Skip to content
Silicon Valley GirlSilicon Valley Girl

$4B Founder: The Next 3 Years Will Make 100 New Founders Rich

📌 If you're building with agents — visit https://outshift.cisco.com/?utm_campaign=fy26q3_agntcy_ww_paid-media_ioa-svg-outshift_podcast&utm_channel=podcast&utm_source=podcast to Learn More or Join Us at AGNTCY.org (https://agntcy.org/) Aaron Levie built Box from his college dorm into a $4 billion company. 64% of the Fortune 500 runs on his platform. He meets with 20+ enterprise CIOs every month — he sees AI deployment data nobody else does. In this conversation he says the next 3 years will create the next wave of giants. He explains which jobs disappear first and which ones get bigger. And he tells me why he still wants a human at the beginning and end of every AI workflow he runs. *Timestamps:* 0:00 — Intro 2:44 — What to Tell Someone Scared of AI Layoffs 4:43 — Why Agents Always Need a Human Supervisor 15:00 — Why Enterprise AI Adoption Is Slower Than Silicon Valley Thinks 19:07 — What Aaron Looks for When Hiring Right Now 20:31 — Top 3 AI Tools Everyone Should Be Using 28:17 — Why the 3-Year Window Is Real 30:39 — Where the Real Market Gaps Are Right Now 33:04 — Which Industries Have the Biggest Opportunity 44:19 — Which Jobs Will Disappear in the Next 5 Years? 51:05 — Final Advice for Entrepreneurs Starting Today *Links:* 📩 Follow my Newsletter: https://siliconvalleygirl.beehiiv.com/subscribe?utm_source=youtube&utm_medium=video&utm_campaign=futureproof-sub&utm_content=AaronLevie 🔗 My Instagram: https://www.instagram.com/siliconvalleygirl/ 📌 My Companies & Products: https://Marinamogilko.co #podcast #AaronLevie

Marina MogilkohostAaron Levieguest
May 15, 202652mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:00 – 2:44

    Intro

    1. MM

      The more I play with AI agents, I do realize that I need a person at the beginning of the process and the end of a process, so I still end up having more people.

    2. AL

      Some of it will be different roles, but I'm very optimistic that we're gonna use this technology to grow more and do more, as opposed to just replace.

    3. MM

      This is Aaron Levie, founder of Box.

    4. AL

      Welcome.

    5. MM

      $4 billion company. 64% of the Fortune 500 uses his platform. He says we have three years to build the next generation of AI companies.

    6. AL

      These market windows happen every 10, 20, 30 years in technology. The mainframe, the personal computer, the internet, the cloud/mobile.

    7. MM

      If you were starting today, what would you do to find the right idea, to test it, and to make first money?

    8. AL

      My first thing would be-

    9. MM

      Five days ago-

    10. AL

      Yeah

    11. MM

      ... you posted this: "We're at a unique moment of history where anyone with high level of ambition and core skills in any area can overcome a lot of historical experience requirements-

    12. AL

      Yeah

    13. MM

      ... with a role." Can you talk more about that?

    14. AL

      So it's this interesting dynamic where, uh, a- a younger group, not necessarily in age, but maybe in skill or time in that domain, so an earlier group in that domain, um, can have as much leverage and, in many cases, even more because of their mindset differences than somebody that is, like, super experienced in a field. Now- Now interestingly, the- the- the advice can go in all directions because you can, you know, you could have somebody maybe too early in that field and then, um, and then use AI in the wrong way and- and get the wrong outcomes. Uh, equally, you could have somebody extremely experienced that decides to adopt the technology, and then they have a total superpower because they understand all of the contours of- of whatever they're working on, you know, whether it's writing code or doing healthcare or doing biotech, and they will be actually, uh, even more capable of- of leveraging these tools if they have the kinda right mindset wiring, uh, to be able to lever them, uh, le- leverage them. So I think that the core idea is that we're just at this amazing moment where if you're super ambitious, you wanna go deep into technology, ideally you're technical or- or becoming technical so you can kinda really know your way around these tools, you can make up for, again, lots and lots of years of- of skills that- that you would've otherwise had to go and develop. And I think that's an incredible thing for democratizing, um, you know, knowledge and skillsets and expertise. Uh, I often am- am building things or designing things or coming up with things that I have, you know, in any other do- in any other version of the world I would never have been able to go do. But now I- I- I know just enough to be dangerous in- in- in those areas, and- and it helps me prototype, it helps me generate new ideas, it helps me kinda work with colleagues faster 'cause I can kinda, like, highlight the way I'm thinking about something, where normally I wouldn't be able to, like, draw on paper what I'm coming up with but- but I just say, "Okay, this is the rendering that we're looking to do." Uh, and so again, I think that's an- an incredible technology that's available to everybody for those that want to- to adopt it and- and lean in right

  2. 2:44 – 4:43

    What to Tell Someone Scared of AI Layoffs

    1. AL

      now.

