- [06:38]
Patrick O'Shaughnessy
You've been really involved there for a while in a story that goes back to something you started back in 2011. But before we go back to 2011 and get into the technology story, maybe you could just give us a taste of what I would call the wild evolutions of the use of technology in warfare that you've seen in Ukraine, both from an offensive and defensive perspective. Because again, you hear dispatches in the US At a distance and don't pay close attention. You've been up close and personal a lot. Talk about the role of technology and what the cutting edge feels like in that very specific situation.
- [07:16]
Ray Ozzie
There's a lot going on over there, and if you take the time and interest to read, essentially, even though it's a horrible situation right now, it is essentially a proving ground for a term that I will refer to repeatedly, which is intelligent machines, intelligent cyber physical systems, the use of AI and machine learning as it pertains to machines, low cost drone manufacturing. The reason I was over there was because of radiation monitoring and the use of technology that we're working on. Throughout Ukraine, however, technology is being used both for offensive and defensive uses. I think everyone's pretty familiar with How Starlink has been essential to getting network connectivity in the field, both among the people and among the machines. But there have been small startups that have formed that have covered a variety of technologies, one of them being the use of decentralized audio monitoring, whether through dedicated devices or even one startup has done it through phones, cheap Android phones that they put out there and they listen. They use machine learning to train them to listen for the specific sounds of certain drones and certain missiles and they correlate the trajectory and project the most likely impact points. A combination of bots and people publish on Telegram where these things might be going. They don't really buy the rules of engagement, they don't really talk about the impact or not because you don't really know who's watching it, whether it's friend or foe who are monitoring these channels. But it's really a very intense environment and they're also having to deal with energy related issues. The power infrastructure is what's under attack the most right now for the broadest part of Ukraine. And so people are having to deal with systems that are tethered to power going out from time to time. Interestingly, at the current time the cellular network is pretty reliable I suspect because it's in both sides interest to keep it up and running. But that's also a potential vulnerability.
- [09:46]
Patrick O'Shaughnessy
When you think about the role of the technology that you've developed, maybe you can just explain what it is. And I'm fascinated by the cell phone example where you create this, I'll call it like a mesh network of signal and processing and then action that happens on the back end of that processing. Maybe just explain the nature of this mesh network idea or you call them intelligent machines. I think about it when we last talked about it as this incredible system that is nothing but the collection of very simple individual components. But it's the combined mesh network again I'll call it. That's so fascinating and I'd love you to just explain literally what the technology that you've developed there is and does.
- [10:26]
Ray Ozzie
Sure, Patrick, in a minute I'll go back to the origin of my company and some of the things that we've developed. But essentially my company blues its core business is essentially enabling anybody who has a physical product to turn that into a data driven intelligence service very, very easily and quickly so that they can think about the product that they're trying to develop and not the difficulties and challenges of the technology, backhauling data and so on. And one of the devices that we have worked with a Nonprofit Safecast to build is called Radnote, and it's essentially a standalone autonomous radiation detector. A Geiger counter wrapped in some communication and power management gear. And essentially it's super easy to install. The person who installs it just turns on the power and mounts it on a pole or a tree somewhere. And it's solar powered. It is disconnected from the power infrastructure. It's both solar powered and it's got a primary battery in case for some reason, the person mounting it is under duress and they don't have time to site it the right way. It'll last up to 10 years on a primary battery at a lower sampling rate. But essentially it measures periodically the radiation levels. It reports those radiation levels to the cloud. The newer generation of it that we're working on, if it can't get at the cellular network which it uses for backhaul, it will use satellite to backhaul at a lower frequency so that there's less data. The device dynamically adapts its sampling rate based on the amount of available energy and the communications, radio access technology. In the Ukraine, we're working with a nonprofit over there who had approached Safecast to get these installed. The reason it's essential both to the citizens and to the government is that the power infrastructure is under attack. And there is an extensive radiation monitoring network throughout Europe, throughout the world. But when the power infrastructure is taken out, most of those things go offline. And having a backup is very important in the Ukraine. I'm sure many of your listeners understand that Ukraine is where Chernobyl is, but it's also the home to Zaporizhzhia, which is the largest nuclear plant in the Ukraine. And it has been taken by the Russians, and it's being used by Russia as a launching point for missiles and drones, because they know that no sane military would attack back a nuclear power plant. And so the citizens are fairly worried whether it's will Chernobyl come under attack again? It was one of the earliest things that was taken, then it was taken back by Ukraine. If the power infrastructure is damaged around Zaporizhzhia, will it cause the spent fuel cooling systems to fail? Will there be radiation leaks, or will there be a nuclear device detonated? People are just concerned. And so the way that the RAD note works, this device is that because it is solar powered, it moderates how often it samples or how often it uploads. So let's say that perhaps by default, it might upload once a day or once every 12 hours. A large number of Samples. Well, if one of the devices happens to report a higher radiation level, it will increase the tempo of its uploading and it will notify the devices in the geographic region surrounding it to also increase their sampling rates to give the most relevant data, balancing the energy usage and not. So the people in the academic side call this swarm intelligence. It's a fancy world. You could think of it as collaborative machine intelligence, but it's machines operating in networks among themselves to do intelligent things. And this is a very specific application of that type of thing. But the reason that I started Blues is because I believe that ultimately all physical devices are going to become intelligent machines where that intelligence is in some mix on the devices in the cloud, there's kind of a fusion between the devices in cloud and the devices and devices. And how that fusion happens varies based on the cost parameters of the device, what the use case is, and so on.
- [15:26]
Patrick O'Shaughnessy
I want to talk a ton about machine swarm intelligence and all the different applications that might unlock in the future. I think it's the most fascinating topic, but I always love origin stories, and I'd love to hear how this idea first came to you back in Japan in 2011. Could you tell us that story?
