0:00The more agentic applications that you're building, the more human in the loop you're introducing.
0:05We're in the coding acceleration.
0:08We're in the beginnings of a What were your predictions that have changed?
0:13So with the mathematicians who go, well, okay, this problem has worked in the last decade. AI solved it in an hour.
0:20I barely see any companies hiring [music] for entry-level folks. Like any 20-year-old who are going to college, they're like, I don't even know what job role should I apply for?
0:28Your mid-level, mid-career person is going to be much slower to figure out how AI can adapt your business and then the next generation. You got to hire at least some of them.
0:38Do you think the society as a whole is catching up at the same pace?
0:44Good writing is not as much of a human skill anymore. Like anybody who has a $20 subscription can write amazing things.
0:51Yes. None of that is really great writing. It's good writing. I I frequently say that the future is sooner and stranger than you think.
1:04[music] [music] Hi everyone. Welcome back to Control Alt AI and I have a very, very special guest today. It's Reid Hoffman and I'm sitting down with him for the second time. You know, Reid, what was funny the last time we met and I had these like little cards which had questions for you. You told me that the tagline of the show is good, bad and the future.
1:40And you thought it was very funny because it's not the the ugly, but it's the future.
1:45And since since that one year, since I sat down with you, a lot of things have changed. And I remember having this conversation with you when you just launched your book on Super Agency.
1:56And I was talking to you about like, you know, hey, Reid, like tell us more about where do you think this is heading and like, you know, how much agency is too much agency?
2:04What do you think has changed since then? What were your predictions that have changed? And what do you think is like, I told you so.
2:11[laughter] Two days later, he finishes answering the first question. Um, so, [clears throat] um, well, so we're here at the Masters of Scale Summit. Uh, uh, agency is our subject here too. So it continues to be the thing and part of it is that agency is a mindset. Like it's like uh, people start with, oh, if it's agentic technology, does that mean it has all the agency because it's like an agent? And it's like, no, no, just like a travel agent or a doctor, it can be your agent too. It can be an expansion of your agency. So I think
2:39um, we continue to be turbulent about that because the capabilities continue to increase and part of what I think a lot of people are not tracking and I think this is one of the good things about like OpenAI releasing the math results and the Navier-Stokes is that part of the reason OpenAI and Anthropic are so much like, oh my gosh, this huge intelligence is coming is not just through the ChatGPT or the Claude or the Copilot or Gemini that you see, it's because the models that they have
3:09that they're operating with, that they're developing, that they're testing internally are capable of wondrous things. You know, math proofs and so on. So and you know, part of the reason I start with that kind of math example is you know, a bunch of mathematicians said, oh my gosh, this is this is crushing to me and but my heart goes out to them and it's a challenge it's a challenge these technological transformations.
3:30But the same thing like when you were a weaver and the power loom came along, it's like, well, look, this is part of how we get progress in society, how we have a middle class, how everyone gets, you know, you know, clothing that they can afford for themselves and their children and all the rest. And that's part of the increase of productivity and there's infinite math.
3:46So what the mathematicians who go, well, okay, this problem was working last decade, AI solved it in an hour, right? You go, great, what can you do great, uh, directing the AI to solve even better, more important, relevant questions? And that's kind of I think the agency. So I think, we are, you know, in the, what is it roughly probably a year and a half or something or since we last talked.
4:11You know, obviously the capability is gone massively up. We're in the coding acceleration. We're in the beginnings of recursive self-improvement. It's part of where you get the the coding acceleration. And you know, this is what I would say is even when you're kind of like I'm predicting that it will be fast and important, it's like wow, you you're really feeling that acceleration of what the capabilities are.
4:38No, 100%. So when you were talking about this coding agents, right? What we are seeing and we were just speaking about this on how building software like this first layer of user experience has become so easy and so cheap. [snorts] That is also created a lot of disposable software.
4:56So there is everybody's talking about code abundance. A lot of people, especially the reason I wanted to bring this up is a lot of engineering managers that I speak with, they're like the amount of code reviews and PRs that gets thrown at us is insane.
5:12And that goes back to this human in the loop that we're trying to achieve, which sort of is a I'm just playing devil's advocate here, right? Like when you have human in the loop, the more agentic applications that you're building, the more human in the loop you're introducing. And at the end of the day, the amount of time humans are spending in the loop is probably higher than what they used to just doing things from scratch. So [snorts] in this like
5:39tug of war of how much agency versus how much human in the loop, who really decides on what that balance should look like and like would it be the users? Would it be the companies?
5:51How do we really get that right balance of not losing productivity, but not like giving up human in the loop as well?