    2. MM

      What would you say to someone, uh, who's watching this but they've also heard a lot of news about layoffs-

    3. AL

      Yeah

    4. MM

      ... about college grad- graduates not getting enough jobs because they're being replaced by AI.

    5. AL

      Yeah.

    6. MM

      Uh, w- what would you say to those people?

    7. AL

      Yeah, I think there is... I think we're at a moment right now where, uh, and these- these happen in history, you know, every couple decades or every, you know, 50 or 100 years where there's a major technology disruption or transformation, and there's a lot of questions around, okay, where does that show up? Who are the- the people that get enabled by that, and they can do even more? Who are the people that maybe get displaced by that, and what do they do next? So we're in one of those periods where it's a, it's a serious topic and a real conversation. Uh, I do think that some of the negative kinda commentary and messaging out of, you know, the industry or even, you know, kinda political, uh, you know, institutions, uh, probably is over-weighting the negative side and under-weighting the positive side. Uh, for instance, I'll give you one example. So there's this sort of death of the software engineer topic that- that comes up, and that comes up because these AI models are really, really good at code generation. So they're really good at writing code and, like, you- you look at them and you're like, "Oh my God, that's incredible how much code it just wrote," and it- it wrote that code as- as well as another engineer would've. And that's all totally true, but to get that code into production, to make sure that it's secure, to have it maintain an application on an ongoing basis that doesn't get hacked, to make sure it's integrated across all your other data systems and database and infrastructure, that still requires a tremendous amount of knowledge and expertise in the field broadly of coding and in software development. And so the people that are gonna be able to best leverage the technology are actually gonna be software engineers using code agents to be able to generate vastly more, you know, code output than they would've been able to before. So-

    8. MM

      But that's-

    9. AL

      Yeah

    10. MM

      ... that's today. Do you ever think about, like, in five years AI's gonna be able to do that? Uh, I don't know, look at the market, strategize around some problem that-

    11. AL

      Yeah

    12. MM

      ... the market is not solving yet.

    13. AL

      Yeah.

    14. MM

      Build a company, [laughs] develop software,

  3. 4:43 – 15:00

    Why Agents Always Need a Human Supervisor

    1. MM

      and that's it.

    2. AL

      Yeah. There's a lot of data signal that isn't, you know, digitized in a, in a format that the agent can go with, and there's a lot of ways the agent can get confused by accessing the wrong information or doing the wrong thing that you didn't intend. And so for all of these reasons, it leaves humans in- in some kind of supervisory capacity for what these agents need to go do. And so does it need the same number of humans as we have today for the exact same workflow? No. Not- not usually. But are there a lot new workflows that businesses will now do because they have access to those agents? That's the sort of bet that I have. And so the way I kinda think about it is, is if you think about that five-year-out scenario. Let me, let me paint a, a slightly different one. I'm a small business. You know, pre-AI I was three people. We were selling something online. It was a good business. It sort of paid the salaries of these three people. But let's pretend I, I had even more ambition and, and I wanted to go after a bigger market. What do I do if I'm, if I'm those three people? Well, it's like I have to hire a sales team, I have to hire a marketing team, and a lot of people are just like, "That's a really high barrier to entry to grow my business, you know, meaningfully."Now, enter agents. And you're like, "Oh, I want an agent to go and generate this marketing campaign," or, "I want this agent to go and build a better website that delivers a better experience for my customers." Well, what happens next? If it works, now you have more customers, now you have more supply chain issues, now you have more customer kind of interaction challenges. You have new features they want you to build. Then all of a sudden, because you had agents go and get you some of the way to getting some of the work automated, my hunch is that same three-person business becomes five people or becomes 10 people because they now have automation that's augmenting the prior constraints and limitations that they had. I think that's gonna happen as much if not more than the scenarios where you have a company that is sort of saying, "Okay, I have 2,000 engineers today. I'm gonna have 1,500 in the future." I think it'll be a much more diffuse set of growth that happens through the economy.

    3. MM

      Yes.

    4. AL

      Some of it will be different roles, but I'm very optimistic that we're gonna use this technology to grow more and do more, as opposed to just replace.