- [15:42]
Ray Ozzie
Absolutely. Well, I had just left Microsoft at the beginning of 2011. I had a little bit of time in my hands. And what happened in March of 2011 is something many people remember. There was a tsunami that hit Japan and that took out Fukushima nuclear plants, and a few of them melted down. People were terrified in Japan. People around the world were worried. After Chernobyl, there were plumes that went over many parts of Europe. When Fukushima happened, clearly there was a plume that radiation was released. However, the location of that plume, where it flowed because of the prevailing winds, the degree to which debris was still pouring out of the reactors themselves, it was all a mystery because tepco, the power company, and the Japanese government, frankly, had their hands full dealing with the event. Their top priority was not reporting data so that people around the world could see it. And the government was beginning to do relocations of people from place to place. And this just made people more nervous. Are we being relocated to the right place? Who should be relocated? Is it going to reach Tokyo? These were just all questions people had. And so I believe probably within three or four days of the meltdowns, whether it was by email or messages or other mechanisms, many people in tech tried to say, what can we do to help? And so I think it was the first week of April, which was a couple weeks after the event about 30 of us flew to Tokyo and just met at Digital Garage in Tokyo just to see what to talk about. What could we do the task, just within a day or two, it became clear that the most useful thing that we could do would be to find a way to measure radiation extensively on the ground and to publish that on the net. There were a couple of people who were scientists who were competent enough to understand the mechanisms by which the body absorbs or interacts with radiation. But that wasn't our primary job. We thought if we could competently measure the radiation and publish it, then people who were experts could look at it and make their own assessments. And so we very quickly, within a few days, put together a prototype with a laptop and a commercial Geiger counter, some GPS hardware that we got at Akihabara, strapped it onto a car and just started driving around Japan into the exclusion zone as close as we could and very rapidly began to generate a very high quality map of the various radiation levels for the technologists who are listening. Essentially the nature of the device was that it would measure every five seconds and correlate load initially onto an SD card, a log that would log, correlate time, GPS location and radiation level gamma. And if you go to safecast.org and click Maps and explore, what you'll see is that safecast today is the world's largest open radiation data set that exists certainly every road in Japan, but also many, many, many other places globally. And there's a time axis also, because essentially over the many years that we've been measuring radiation, we've measured the same place many places many times. And so you can actually look at the levels and see where they've decreased, where the radioactivity has gone into the soil and so on. But this ended up being a very, very useful data set. This isn't just data for data sake. You can see if you look at the maps today, you can see where the plume went. It went slightly northwest of the plants. You can actually see that unfortunately, some people were probably relocated from places that were more safe to places that were less safe. If we had known at the time, in the days following it, where that radiation was going now, the levels were not life threatening. The most life threatening aspect of Fukushima ended up being the worry and the concern and the stress that it put on people and the way that it broke up families, because unusual culturally, but the people who were from the region were looked down upon differently by people who were not from the regions that were more heavily irradiated. People Lost their jobs. And so there were many, many, many impacts. But stepping way back, what that experience taught me was the importance of environmental data. The importance of having baseline environmental data when there is no issue. The importance of using commercial, off the shelf, inexpensive sensors to instrument the world so that we can know more about our environment, so we can be more in touch with our environment, so we can understand what is anomalous. If you go back to Flint, Michigan and look at what happened with water quality there, who would have thought that probably, I mean, it was heavy metal sensors at the time were not that inexpensive. But you have to ask yourself the question, if the water quality infrastructure, if the water distribution infrastructure had been monitored, would we have discovered certain things earlier? Streams today, the soil, the air quality, even sound levels within urban areas. There's really no reason at this point in history anymore why we shouldn't be broadly monitoring the earth and listening to what it has to say to us.
- [22:18]
Patrick O'Shaughnessy
One of the investment ideas that we've always loved is the notion of products that bring things from static to streaming. And obviously that's literally what all of this is an example of and what Blues does. Can you talk about the technical challenge or challenges that you encountered, starting back in Japan through to the present day, of making a machine intelligent? Because it seems obvious, like if you could just snap your fingers and have every machine capable of taking readings, streaming that data up, that there's tons of applications where it would be valuable. But the world isn't wired up that way, so there's some sort of friction. Maybe start with why it's an interesting technology challenge to enable this, and then we'll get to what it might offer us in the world if we could connect all these machines up and make them smart.
- [23:04]
Ray Ozzie
Absolutely. Well, when I say intelligent machines, there's a lot that is packed inside. Yeah, yeah. We tend now, because everyone's talking about AI, LLMs and so on, we tend to focus on the word intelligent. But in reality it means, as I said, number one, the fusion between the cloud and the machine. It is very difficult to imagine these days. And moving forward, a standalone machine, pretty much everything now takes advantage of the fact that we have got ubiquitous networking and so connecting. Intelligent machines, of course, means connected machines, machines that generate data, machines that are secure so that they are not subject to attack and take over, machines that as appropriate, encode their data confidentially, machines that have strong identities so that we understand the provenance of the data. And of course, yes, finding the right balance of putting edge inference into the machine itself or Potentially uploading the data raw and doing the inference up in the cloud. Sometimes it's appropriate to do detection and alerts on the device, sometimes not. Intelligent machines also have to manage their energy much as we do. If it's tethered, if it's plugged in, it's got an ample supply of energy. If it's remote in the environment, it's going to have to use some sort of energy harvesting. There is solar, of course, and wind, but there's also harvesting new technologies for harvesting wireless, grabbing broadcast waves over the air and grabbing a little bit of energy off those waves and storing, using charge pumps to store it. There are all sorts of new ways of doing this. But in essence, the challenge that I started seeing at Fukushima was that the initial devices that we used were essentially battery powered for short term drives in cars. The most we were ever measuring was four hours, six hours, something like that. Batteries are fine. They were standalone in that they recorded to SD cards. It was a fairly straightforward use case in the initial generation of devices. The second generation of devices were a little bit different. We had the opportunity to get deeply into the exclusion zone, where the radiation levels were extremely high. The way that we got in there was interesting. The residents of Japan who had been relocated are afforded the opportunity to visit their homes once a year. And so we were able to get equipment deep into the exclusion zone. However, at the time we were first deploying, there was no power. There was no cellular actually at the first time. And so backhauling the data was extremely challenging, as was power management. Now, this group of volunteers that were working on these devices, it was a mix of hardware hackers, software hackers, some people who had cloud expertise at the time, many people who were great community people who reached out to the residents and so on. But I would say it was mostly the number of professionals, deep professionals in software, in firmware and hardware, was less than the number of well intentioned amateurs. That group, while passionate, had a really difficult time getting this autonomous solar powered thing going. It was a little bit more difficult of a challenge both at the hardware and firmware level. So I thought, how hard can this be? And I had started back in the 70s as a double E and I learned software over time. I thought, how hard can this be? So I got my hands dirty and.
- [27:15]
Patrick O'Shaughnessy
Turns out it was hard.