    5. MM

      So Aaron and I have been talking about building an AI native business. One quick thing before we get to it. Look at your work right now. You maybe have one tab for emails, one tab for research, one tab for decks. They all run inside your company. The problem is none of them really know the others exist. Here's what that looks like for us. We record a podcast episode. To turn it into a LinkedIn post, somebody on my team opens the transcript, copies it to another tool, prompts a writing agent, then copies the output somewhere else. We do that for every single episode. The agents already exist. The bottleneck is that they can't pass work to each other. A person has to sit in the middle and move files around. That's what Outshift by Cisco is here to solve. They call it the Internet of Agents, an open infrastructure where your transcription agent can pass a file directly to your writing agent, which can pass the output to your scheduling agent, and there is no human in the middle. No manual copy-paste. These agents are coming from different vendors or might be built on different frameworks. Doesn't matter. Verify who they're talking to and move the work forward on their own. It runs on existing protocols like A2A and MCP, and it works with whatever you're already building. The open source project is called AGNTCY.org. It's a Linux Foundation project. Outshift by Cisco was a co-founder with 80-plus members contributing to it today. If you're building with agents or just watching where this is going, go to outshift.com, discover the Internet of Agents, an open interoperable internet for agent-to-agent collaboration. Now let's get back to Aaron. And because we're doing more, we're basically consuming more, right? And solving more problems, so we're becoming a more abundant world.

    6. AL

      More abundant, and you, you, you can't escape some ultimate constraint. There's always some constraint in the system. There's a new bottleneck that emerges. Um, I have lots of things that I've tried to automate where at the end of the automation, the very next thing you have to do is a human has to do some work. It has to follow up with the customer 'cause I just can't fully automate that entire process. It has to, you know, update data in some system. It has to go into three meetings and, and, and kind of coordinate with some other kind of, you know, set of, set of people. It has to go to the customer's site and do some implementation. Um, so there's always constraints in the system. We just haven't identified all of the new ones that happen when agents kind of arise. Um, there was a funny article, uh, about a week ago in the Financial Times where lawyers are now being inundated with questions from their clients because their clients are going to AI agents and asking questions about legal issues, and they're drafting documents or whatever.

    7. MM

      Yeah.

    8. AL

      But guess what? Like, if you were to go draft a contract right now, the very next thing I predict you would do is you'd go and send it to a lawyer and say, "Can you just make sure this is, like, gonna, like, you know, hold up in court?"

    9. MM

      Mm-hmm.

    10. AL

      Because in the 3% chance it's not, which is basically maybe the hit rate of, like, like, what an agent will get right or wrong, that's not worth the risk of-

    11. MM

      Absolutely

    12. AL

      ... of saving $500 of talking to that lawyer.

    13. MM

      Yeah.

    14. AL

      Like-

    15. MM

      Same with, like, financial advisors, right? You still wanna run something through a human. [laughs]

    16. AL

      I am, like, not that interested in automating my tax, my personal tax process.

    17. MM

      Hmm.

    18. AL

      Like, I am totally fine with, with, you know, the, the one-time fee to just make sure that that is just like-

    19. MM

      Everything's clean

    20. AL

      ... a, a clean process-

    21. MM

      Yeah

    22. AL

      ... from, from somebody that has, like, done this for 10 years or 20 years or 30 years, and, and there are just some parts of the economy which is naturally already where, you know, dollars tend to flow, where you're like, "I just want this done well."

    23. MM

      Yeah.

    24. AL

      I, I want my doctor to be really good. I want my lawyer to be really good. I want my tax advisor to be really good. I want them using AI because if they could somehow, like, review more of my data or look at more of my patient history or look at more of my, my legal history, that would only be a net positive. But I want that person ultimately to have some degree of accountability that's on the line. These agents have no accountability. They're not on the line.

    25. MM

      That's, that's the thing.

    26. AL

      They're not on the line for anything.

    27. MM

      Yeah.

    28. AL

      They're gonna disappear in, in two seconds later.

    29. MM

      Yeah.

    30. AL

      And they have-

  4. 15:00 – 19:07

    Why Enterprise AI Adoption Is Slower Than Silicon Valley Thinks

    1. MM

      Exactly.