- [27:16]
Ray Ozzie
Yeah, well, yeah, I learned why it could be hard. I mean, to a professional software engineer, it was fairly straightforward. I mean, it's just building state machines, understanding how much energy was in the battery, what the Charge rates are going to be looking, doing roll moving averages of charge rates over periods of time. So you could understand if you were an extended cloudy period, cranking down the sampling rates and so on. But it was also very challenging building low power hardware. I had never developed low power hardware and this is microamp level, not nano amp, but certainly not milliamp level draws building the entire system so that it would conserve power. And furthermore it was building an enclosure that would withstand the elements and survive with one visit per year. We only had one opportunity to visit these devices. So between the radio access technology challenge, given that there was no cellular, and the energy management and the physical, it was just tough. And I learned by trying to build this thing over many months that it shouldn't be this hard. I've been in technology for many years and things really only take off when they are made easy enough that essentially people can focus on their use cases instead of all the technology elements. In this particular case, the first device we built was called the Solarcast. It would measure radiation and air quality level and transport backhaul that data over Lora. And then up in the hills surrounding the exclusion zone, we built little Lora gateways. Lora is a, I believe the acronym is from long range, but it's a radio technology, very low bandwidth that operates in the 900 MHz ISM band. They call it the ISM band in the U.S. but industrial, scientific and medical band, but it's an unlicensed band WI fi where anybody can use that band, but depending on the encoding, it can be used for very long range or very deep penetration. They're both very useful. And Lora is a proprietary technology from a company called Semtech that wraps a little modem that encodes and decodes that technology for use by little IoT apps. In our case at Safecast, we use Arduino as the operating environment and IDE for the mcu. And we would package up and compress and send that data in little 200 byte packets up to the hills where they then went into a Raspberry PI gateway and were using Balena operating in a distributed operating environment to manage the gateways and to then bring it to our cloud based service. And then over time when the cellular networks came back up, we actually hadn't fully designed the device yet. So we made them both Lora and cellular capable, those extreme devices working. There are still many of these devices that if you go to the Safecast map, you'll see are still broadcasting those radiation levels. But I felt that it was just too difficult and I had not even gotten into the security aspects of it. The devices that we developed for safecast, it was all out in the open. The identity of the device is very crude. It was just simple identifiers that we assigned into them. It was not professional grade. I had just come from Microsoft, and whether it was at IBM before that or Microsoft, I had many friends and colleagues who built systems for large enterprises. I built a system that called Azure.
- [31:27]
Patrick O'Shaughnessy
That some people may have heard of.
- [31:28]
Ray Ozzie
That people may have heard of, that a number of those customers were using for their cloud cloud services. I called them up and I said, aren't you guys connecting devices, IoT devices to the cloud, whether AWS or Azure? And what I found was a litany of failed projects. Enterprises who had started prototyping, getting physical devices connected to the cloud, generally using WI Fi. And the prototypes were very easy to build. But over time they learned of the challenges that wifi doesn't work for commercial products. And then when they tried to use cellular, they had similar challenges to what we ran into. Dealing with carriers to get the appropriate data plans was not right. Getting devices certified was difficult. So I started a company, Blues, to try to make that easier.
- [32:25]
Patrick O'Shaughnessy
Maybe just talk us through the stack here because we've referenced a lot of layers of it. You've talked about Lora, you've talked about cellular, talked about wifi, you've talked about satellite. It seems like such a straightforward task to just have a thing on a piece of hardware and send some data up and capture that data. Obviously Blues is entirely predicated on making it really easy, but what is that stack of hierarchy that you had to build for v1 of the device and then we'll talk about how it could be used.
- [32:52]
Ray Ozzie
So success, let's just keep in mind what success is. Success at the highest level is to enable companies of any size either to build physical products from scratch or to transform existing physical products into data driven intelligence services. People want to understand the nature of devices generally for two reasons. Either to do preventative maintenance, meaning understand whether those things are operating correctly, anticipate their failures, and so on, or to improve the customer experience. There's something about the product whose experience could be better if you service enable it. The customer environment is that we want people to be able to focus on their use case. They are professionals in some domain they might manufacture refrigerators, manufacture generators and so on, and they want to focus on that use case. They just want to get the data to the cloud. At the lowest level there is Hardware because it's a physical device. And so today, because the hardware industry tends to have evolved as a component industry, the components that are involved are things like modems. Modem is a modulator demodulator. It's a thing that encodes data for transmission either on a wire or on a wireless network. In order to power a modem, it takes energy, so there is a power supply involved. If you are going to send data to somewhere, you need a protocol stack to encode the data that goes over the wire or over the air. And depending on the type of radio, those protocols may vary. Sending something over satellite, you want it to be very small and you want it to be packet oriented. There is a protocol stack. If you want to do it securely, you have to. In modern cloud based systems, you want to have digital identity in some way, shape or form. So how are you going to get digital certificates loaded onto those devices? This is an extremely challenging thing, especially if you're not familiar with that domain. People are very familiar with certificates and digital identity. In web browsers. Everyone uses SSL or TLS to secure connections. How are you going to get those, the root certificates and certificates in general digital identities onto devices? So this is a challenge. Putting hardware secure elements onto a device is a challenge. So anyway, these things stack up and then when it comes time to actually have a service that transmits. Cellular is the most powerful technology in my opinion. That's out there because it's one of the few technologies where in order to do cellular, you've already got a secure element on the device to do digital identity. We call them sims, SIM cards or SIM chips. You can put cellular into a device. And most everywhere on earth that there are people. Not so much rural, there may be cellular coverage problems, but most places where there are people, there is some sort of cellular coverage. And you don't have to configure it like you have to do with a wifi network. You don't have to build your own tower network like you do with something like Lora or WiFi, where you have to install your own gateways. It's already done for us. However, in order to use cellular, you have to get your device certified by carriers so that it doesn't abuse, whether intentionally or mistakenly, the shared resource of the cellular network. So all of these things stack up. And so my concept in doing blues was that we would create a fused hardware software service offering. The hardware offering is called the notecard, the service offering is called the note hub. But essentially the hardware manufacturer would design their product, they would put the note card in it and we would offer note cards with different radio access technologies, different types of cellular GSM, LTE, you know, 2G, 3G, 4G, 5G, those note cards, or Lora or a satellite. The little note card would have on it the sim, the secure element. It would have flash memory so that it could do buffering of data coming from the host mcu. It would be protocol agile so that the person doing the host MCU wouldn't have to worry about either security or how to encode the bits and bytes to get them efficiently to the cloud and that essentially it would just work so that the person could worry about their use case and worry about the cloud service that they're trying to do. And we would take care of everything in the middle.
- [38:24]
Patrick O'Shaughnessy
Your whole career, if I think about it in the biggest possible sense, seems to be about enabling connectivity and collaboration in some very, very broad sense. If you think about Lotus Notes, if you think about Azure, if you think about Talko, if you think about Groove networks, everything sort of has this common theme of using technology to hook things or people up and help them work together more efficiently. We're going to go back to Azure and all those other exciting chapters and the lessons you've learned watching technology paradigm shifts, which is something I think you're in the top.0001% of people on the planet to explain and riff on. But I want you to imagine first, I'll call it the five year hence future or something like that, where because of what you're up to, the world of machines can be more connected, gathering more data, using more data. I'm of course interested in how you think this might impact artificial intelligence, whether or not maybe this is a source of data to train more models on and so forth. But just imagine the intermediate term future for us and why the promise of connected swarm machine intelligence is so interesting. And maybe now's the time to get into some of your favorite early examples that you've seen of things that get enabled. You're a builder of enabling technologies, so I'm always interested to know what is being enabled at the frontier that you're seeing because of a solution like this.
- [39:45]
Ray Ozzie
Let me just step back and lay out a little bit of background here. What excites me as a technologist that's been in the industry for a long time is not so much just technology for technology's sake or specific use cases. I get really excited by broad, broad, broad use of a technology by the masses. And when I say the masses. What people generally think of is consumers. But what I really mean is that there are an extremely large number of small to medium sized enterprises globally. If you think about small businesses, even just in manufacturing physical products by unit count, small to medium sized enterprises are constitute about 99.9% of all manufacturing firms. My revenue, it's probably 35 to 40%. Even earlier in my career. If you look at the world of tech, there's big tech and there is the rest, and there are a lot of companies in the rest. And if you look at just anything like spreadsheets or email or any kind of communication tools, there was a phase of computing, just simply computing. At first, the largest corporations were able to afford getting mainframe computers. And then over time, companies like DEC and Data General and Prime brought computing down into the super mini computer and minicomputer size realm, which enabled more and more companies of different sizes to get access to it. Eventually, the personal computer revolution helped everyone. And the spreadsheet especially became an empowering tool by which everyone enjoyed the benefits of that technology. With regard to physical devices, we have been in a situation since probably the early 2000s where the largest companies, the automotives and John Deere, began experimenting with cyber physical systems, with putting communications technology and IoT technology into physical devices. In cars. It began with OnStar. Some of your listeners may remember this. You press a button and suddenly people would know where you are. And this was a very fancy technology in the early days. But increasingly, generation after generation, it turned into more. It became integrated with the maintenance value proposition of cars. It became integrated with the entertainment value proposition of cars. Many people aren't even aware of this, but many cars have two modems in them now. One is paid for, completely subsidized by auto manufacturers. And they get engine performance data. They monitor the performance of the vehicle. It's not supposed to go into that realm, but they gather that data and they sell it to the dealers and they have part of their value proposition. And their business model has changed over time as independent from the modem that gives you a wifi hotspot or that refreshes the maps. John Deere, they've gotten a lot of negative press in the past years because building essentially closed systems that have caused many farmers to rebel because it's really transformed the nature of farming. But in essence, they were innovators in instrumenting farm machinery with location, with things related to the preventative maintenance of the vehicles. It's really transformed in many ways the production of food. And those are big companies that could afford to do this. They have gargantuan R and D budgets like many Silicon Valley companies. They can hire the best of the best. They can hire the best security people, the best people who understand power management, radio access technologies and so on. But my question from where I've come from is how does this technology evolve to allow all small to medium sized enterprises who manufacture physical goods to enjoy the benefits of building intelligent machines and intelligent services? Again, this happens in every sector. All of us when the web first came out, hard to remember, but we ran email servers within our small businesses. We ran little payroll systems and accounting systems on servers within our departments. Eventually, because of the democratization of technology, Those things became SaaS services, they became part of the cloud. And now no one would think of doing those things. If you were a small to medium sized enterprise, no one would host their own systems because you can't hire the security people. And why would you, why would you host your own email system as an SME when you are a company building refrigerators, generators, propane tanks? What you care about is that value proposition and how your systems serve your customers. And so essentially what these folks just want is a data pump. They want the data to go from the devices to the cloud. And increasingly they are also wanting to do things like edge inference. They are wanting to detect anomalies either by putting some intelligence onto the device or by getting the data in the cloud and doing that inference up there. That's what we would like to help them do.
- [46:04]
Patrick O'Shaughnessy
You've seen several of the most interesting major technology disruptions and been a part of creating the solutions. You wrote an incredibly famous memo called Internet Services Disruption. And I think it was 2005, something like that, which predated your creation of Azure at Microsoft. Can you just summarize? That was maybe the most important transition other than the one we're going through right now with AI in this whole history. What did you see back then? And maybe this is an excuse just to talk about technology paradigm shifts in general, which I know is a topic that you're passionate about and have ushered lots of them in through history, but maybe zoom in on that Azure, what it was that you saw, why you wrote that specific memo to the senior team at Microsoft, and then what happened immediately afterwards. And then I want to use those lessons to apply to today.
- [46:51]
Ray Ozzie
I started at Microsoft in April of 2005. I had known Bill Gates, Steve Ballmer, since when I was at Software arts working on VisiCalcast. We can go back and visit that era, but I knew many, many technologists at Microsoft. And Bill and Steve essentially said, look, we've been through many shifts. We have so many things going on at the company. We're in a great cash position right now. But Bill in particular, like Andy Grove, he has this perpetual view that the paranoid view that two guys in a garage or somebody is going to discover this thing that's going to take the company down. And I was coming in, they knew I was coming in from the outside. I had a different perspective. In any company there is a danger, particularly a successful company, there's a danger of only ever seeing things incrementally as a function of the status quo. So they said, if you believe that there is a big disruptive change happening, lay it out and fix it. Lay it out and prepare us for it. Now what I had been working on at Groove gave me some opinions. Groove was the company that was bought by Microsoft that brought me there in 2005. Groove was actually a collaboration tool that was based on a peer to peer architecture. This is the Napster era. But it was also the dawn of the web era. You know, 97 was when I had founded Groove and I became very aware, both because of my enterprise customers and defense customers, that there was this hybrid thing going on where people were forming teams and communicating peer to peer directly, but people were also talking to companies in a hub and spoke manner, enterprises. I began to form this mental model of every enterprise having being a hub and doing all of their customer service in a very broad way from the cloud, from the center, and people forming teams and using the Internet to do those team infrastructures. Groove itself became part of SharePoint. That was more the team side. But where that memo came from was the fact that when I looked at Microsoft, it was, I don't think the word stuck is quite the right word, but they were centered on a previous era in all of their product lines. Microsoft Office was built for a box era. The box was a PC. It was software you put on the box to deliver its value proposition standalone. Windows Server was a box value proposition. You put the box in a department or you put a bunch of the boxes into a managed service provider and you served hundreds or thousands of users from those boxes. Even Xbox was a box that served gaming in the home. It was a very box mentality. In the meantime, here is a company, Google, who's at the bleeding edge of a completely new era. They were born into a services era. And I believe you have to take on the mentality of the era and then try to back Solve what that future environment is going to be like as opposed to trying to incrementally get there. So when I wrote that memo, I looked at each business unit and I said, what would a that future world look like? What does it look like for a connected Xbox world? So that means Xbox Live. What does it mean for a connected productivity world? That would be what Today we call Office365. At the time, we called it Office Live. What would a connected computation and storage operating environment be like? Well, that is what eventually became Azure. And that memo was a way of me communicating that vision to the company in a way that was culturally compatible with the way that that culture receives disruptions. Bill had written several different memos at the time. When I came in, Bill had asked me to assume his role because he was on his way out to do work on the foundation. So he gave me a lot of coaching on that when I took on his role as chief software architect. So that was kind of the kickoff of that era.