    2. AL

      And so that part exists very broadly throughout our organizations and throughout the economy. And so I think, I think what some people in the AI ecosystem that lean more to the sort of rapid takeoff, you know, kinda quick takeoff scenario is that they're th- they're, they're thinking that because the agent can do lots of stuff really well, that that sorta diffuses across the economy in a way that is sort of this destructive scenario. And, and, and I have, I don't know if it's a benefit, but, but it's, it's [laughs] certainly a, a reality. Like, I have the fun, pragmatic reality of, like, I work with enterprises day in and day out, and these are enterprises outside of Silicon Valley. They're in the real world. They're the manufacturers of our products. They're the banks that we, we, we, you know, kind of bank with. They're the life sciences companies that develop drugs. And, and what these really amazing researchers and thinkers don't do is they don't talk to those people-

    3. MM

      Mm

    4. AL

      ... who are actually implementing these systems. And so they see this incredible capability takeoff, but they don't realize the diffusion of that AI across our organizations is ultimately constrained by and bound by 30 other things that doesn't really relate to the super intelligence that's in that model. It relates to how do I implement this thing in a safe way with the right safeguards so it doesn't blow up my, my, my data structure? I, I just think that the timescales are wrong. Uh, the way that people imagine the AI being implemented in society is, is, is generally wrong. Now, there's a one real risk that I agree with, which is there is cybersecurity risks. There are risks of mis or disinformation challenges. Those are very real. We need to work through those. But I'm, I'm much less, uh, inclined to believe the, this thing takes off, it replaces all white-collar work, um, and then we're in some really bad scenario on that.

    5. MM

      But we still hear all the news about layoffs happening due to-

    6. AL

      Yes

    7. MM

      ... AI. Do you think it's due to AI?

    8. AL

      Some of it is, is definitely not due to AI. It's over-hiring during the kind of zero interest rate era, the COVID era. Um, so there's some phenomenon that, that, that kinda relates to that. I would say that some of it, it definitely could be related to AI. Like, there are some organizations that they're like, "Listen, I had 3,000 people working in engineering before. My product roadmap is sort of not... doesn't need to, to triple in, in sort of scale. It needs to grow by 50%. And so I think that if each engineer can be, you know, 2x more productive and my roadmap only scales, you know, 50%, then, uh, then, then I think that there's some sort of savings there as a result of that."

    9. MM

      Mm.

    10. AL

      And then they might do a, a layoff in that scenario. So I, I think that is real. It, it's not something that I can, I, I, I can sorta gloss over. But what I see from customers, and you can go online right now and, and, and I, I, I guarantee if you took five random companies in the Fortune 500, just as an example, take five random companies. I guarantee that every one of those five companies is hiring software engineers right now.

    11. MM

      Yeah.

    12. AL

      And so where are they hiring their software engineers? They're hiring them... There's a, there's a, a, a, a, an interesting posting right now on, on Eli Lilly's career website-Which is a lab software automation engineer. This is a role to use AI to help sort of automate and, and increase review of lab results and automate the lab process in, in life sciences discovery. The kinda general idea of like AI is gonna destroy software jobs, that is not playing out empirically, and I, I, my predict it will not play out ultimately.

    13. MM

      Hearing that story, the thing I keep coming back to is that it wasn't talent, it was the system around her, and it made me think about something very simple. Most people use Claude like a search engine. They type in a question, they get an answer. Most times they're not really satisfied with it, and they close the tab. I did the same thing for months, and I was looking at people who were saying AI is changing their life, and I'm like, "Mm?" Then I spent one afternoon setting it up properly, uploaded a few files about how I think and how I work, and it completely changed. I wrote the whole process up step by step. You get it when you subscribe to my newsletter, Future Proof. It's free. The link is in the description. Interesting. So you think we're, we're gonna get more jobs-

    14. AL

      Yes

    15. MM

      ... in the next few years. When you're hiring now, how is it different from hiring five years ago? What are you looking for in a candidate?

  5. 19:07 – 20:31

    What Aaron Looks for When Hiring Right Now

    1. AL

      So I think right now is a great oppor- great time to be going deeper technically. You don't have to like... Doesn't mean, like, you have to, like, be vibe coding all the time and building entire products, but you should try and really understand what is the agent doing? How does it work? How does MCP work? How do CLIs work? How do skills work? And, and getting really well-versed in that. The people that are doing that will have a huge leg up in the next kinda three to five years because all of these companies will be hiring for people that can do that within their workflows.

    2. MM

      Mm.