- [51:40]
Patrick O'Shaughnessy
What do you remember most learning from him and how he operated? Learning in that handoff and then working with him afterwards. Is there anything portable to others that you think if others just adopted this mentality or way of working, they would do a better job than they're doing now? Because obviously Bill is famous for just incredible productivity and paranoia, I guess, is the word you used. Is there anything else that you remember from working with him that would interest or surprise people?
- [52:07]
Ray Ozzie
Do you think Bill is a very unique individual? We've talked about a number of different aspects that he has and that the combination of Bill and Steve had over time. Some of the patterns that you would recognize are Bill and Steve are a hacker hustler pair. They did very, very well in that realm. On the other hand, Bill himself was very unique in the industry in the early days when I was growing up, because he was one of the few who understood all three of technology. The market, the technology use case, and business, all three of those in one individual. Another aspect of Bill that in recent weeks has gotten a lot of airplay is founder mode. He's very detail oriented and he had no fear of diving into the deep detail of technology. He would create groups in the company, would, over a very long period of time, create forcing functions by scheduling meetings with Bill where he would grill people with questions at every level of the stack. Steve did the exact same thing on the business side. Steve, when he held global business reviews, understood that business at levels of details better than pretty much anybody. In many cases, better than the business leaders themselves. It's quite a unique organization and I've competed very, very hard with Microsoft. They were the archenemy when I was doing Lotus Notes, even though I knew the people very, very well. Exchange SharePoint was the direct competitor of what I had done at Lotus Notes. But I'm extremely pleased to see what Satya and the organization, the senior leadership, has done to fulfill the promise of what we had dreamt about in that era. When I had written that memo, what I did was kick off that era. What has become of it really is so many people deserve so much credit for what that organization has become.
- [54:19]
Patrick O'Shaughnessy
I went and read the whole thing yesterday and I encourage people to read it. The main reason that it was surprising to me how incredibly detailed it was. It's quite long and it goes segment by segment. Not just to say at a high level, here's an opportunity that we should think about, but very, very detailed specifics around what's changing and what might be the zone of opportunity for each part. As you laid out of Microsoft, if you could imagine writing a similar memo today just to the world, to all these businesses, the whole set of companies that might use and deploy technology to make their product or make their product better, what do you think the key points would be that you would point out that everyone needs to be mindful of? You sat there in the cloud transition in the early days and then created Azure. Great businesses of all time, along with aws. And now we seem to have a new set of enabling technologies, whether it's blues to make machines intelligent, whether it's LLMs to make everything embedded with reasoning capabilities or better search capabilities. If you were writing that memo to the world today, what do you think its key points might be?
- [55:24]
Ray Ozzie
I think I would open by saying what the key enabling technologies are. Certainly today, one of the most significant is AI, AI and ML. Much as cloud was an enabling technology there. Another enabling technology that is, I think, too easily dismissed is something that Chris Anderson referred to as the peace dividend from the smartphone wars, which is the inexpensive, ubiquitous semiconductor technology. If you can imagine it in hardware now you can build it, whereas that was only really possible in software before. So I would begin by framing these things and I would say because it's very difficult for people to conceptualize this, they need to be reminded there will be a point in the not too distant future that all of these exciting enabling technologies will be commodity. AI technologies are going to be no different than relational database technologies were back in the day. They will be runtime libraries that can be built into every application solution, there will be open models that are available. And the question isn't really how are you going to make money from the technology? The question is, how is that technology going to reshape society and business? And so after the framing, I would go use case by use case and try to jump to the future and look backward and say, within this use case, how will it be fundamentally transformed? If you were born 10 years from now, 15 years from now, and all of that technology is humdrum, what would it look like? How would it transform the customer experience? How would it transform the developer experience? How would it transform business value propositions within different industries? Within Microsoft, it was division by division because those were the relevant use cases. What's exciting to me is visiting not just software use cases, but also hardware use cases in mind. You've got manufacturing, transportation, logistics, the environment. Sustainability is something that is weaving itself through so many different sectors, energy and sustainability, sector by sector. How have these new technologies, how will AI and the semiconductor revolution and the revolution in wireless technology impact each of those use cases? And then if you can envision those use cases, you will immediately see the opportunities as to what role you personally or your company can have in addressing those use cases. The reason I did it in the memo was because people have a much easier time understanding how something can be actionable. The more you localize it, the more you can contextualize it in terms that people live on a day to day basis.
- [58:48]
Patrick O'Shaughnessy
Could you give us one example of what would excite you the most? If I could drop a note card into every device in X fleet of machines in the world, is there one that comes to mind that would just be the most exciting for you to just snap your fingers and have that be true and have data be streaming from that fleet of devices up to the note hub and to the cloud.
- [59:10]
Ray Ozzie
It's a really difficult question to answer because from where I sit, the reason we're doing it is because I believe the number of use cases is inherently broad, just like software itself. And the things that excite me the most emotionally, perhaps as a citizen, are the environment and sustainability related use cases. I mean, maybe I should just summarize a little bit of how people are using it. We have a company called True Manufacturing, a large customer who's a private manufacturing business in the Midwest, in Missouri, founded in 1940. They build high end commercial refrigeration, ice making and kitchen equipment. They need embedded intelligence in their equipment both for preventative maintenance and to improve the customer experience. So they used to Just ship these refrigerators, whether it's to a small sandwich shop where you see one cooler with drinks in it, or it could be somebody like a Walmart or a Target, where you see rows and rows of glass windows. Those used to be shipped, as I was saying about Microsoft, like boxes. And the company itself never actually had touch with the ultimate customer. They sold to a distribution channel and the distribution channel then sells to the customer. In this new era, the manufacturer, every manufacturer, even if they have a multi tier distribution network, gets an opportunity to understand how their product is being used, what the failures are, what they aren't. In order to help the channel understand and the customer themselves how to repair them better, they can tune the energy management of those devices dynamically based on the actual in situ loads and they can give a better customer experience. In the true manufacturing case, some of those coolers are holding beer and some are holding vaccines. They can tell those customers about spoilage and give them actual data and it's transforming their business. Another one, American Crane, is a small business that makes very large cranes. Department of Energy, Department of Defense. For all sorts of sectors, they embed intelligence into the cranes. One of the ones that's most exciting to me is a company I can't name, but it's one of the world's largest automotive and truck battery manufacturers. They manufacture the batteries that go in long haul trucks and in cars. But the trucks are to me the most interesting because by embedding a battery management system and cloud backhaul into every battery, you can help the customer understand when it's best to power the truck by battery or by fuel. You can save fuel and extend the battery life by understanding the drainage patterns of those batteries. Someday we are not going to even consider the concept of somebody buying a battery or a propane tank. If you're on the east coast, many people use propane tanks or oil tanks. The concept of having one of those things that is not cloud connected and doesn't have built into its value proposition, the data is going to be simply unheard of. We have one customer. Again, hard to conceptualize. It's one of the world's largest Porta Potty manufacturers. And why would you connect a Porta.