    3. AL

      So we're definitely looking for people that whose technical acumen is, is growing, whose AI sorta savviness and fluency is growing. Um, you w- you wanna be using these tools, you know, in your, in your free time as much as possible so you, again, understand kinda how they're working and, and what's going on. Um, uh, at the same time, I don't think a lot of the, uh... I, I think it actually still matters that you have, like, some degree of domain expertise. Um, like you're really good at marketing. You understand what customers want. You're really good at selling. Uh, you're good at, at product management and, and interviewing customers and, and assessing markets. Like, those are the, the, like, timeless sort of, you know, skills that transcend any kinda technology revolution. And so AI is just a way of, of augmenting those domain skills. So there, in, in some respects, any role that we're hiring for, marketing, sales, finance, engineering, et cetera, we need all of those domain skills, but also we need you to now be-

    4. MM

      Technical

    5. AL

      ... increasingly kinda AI fluent or a little bit more technical.

    6. MM

      Uh,

  6. 20:31 – 28:17

    Top 3 AI Tools Everyone Should Be Using

    1. MM

      can you recommend top three apps that people should be using?

    2. AL

      Um, you know, pr- pr- I mean, it probably won't be much of a surprise. I would, I would download Codex. I would download Claude, uh-

    3. MM

      Even for non-technical people, Codex?

    4. AL

      Yeah, 100%.

    5. MM

      Yeah?

    6. AL

      Well, especially, uh, partly is, is because Codex is becoming more, uh, inclined, uh, toward knowledge work use cases.

    7. MM

      And what should they be doing with it? Like, automate a process within their work?

    8. AL

      Automate a process. Give it just a crazy problem and see what happens.

    9. MM

      Mm-hmm. Mm-hmm.

    10. AL

      Like, uh, go do this research in this market. Um, you know, wire up multiple MCP servers to data sources you have, um, so you understand kinda how does it work. Like, how is it querying that, that other system? How does it, how is it accessing my email? Like-

    11. MM

      Yeah

    12. AL

      ... like, oh, scary. Oh, no, actually I understand it now. Like, like, get, get a sense of how that all kind of is working together. So I think just any one of the top AI tools for productivity, maybe for coding, uh, is a good way to get started and, and it'll already get you like 90% of the way there.

    13. MM

      So Codex.

    14. AL

      Uh, probably Claude CoWork-

    15. MM

      Mm-hmm

    16. AL

      ... Perplexity. Like, these are some-

    17. MM

      Yeah

    18. AL

      ... just easy ones to just get started with.

    19. MM

      Yeah.

    20. AL

      And you'll have a good sense of kinda what the market looks like.

    21. MM

      Do you have any examples of workflows that you've automated for yourself and you'll never go back to manual?

    22. AL

      The kinda things that I'll never do again is, like, I'll never do, like, market research in a traditional way.

    23. MM

      Yeah.

    24. AL

      So I'm often asking an agent to, like, go and analyze, you know, 100 different companies worth of trends or information. I'll just never do that again.

    25. MM

      Mm-hmm.

    26. AL

      Like, I'll never go to Google and type each company in and, and do the research. Like, I'm gonna have an agent go and, and fan out, do all that, and then maybe I'll click, like, all the underlying sources and verify something or double-check something. So lots of market analysis. I'll never, I'll never open up code editor and, like, you know, type code again. And I, I wasn't for the past, you know, many years anyway, but, like, the reverse is true, which is now I can actually, like, get prototypes built when-

    27. MM

      Yeah

    28. AL

      ... I couldn't have before. Um, so anything coding related. Even design is like you just go to ChatGPT and you're like, "Hey, I need this idea done. Could you just, like, make it like this?" And then it gets you, like, image, uh, the new image rendering model gets you, like, 75% of the way there. You hand that off to a real designer, and then they kinda do the, the full thing. Um, so there's a lot in the ideation, the creative process, the market analysis, customer research, all of those domains that, um, that I w- uh, am heavily using AI for.

    29. MM

      Is there a certain way you structured memory? Uh, like did you... Uh, 'cause I hear some people, like, upload personal constitution, like their principles of work. Is there anything like that that you've done?

    30. AL

      Um, I'm less fancy on that front, um, and partly because I don't even know what I would write down-

  7. 28:17 – 30:39

    Why the 3-Year Window Is Real

    1. AL

      Um, well, uh, [sighs] I mean [laughs] I mean, it could be three and a half years.

    2. MM

      [laughs]

    3. AL

      Uh-

    4. MM

      But it's, like, not-

    5. AL

      Yeah

    6. MM

      ... 10.

    7. AL

      Yeah.

    8. MM

      It, it looks like we have a very limited-

    9. AL

      Yeah

    10. MM

      ... gap-

    11. AL

      Yeah

    12. MM

      ... in the market-

    13. AL

      Yeah

    14. MM

      ... where you can build something great, because then it's gonna be another, like, boring 10 years.