- [63:05]
Patrick O'Shaughnessy
Potty can use our imaginations.
- [63:07]
Ray Ozzie
Yeah, you know where they are. You know, if some kid has kicked it over, you know, the number of door opens and closes, you know, the using gas sensors, the relative smell, you can change the value proposition for the customers and improve the products fundamentally. And this will be simply ubiquitous healthcare. We have a number of customers on the healthcare side. Some of this started during COVID where a very innovative entrepreneur in the Caribbean, where they don't have as much hospital infrastructure built equipment that they would put for in home care that would let doctors remotely monitor a broader population. We have one customer so far, Ocean, who actually motivated us to do satellite where they have embedded these little buoys. People think of weather buoys as few and large in the ocean, but they airdrop small buoys into the ocean and they're the world's largest privately owned network now of ocean sensors. They're doing this for sustainability. They help maritime shipping optimize the loading and unloading and the traffic patterns of where they go based on really micro optimizing what the ocean patterns are like in real time at any given point in time. And for some buoys that happen to be near the coast, within several miles of the coast, they happen to automatically use cellular. For the ones that are in blue water, they automatically can use satellite. So anyway, I'm obviously super excited about this at a very high level. Again, I believe people are very excited about LLMs and chatbots and that's the center of conversation right now. When you think about the stories I was talking about Ukraine, when the stories I'm talking about are customers here I am excited about intelligent cyber physical systems in a commercial sense. Everything that you buy will be in some way, shape or form having a service based value proposition. Much as I talked about in that 2005 memo.
- [65:28]
Patrick O'Shaughnessy
I'd love to ask about timing. When you think about the history of AI and for a long time it was called machine learning. You're one of the few that says ML alongside AI because most people are just interested now in the new post ChatGPT era. But this technology has been building for a long time and then it had an explosive moment. And I think prior to ChatGPT you might easily say it's been disappointing the amount of commercial applications of machine learning, at least from the perspective of the customer. It was fueling lots of stuff behind the scenes. It seems like IoT is somewhat similar in the sense that in 2015 or something there was probably an IoT ETF. It was one of these themes that people talked about in technology. As this is coming, it's going to be ubiquitous and then maybe it was disappointing in some sense because it took longer and AI now seems to be on a similar trajectory. So these things are going in close tandem. What have you learned about that phenomenon where IoT was a big thing? Why wasn't it a bigger thing in 2015. What has changed now that makes this incredible vision of a connected world, a connected physical world that you've just laid out newly possible? I'm just fascinated by the timing of when these things take off versus when they're first talked about.
- [66:40]
Ray Ozzie
Well, maybe I'm oversimplifying, but let's start with AI. The fundamental thing is accessibility. Machine learning has really been powerful and has had great returns within certain sectors. The people where you could hire data scientists, where you have a good flow of data, where you had enough of an investment to tag that data so that you could very clearly lay out to the machine learning systems what an anomaly would look like. And there are many sectors in manufacturing and others in logistics where that data where machine learning, even in medicine, where that technology has been used. But once you see a chatbot and everyone can just talk to it and understand and get some value from it, suddenly people can imagine an impact in a form that they can understand. And once Llama was released and developers could start to see how they could integrate these things with their own systems, suddenly things take off at a rate that they just couldn't before. With Iot. Let's not forget that there is hardware involved. And many of us, as you said, it's maybe 10, 15 years ago, there was a lot of potential. And the excitement again, I believe, is accessibility. As long as it's a technology that can only be utilized Bay Area companies or companies that are extremely rich with big IT budgets, it will not fulfill its promise. Whereas once it becomes possible for most everyone to do it, suddenly it's much more relevant. I am mostly excited about what I call commercial iot. There is an entire sector where there's another, even broader disillusionment I think, that has yet to come, which is consumer IoT smart home technology. We're in a phase right now where everybody's getting smart light bulbs or everybody's hooking things up to Wi Fi, and over time, some of these things will lock in. But there is going to be a lot of disillusionment because people are connecting things just because they can, not because they should. As people move from home to home and realize the amount of equipment they've got to retrain or reset and reconfigure, some of the bloom will be off the rose there, but on the commercial side, it's really taking off. We went from an era of, I think, early trial, then a trough of disillusionment as people couldn't figure out how to backhaul now onto the General satisfaction and excitement with the potential of the technology.
- [69:45]
Patrick O'Shaughnessy
I always love that framing that Kevin Kelly has in that book. I think the book is called what Technology Wants and what Technology Clearly Wants today is data. Novel tokens to train models to push through models to inference everywhere, all the time. How are you thinking about LLMs, AI and the way to commercially embed these new technologies into your product or service? You said something about Bill Gates, which I think is true of you as well, which is not just cool new technologies and enabling technologies, but things to do with them that are valuable to customers. A commercial sense alongside a technology sense. What advice would you give people out there that are building businesses, building products that are interested in AI and LLMs and all this stuff? It's successfully commercially applying it because it seems like a common trap is watch the sports. I watched some tennis this weekend and every ad is like AI this and AI that. And it seems like it's pitching AI as the feature versus the thing that AI enables, which seems like a big mistake. Tell us how you're approaching this and thinking about this and how others might successfully do the same.
- [70:54]
Ray Ozzie
Assuming it's going to be an embedded technology. I mean, let me actually step back. There is artificial general intelligence and speculation about whether we're going to get to that point. And I'm really not talking about that right now when we're having this conversation. I'm really just talking about the technology at its current or slightly projected level of maturity. As I look at it right now, I divide things into LLMs as a human interface accelerator or supercharger, where it can scale in ways that humans cannot in ingesting large amounts of information. And there are questions that it's able to competently or semi competently with the hallucination asterisk, it can very competently answer a broad range of questions that today it's very expensive to answer or may not be answered at all. You see this in coding a lot. Some people with regard to programming think about it, writing the code, but in actual fact, it also answers quick questions about algorithms and APIs. It's an amazing UI to the world's API reference manuals that are out there. And sector by sector, this is true. Imagine AI having ingested all case law, all schematics that have been built by engineers. It is really going to be a supercharger. From a human interface perspective, I believe the greatest potential for entrepreneurs right now is to look at it sector by sector and say, what sector do you have a personal interest or connection to and how can you jump forward and envision that sector transforming fundamentally and then back solving? What are the steps going from here to there in the IOT world? The way that I'm looking at it is both from a human interface perspective and from the classic perspective, which is anomaly detection. The past way of using machine learning is to manually tag data and to generate very clean data sets and then use machine learning algorithms to do that. I'm sure that's still happening in many sectors, particularly life critical sectors. But the modern thing to do is to just gather as much data as you can. Don't even clean it, just gather data. Gather data. If you have the transport bandwidth, bring that data to the service, don't do the inference and reduce it on the device. Bring it up to the service if you can, and just record vast amounts of it and then let LLM technology, some of which has not been developed yet, but the promise is to use some of that technology with huge context windows to learn what an anomaly actually looks like. But there's also exciting technology in the human interface. Some of the stuff that we're doing haven't released yet, but we're experimenting with the concept of dashboard by query. If you have a lot of data, there are many things that you can do to help operations, tech personnel, or even the people in the business unit who are looking at IoT data to say, oh, which ones have a higher frequency of X happening or which ones haven't reported? You can create these dashboards. That's the classic way of getting value from large data sets. Another way of getting value is using analytics tools to do queries. An interesting thing that we're experimenting with is what if you could use LLMs to describe the query that you would like to do and then generate queries and then generate charts based on the problem that you're trying to solve, as opposed to being so hyper explicit? I think there are going to be many possibilities in that realm as people do more and more experimentation.