    15. AL

      Basic theory is, like, you know, these market windows happen every, every 10, 20, 30 years in technology. Uh, the mainframe, the personal computer, the internet, the cloud/mobile. So there's already been kind of four of these eras. And if you look at the biggest companies, you know, in, in, in tech, they generally correspond with-

    16. MM

      Mm

    17. AL

      ... when these windows open. There's a couple ones, a couple sort of examples that don't. Facebook sort of didn't correspond with any particular window. It was more of a, a social change that occurred as opposed to a technological change. But most other, other things, Google, Amazon, Microsoft, Apple, uh, you know, sort of the, the real turbocharging of IBM and, and in that era, and Intel and, and so on, they kinda correspond to a new technology kind of found at the foundation level emerges, and then, and then you have this opening where a bunch of new companies kind of respond to that. In our era it was, it was Salesforce and Workday and, and, you know, s- sort of enterprise software companies like Box that, that sorta were able to capture that moment. And then in mobile it was, like, Uber and, and, you know, DoorDash and another set of companies. So we're in a, a window right now that has all of the makings of that. Which is AI is now emerging. Companies need to d- companies are gonna want to apply this intelligence in various areas. And so there are gonna be a lot of applied AI companies that, that bring that intelligence to, to businesses, to society, to consumers in these applied use cases. And the only reason it's not, like, you know, 10 years is because-There's a lot of network effects remotes that get built. So if you build one of these companies and you're capturing data from the customer and you're improving the feedback loop of the agent, that'll just make your technology better and better over time.

    18. MM

      Mm-hmm.

    19. AL

      Whereby, at least on paper, that, that product should become more, uh, sort of, uh, eh, strengthened in its competitive advantage over time. So that's why it's like, yeah, it's not like an infinitely long window-

    20. MM

      Yeah

    21. AL

      ... because, you know, it's very hard to disrupt Walmart today because c- customers have been using it for decades.

    22. MM

      Yeah.

    23. AL

      Um, and, uh, and so you kinda wanna be in one of those spots as these markets are, are emerging.

  8. 30:39 – 33:04

    Where the Real Market Gaps Are Right Now

    1. MM

      Are you seeing any gaps in the market where a startup should be working on right now?

    2. AL

      Um, uh, still, I mean, tons.

    3. MM

      Mm.

    4. AL

      Uh, but, um, uh, I, I think there's still, like, y- y- I think, you know, everyone sort of knows the example of, like, Harvey right now for legal.

    5. MM

      Yeah.

    6. AL

      I think there's still lots of, of job functions, industries that will have their Harvey.

    7. MM

      Mm-hmm.

    8. AL

      Like, I don't think we've heard the end-

    9. MM

      Mm

    10. AL

      ... of the, the Harvey for X. I think there's gonna be new infrastructure that gets built out because these agents are gonna need new kinds of tools beneath them. Uh, there's, uh, th- there's all this new interesting stuff around what... when agents are doing work within software, they need more headless technology that they have access to. They might need payments. They might need... And so, you know, Stripe and, and their, their, uh, th- this new company, Tempo, is, is providing payments for agents. Well, now if an agent can pay money, then you can start to think through, like, well, what would the agent pay money for? And there might be new businesses that emerge-

    11. MM

      Mm

    12. AL

      ... that the agent is now going to transact with. Like, they're gonna need data probably.

    13. MM

      Mm-hmm.

    14. AL

      They're gonna need infrastructure. Uh, they're gonna need to do tasks for you in the economy. Like, there's lots of things that you can start to imagine that will become these new business models-

    15. MM

      Yeah

    16. AL

      ... because of, of what happens with, with agents doing this work. They're-

    17. MM

      With a whole new layer of, uh, active creatures in the market-

    18. AL

      Yes

    19. MM

      ... which are agents.

    20. AL

      Yeah, 100%.

    21. MM

      Yeah. Uh, w- if you were starting today, can you walk me through a plan? Like, what would you do to find the right idea, to test it, and to make first money?

    22. AL

      My first thing would be some mix of, like, you know, assume that we've got the most intelligent sort of system on the planet, so that we have this incredible AI intelligence, and just imagine that emerges. Then the question is, where in the economy would that add the most amount of value? And then try and think through, like, like, are there spots where, like, an incumbent isn't effectively responding to that? Um, uh, so that'd be, like, one framework.

    23. MM

      Mm-hmm.

    24. AL

      Another framework would be, like, where in the economy is it hard to deploy agents because there's a lot of other kinda systems that, that those agents need access to? And that's usually where, like, there's lots of work to be done to get the agent to work within the environment. I'm pretty excited by a lot of these new, um, kinda professional services, IT integration consulting firms that are emerging. Because, uh, when you go to the real world, and you're like, "Oh, would you like to automate your work with, you know, Cowork or, or Codex or any of these systems?" They're like, "Yeah, that'd be awesome." And then they show you their environment, and it's like, ooh, like, it's gonna

  9. 33:04 – 44:19

    Which Industries Have the Biggest Opportunity

    1. AL

      be a lot harder than you think.

    2. MM

      What markets?

    3. AL

      Anything.

    4. MM

      Anything.

    5. AL

      Everybody.

    6. MM

      Mm.

    7. AL

      Uh, healthcare, law-

    8. MM

      Mm

    9. AL

      ... life sciences, um, bank- I mean, just every industry. Um, because if your company's more than s- five years old, pre-AI, your data is all over the place.

    10. MM

      Yeah.

    11. AL

      You've got 30 different systems you're working with. Your workflows aren't documented, to the prior point. So that's a lot of change management you need to go do to implement agents. So what does that spell? That spells opportunity for new services startups. That spells opportunity for the existing Accentures and Deloittes of the world. It's kinda like a, a little bit of an up-for-grab market at the moment because of how much work there's gonna be.

    12. MM

      How do you decide between, like, building versus-

    13. AL

      Like, uh... maybe the only thing, like, Mark Cuban has had this riff, and I, I fully agree with it. There's gonna be, like, a lot of opportunity both for companies, but even just these will be roles that if you're, like, graduating right now, you might wanna think about is, like, who's the person that shows up at the 10-person consulting firm in Minneapolis-

    14. MM

      Mm

    15. AL

      ... just to, like, pick a non-Valley location. Who's the person that shows up that helps them take advantage of AI?

    16. MM

      Yeah.

    17. AL

      Because they don't have, like, a big IT department. They don't have a way to wire up their a- agentic workflows very easily. That's gonna be, like... Th- there's gonna be tens of billions of dollars, hundreds of billions of dollars that get made between jobs and services firms in just, just that over the next decade.

    18. MM

      Also, as an entrepreneur, when I'm thinking about that, but what if Claude just makes the process really easy? I don't know. You just deploy an agent. They build a specific agent who goes into your email, whatever you have, your Box-

    19. AL

      Yeah

    20. MM

      ... and creates the whole ecosystem for you. How do you think about that? 'Cause these companies are getting more and more powerful, right?

    21. AL

      If I took the, your exact scenario, and I'm like, okay, um, an agent's gonna read through my entire email inbox in that, in that scenario, and then it's gonna access Salesforce. And then it's gonna have some kind of, like, workflow that participates in. Like, even me as a... I've been building software for 25 years, and I use every single tool that has ever been produced in AI. Obviously, not literally, but, but, like, pretty much. I don't feel comfortable implementing that workflow right now.

    22. MM

      Mm.

    23. AL

      So the idea that that 10-person company is gonna go and set that up a- just because Claude became super powerful, I am skeptical that we ever get to that point.

    24. MM

      Interesting.

    25. AL

      Because-

    26. MM

      Okay

    27. AL

      ... b- and the reason why I'm not comfortable with that is, like, I don't... I have to think through the guardrails of, like, uh, what happens if somebody emails me and then says, "Hey, Aaron, I, um, you know, it, you know, you... I need you to pull up this Salesforce record for me that, that you told me you would, you would look, you know, look at. And, um, and you can send me that information." Well, if my agent has access to my email and my Salesforce, then the agent should, by design, answer that email question and go pull in the Salesforce record and then send it out.

    28. MM

      Mm-hmm.

    29. AL

      That's, like, a non-starter.

    30. MM

      Mm-hmm.

  10. 44:19 – 51:05

    Which Jobs Will Disappear in the Next 5 Years?

    1. MM

      Any jobs that are gonna disappear in the next five years?

    2. AL

      I think there's gonna be work that gets compressed. And, and then, and then I think you're going to take those people and often re- repurpose for, for, again, more of the agent manager escalation path or proactive versions of that work. A very kind of clearly obvious one is, is, and this is something that, that companies have always, you know, tried to sort of automate to some degree. Like, like if you're emailing a company and you're saying like, "I need you to reset my password," that is probably not gonna be a person on the-

    3. MM

      Like customer support, right?