- [75:31]
Patrick O'Shaughnessy
Is there any other enabling technology we haven't talked about that you wish existed? You helped create Azure and that was a whole set of Internet services box to streaming, to use your language. Now you're creating the ability to pump data from anything and do lots with that data. LLMs are creating reasoning capability with big context windows and the reasoning engines themselves. Is there anything else that you wish you could snap into existence as an enabling technology that you've thought about or that you think might be coming that's relevant for this big picture?
- [76:02]
Ray Ozzie
There's so much that's in flight right now concurrently. I want to be super honest with you that I'm just trying to absorb all of this thing. We haven't even talked about radio access technologies, but there's as much innovation going out on the radio access technology space as there has been in the semiconductor space and as there is in the AI realm. Satellite has the potential of delivering 100% ubiquitous coverage. Where we didn't have that before, we've had cellular systems out there that handle where there's population. But now with companies like Skylo or ast, there are possibilities that cellular based backend technologies will be able to be used with new radio access technologies to deliver ubiquitous connectivity. What solutions will happen with that? With some of the technologies that have happened with 5G and have not yet been absorbed into solutions and are coming with future generations, there's mind boggling bandwidths that are possible on a point to point basis, on a line of sight basis and so on. There's just so much innovation in the area that I'm closest to right now. Where I want to see innovation is the easy button for physical products, for hardware design. Right now there is a shortage of really strong talent who understands how to design and build hardware devices. It's not a black art, it's a skill. And I really have tremendous respect for people who have learned that skill through apprenticeship and formally to build low power devices, to build very energy efficient devices. But there is still a number of spins that must happen. You must conceptualize the solution, prototype it, spin it into an initial hardware prototype, put it into use, refine it, go through that same process again. When I started in this industry, that's what software was like. We had card decks and you had to, I mean, I know that I'm showing my age here but you had to go type out the card deck, you had to go give it to the operator with jcl, you know, cards on the front and back, run the job, get it, go see if it did the solution. We're still in that era with regard to physical devices over time I believe AI and LLMs, they hold some promise in that realm. It's very difficult to see today, especially if you're one of those people with those skills. But shortening the cycle time for creation of physical products I think is really exciting. And we're not there yet.
- [79:12]
Patrick O'Shaughnessy
One of the things we're seeing is the world of hardware usually plus software, the two are pretty inextricably linked now is attracting some of the most ambitious talent in the world. Software is so mature there's so many resources for how to develop software. There's so many playbooks, it's so well understood. There's software for everything. It'll keep going, but it feels like a pretty well worn path. Whereas there's no similar set of books, playbooks, ideas, classes, et cetera for developing hardware. What do you think is the most significant difference between those two things? What do the best hardware development people? You said it's too small of a sample, but the ones that are good at it, what do they do especially? Well, that's most distinct from the great software developers? Because I would love there to be just more of these people out there and so I'm curious about them.
- [80:03]
Ray Ozzie
I actually believe the core skill is similar in some ways. It's just that they're different disciplines. The thing that I grew to respect very early in career, we don't talk about a lot these days on the software side, which is systems thinking. Let's take it into a completely different discipline, architecture. You can have architects and others in that discipline who understand very small units of design. Then there are the few systems people who can build an entire skyscraper. They understand how it all fits together and it's a fractal problem. There are some people who may understand some little part of a valve, and there are others who understand how to build a rocket and others who understand how to go to the moon. In software systems, there are people who write device drivers and who understand how to do unit testing on device drivers or on a piece of a user interface. And there are those people who really, through their skills have learned to build an operating system and a very, very large application. In hardware. There are analog engineers who really understand certain aspects of it. There are people who design secure elements and understand hardware side channel attacks. It's very difficult right now to find semiconductor people who bridge up into firmware, the firmware people who get cloud. And yet it's all holistic now. An intelligent machine has to be from the chip level, maybe using an application specific inference engine, all the way up through that cloud service. The more you can have a systems understanding, the more valuable you will become and the fewer people there are. I don't know how LLMs, which are the accumulation of the knowledge of the really deep specialists and the writings, whether it's code or writings or schematics of people operating at the systems level, I don't really know yet the capabilities that are going to happen at that level. It's going to be an exciting time.
- [82:39]
Patrick O'Shaughnessy
Going back to those two big themes that I see throughout all your thinking of connectivity and collaboration, if one of those two themes is at the thrust or at the center of a product or service that someone is building, I'm curious to know what advice you would give. Let's say I'm building a workplace collaboration tool of which there are now many, but there's probably lots of places where we're still doing single player work when we should be doing multiplayer work or something. What have you learned about what matters in products where the core job to be done is collaboration between teams? If you were teaching a seminar class or something of a group full of YC startups or something that were all doing collaboration tools, what would you tell them? Look, I've done this a bunch. Here's what matters. Here's what matters less. Don't make these mistakes. Focus on these things.
- [83:27]
Ray Ozzie
Well, technologists like to focus on technologies and platforms, and they tend to intuitively get excited by those problems. But fairly early in career this was in the Notes era. I learned that business value rises nonlinearly in proportion to domain specificity. That is, you can have an email system or a shared forum where people swap information and ultimately that platform will become a commodity. It's just a matter of time. Whereas a Salesforce automation system or a legal document management system, because of its domain specificity, is built on the same underlying technologies. There is a chat window to the right, there's a document management system underneath. There is a query mechanism to help you find everything. There's an access control subsystem. If you really want to make a difference to the world, focus on a domain. The more you can focus on the domain, the more value you will be creating. You can make a lot of money both ways, but you will make a lot of money as a platform provider over the long term. Only if you're a big player. If you're one of the few big players, everybody makes money on picks and shovels early on in a domain, but over time, the only ones who make a good living, who earn good money on commodity technologies, are the few biggest players. Whereas there will always be immense value in domain specificity.
- [85:27]
Patrick O'Shaughnessy
If you think about the explosion of technologies happening today that you're in the midst of, and compare it to other explosions that you've personally been a part of across your very storied career, how would you stack this one up against the others? This versus early days of the Internet, or cloud or mobile or PC or Pick your era and paradigm shift. What does this one feel like to you versus the others?
- [85:55]
Ray Ozzie
I have to say, number one, that I'm just super fortunate to have been born when I was and be able to be alive during this velocity of interesting problems to be solved and interesting technologies to solve them. There has never been an era like there is right now. And as far as I can see, there will never be an era like tomorrow. Meaning they're all stacking on each other. We still have not internalized the innovations of the past today and we're still building them on top of each other. Every one of the things that I mentioned is still growing. Computation is still growing. It's not growing in one cpu, it's growing in parallel and in application specific processors. Storage is still growing. Networking is growing from the wireless realm with all these different radio access technologies. Semiconductor technology itself is still growing. We've got chiplets now, we've got a clean separation between the parts of the stack in that ecosystem. AI different types of databases. We haven't even scratched the surface of LLMs. I just couldn't be more excited. And the reason I'm in this industry and engaged to the degree that I'm at is that you can imagine and then reimagine different domains based on new mixtures, new compositions of technologies that are appearing every day. And what is the new technology that's going to happen five years from now? It's fun.
- [87:55]
Patrick O'Shaughnessy
How would you encourage people to think about business strategy? M and A Putting your business hat on here as one of these paradigm shifts is unfolding Sounds like we're just in for a permanent paradigm shift. All the time, in all directions based on that answer. But the creation of enterprise value, not just product value for the customer, but enterprise value for the business. You've done this a lot of times. You've built companies, you've sold them very successfully to places like Microsoft. You've then gone and worked at these places and tried to disrupt yourself. You've seen the full life cycle here. A lot of people listening are in technology, entrepreneurship. That's what they do. They're building technology based businesses. Any word of advice on the enterprise value side of the ledger? We've talked a lot about product and technology, but you've also done the thing that I think lots of people are trying to, which is build enterprise value. Anything notable there that you would leave us with?
- [88:51]
Ray Ozzie
I'll say one thing which I think is a little different than what certain people, particularly leaving college, might hear. But I believe that you can't develop for the enterprise unless you understand the enterprise. And you can't understand it unless you live in it for a while. I strongly encourage people who are leaving college to go do a tour of duty of several years within a big company. You could say the same thing about government. You're not really going to understand the organizational dynamics or bureaucratic dynamics unless you actually live in one. You're never going to understand what a non profit is like unless you do a tour of duty inside a nonprofit. Because herding volunteers is different than hierarchical decision making and command and control within an enterprise. I think there is just immense value to be delivered to society by applying technology skills to the problems at scale in enterprise. Because of the nature of enterprise right now you can have huge impact to large enterprises at scale, even by working for a small to medium sized enterprise. I know this is a little counterintuitive, but if anybody really wants an education, go all the way back to the 30s and read the Nature of the Firm by Ron Koch. In those days, the big company was all there was and minimizing transaction costs was the only purpose of the firm. We are in a completely different era now where the firm is actually a very complex network of small entities bound by common economics and purpose. And the organizational form for many industries is no longer one big enterprise. It's lots of little companies that work together to develop an ultimate end product. I'll bet you if you look at somebody like SpaceX, they didn't develop everything themselves. They have a network of subcontractors who specialize in machining this part or doing this and that. They probably use their capital and ability to hire skills only on certain skill sets. But anyway, my recommendation is, if you're a young entrepreneur and you don't have a lot of experience, don't just gravitate to building a consumer app or a consumer device or something like that, you can learn by building anything. But they're not teaching you what are the things that are going to make the most impact on society. And probably they're not giving you the things that will be most valuable for your entire career. You learn immense things by working with people in teams, by working on units of larger products and larger entities. And you will learn how these enterprises buy things, why it's important to understand the management of complex systems, not just the creation of complex systems.
- [92:28]
Patrick O'Shaughnessy
Incredibly unique and powerful way of thinking about one's early career for sure. Ray, your career has been so fun to study your business. Obviously we're investors in your business and so it's been even more fun to study this particular one up close. But if you just think about the overall arc and the way that these technologies are stacking and your role in creating some of the steps that we all get to climb is so cool and so fun. I always ask everyone that I interview the same traditional closing question. What is the kindest thing that anyone's ever done for you?
- [92:59]
Ray Ozzie
The kindest thing that has ever been done was on the career side, was done by a man named Mitch Capor, who was part of the Hacker Hustler duo of Mitch Caper and John Sacks, who did Lotus 1, 2, 3. And Mitch could see that I had a passion in a certain realm. This was in prototyping the product that eventually would become Lotus Notes. I was only a small number of years out of college. I was a pretty decent engineer. I worked on spreadsheets at Personal Software for Dan Bricklin and Bob Frankson. Very fortunate with that. And then I went to work on a little spreadsheet work for Mitch and John. But those guys saw that that isn't what was exciting me. What was exciting me was this concept of people collaborating, using the computer as a communication device. And it was early in the industry, it was too early. But he could see that was my passion. And after I did some work for him, I built a product with a few guys, of course, called Symphony. The day that that product shipped, he came down to my office unprompted and said, ray, I know you've been on the side trying to get venture capital. I know that nobody will listen to you. I know that you've been striking out. You've told me, John's told me, let's figure out a way that you can do what you want to do. I don't understand what you're trying to do. I don't get it. But you clearly are good at what you do and you have a passion about it. So let's figure it out. So for the next six to nine months, we negotiated and we found a way to structure a relationship. And it was a very unique relationship. And I spun off and did a startup. It was a very unusual way of being funded, but he gave me that opportunity, and like I said, he didn't understand it. A career is the connection, the sequencing of many relationships and much learning. And that was probably the biggest single leverage point.
- [95:13]
Patrick O'Shaughnessy
It's so cool to think back on Lotus Notes, Grooved Networks, Azure Safecast, now Blues Wireless. So many interesting things that you've been a part of building and that it's started. These path dependencies are so interesting. The fact that it started that way is so cool and hopefully it makes others like it's making me think about who could we do that for? Whose passion could we go encourage and support even if we don't fully understand it ourselves? A wonderful closing thought. Ray. Thank you so much for your time.
- [95:43]
Ray Ozzie
My pleasure. Thank you so much, Patrick.
- [95:46]
Patrick O'Shaughnessy
If you enjoyed this episode, check out joincolossus. There you'll find every episode of this podcast, complete with transcripts, show notes, and resources to keep learning. You can also sign up for our newsletter, Colossus Weekly, where we condense episodes to the big ideas, quotations, and more, as well as share the best content we find on the Internet every week.