    4. AL

      Cus-

    5. MM

      Yeah

    6. AL

      ... customer support. But even that, customer support is this funny one which is, which is like we, we, we think about it as a monolithic thing 'cause we call it customer support. There's tiers of customer support. There's like the first line of customer support that we will s- most certainly automate, which is like-

    7. MM

      Change password

    8. AL

      ... change password.

    9. MM

      Yeah.

    10. AL

      And I don't mean to like, you know, over-minimize that, that thing, but like there's a lot of tasks like that, which is like I need to download this thing. I can't log in. I have this issue, whatever. That we're gonna fully automate. But there's a lot of customer support which is like, "I need you to get on this call with me-

    11. MM

      Mm-hmm

    12. AL

      ... and look at my, my specific problem in my computer and why this thing isn't working."

    13. MM

      Yeah.

    14. AL

      And we just have no way to automate that.

    15. MM

      Yeah.

    16. AL

      Like maybe we'll automate like the next line of the, of the set of questions, but you can't ev- you can't get to the f- the final thing. I had a, I had a friend have a problem with Box two weeks ago. He just sent me some screenshots and there's a 0.0 chance that he would be able to-

    17. MM

      Mm

    18. AL

      ... have asked the question with an agent.

    19. MM

      So it had to be-

    20. AL

      This is not possible

    21. MM

      ... it had to be you. [laughs]

    22. AL

      And well, in this case, it actually, it had to be a senior product manager.

    23. MM

      Uh-huh.

    24. AL

      I had to get the senior product manager to the person.

    25. MM

      Uh-huh.

    26. AL

      But he couldn't have talked to a chatbot-

    27. MM

      Yeah

    28. AL

      ... and answered the question.

    29. MM

      But what about like bookkeepers? Something manual.

    30. AL

      I had, I had an issue with, with my Mac, uh, last week. I spent 10, 20, 30 minutes on AI trying to diagnose it. Never, never worked. Had to call IT. They had to come and diagnose it.

  11. 51:05 – 52:48

    Final Advice for Entrepreneurs Starting Today

    1. MM

      Yeah. Same, same here. I feel like as a society we're really slow to just change dramatically when it comes to foundations.

    2. AL

      Yeah.

    3. MM

      Uh, and college is one of them. Okay. Last question.

    4. AL

      Yeah.

    5. MM

      Advice for entrepreneurs who are starting today.

    6. AL

      I, I would say just, like, back to the earlier point, lean into the tools. Like, learn the technology, see what's, what's, what's possible with it. Um, make sure you're riding the tailwind of, of what's happening in technology. Uh, you don't wanna be, you know, kind of hitting a headwind, uh, where you're kind of going against the grain of the AI. You wanna be, like, riding the, the AI wave out, um, which can mean a m- a number of things. It might mean do things that actually in a world of AI become more important because people don't want AI to do that thing. So it's like, it's like this counterintuitive, like, riding the AI wave might mean do a live events business. Like, do-

    7. MM

      That's what a lot of people are doing.

    8. AL

      Yeah. Like, like, like, do something where we will appreciate this other thing in the economy because AI is sort of so abundant.

    9. MM

      Mm.

    10. AL

      Or, or, um, uh, AI makes getting healthcare questions answered so quickly, so you should probably be doing hospitals, like-

    11. MM

      Mm-hmm

    12. AL

      ... because now more people are gonna be, you know, go- like, like, actually needing real-

    13. MM

      Wellness clinics

    14. AL

      ... Wellness clinics.

    15. MM

      Yeah.

    16. AL

      Like, like, so there's just... Like, so sometimes it's a technology thing that you do. Sometimes it's a thing that the technology sort of is, is related to an underlying pr- broader societal trend that will become more important as well.

    17. MM

      Yeah.

    18. AL

      Build one of these, you know, consulting businesses that helps deploy the AI. Um, I, I just think there's gonna be, like, um, build a childcare service 'cause we're all... Sort of our brains are exploding-

    19. MM

      Please. [laughs]

    20. AL

      ... and we need help with kids.

    21. MM

      Yeah, yeah.

    22. AL

      Like, there's all this kind of stuff that, that, uh, that is gonna need to exist.

    23. MM

      Yeah. Thank you so much.

    24. AL

      Same.

    25. MM

      I, I like your positivity, especially after talking to some scientists-

    26. AL

      [laughs] Yeah, yeah

    27. MM

      ... on this podcast. [laughs] Thank you.

    28. AL

      Thank you.

    29. MM

      Thanks a lot.

Episode duration: 52:48

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode ExgNR94SrfI

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

iOSAndroidClaudeChrome