5:59Well, it's going to be an intense work in progress, right? Because the capabilities change, our understanding of them changes individually, our understanding of them in a group, in a company, changes. The different preferences and interests by individuals, the different preferences and interests by the organizations and companies. So there's a just this huge, you know, convoluted space Mhm.
6:21and dynamically changing over time.
6:23Now that being said, I think that some of the principles are one is having a notion of human in loop, which is sometimes you have to say, okay, it's actually really important to use there's there's various forms of human responsibility, human judgment.
6:35Sometimes actually even today, even with you know, kind of the the the great models of you know, fable and soul and others, there's places where human judgment is actually really important for shaping what the outputs is, for checking it, for directing it in the right direction, for for knowing when it's actually off target or not good enough. And you know, like the easy thing to say is like, you know, even these amazing AI models, they can solve Navier-Stokes, but they can't really write a good novel.
7:05[laughter] Right? You know, like like it's just kind of is is in the you know, kind of spectrum. And they and they're still fairly dad jokey in their humor, [laughter] right? You know, so so it's kind of like, you know, it's like look, there's this kind of jagged edge and a bunch of different things. Now that being said, I think that one of the things so one thing is you have to have kind of not being overloaded, not having human in loop, being thinking about like, you know, adapting and changing over time.
7:29And you don't change to something today thinking, oh, that now I'm set. Now now the next 10 years we work this way. It's like that's not the way it's going to work. It's it's an adaptive like experiment kind of basis for doing it. And now another thing that I think will become a more common concept too is a human on the loop, Mhm.
7:46which is the work is happening, but it's kind of overseen by human beings. It's not and and by the way sometimes the agentic loop will go, oh, now I need your attention. Like I'm I'm like I I I've got something that's like kind of a question. I'm going to wait for you or I'm going to say, hey, there's a question while I'm continuing work, etc.
8:05And I think that will be another thing that will be in this.
8:09And then when we see that it kind of um that the the the amazing productivity starts putting strain, like you were mentioning the senior engineers with code reviews and PRs and so forth, which is like I'm spending all my time doing this. It's like, well, okay, how do we tune that work process? One, like how do we make sure that the that the judgment of what's coming out of there is good? How do we have AI help those senior engineers?
8:33How do we make sure that that um continues to be a like an even more interesting, more cognitively elevating, more fun kind of challenge for that kind of work versus just feeling like you're you're in a you know, like you're trying to surf a wave of AI generated code.
8:50Like part of the code like we're going to have code everywhere, right? There's like part of the AI revolution is that there's going to be intelligence everywhere that there's electricity and connection.
9:02And so and code will be an important part of that. So I actually think that, you know, you're um like you know, for example, when you're saying, hey, I'd like my household to run the following way in terms of how we're all doing our meals and nutrition, you'll probably code up something for that and have an agent help you with that in terms of there's a like it'll be from that to like, you know, um doctors going, oh, if I had this kind of research assistant helping me with this stuff, I'll I'll I'll I'll put that in
9:30code. So so I think it'll be everywhere, but of course, it being everywhere means new things to sort out and manage and we will figure that out dynamically.
9:39I absolutely love this and I want to tell you this because I feel you'll really appreciate it. The way that I've been hiring for people in my startup has been extremely AI native. So recently we had some interviews for like GTM professionals and the way that we're interviewing them is like share your screen, open up ChatGPT or Claude, whatever is your favorite tool, and here's the problem that we're trying to solve. Why don't you show me how you'll be using these tools to do it, right? And that's where like using it as a brainstorming
10:09partner, coming up with multiple ideas like a GTM person could be leaning into partnerships and marketing and building dashboards for something like human on the loop. Like if I am like doing certain automations, how do I put it and on a tracker and on a dashboard and see these things online?
10:27That actually like makes me feel this in a funny way, right? Like I feel there are a lot of folks like us, folks who are in Silicon Valley and probably like you know, some of the other cities who are also as AI savvy.
10:40But do you think the society as a whole is catching up at the same pace?
10:45No, I think not nearly and most of the time most of the society, I mean the the polls on AI are very negative and part of the reason is because what they hear is well there's all these bad things coming. You know, there's going to be your you know, your job's going to get challenged, maybe the data centers are going to you know, this is the fictional one is used too much water or raise electricity, although I think local communities just saying, hey, you can put your data center here but make it good for us, which means lower
11:14electricity costs and help invest economically and so on. And great and then local community should do that and then the you know, state and federal should be oh and then we make sure that that works relative to environment and blah blah blah. So but I think that's what they kind of hear and they need to get more exposure to things that are like valuable for them, right? And so it's part of the reason why like you know, about a month ago I put out this piece saying America should have a dual mandate which is both kind of lead and win in AI because it's the industrial revolution and kind of
11:44what makes but also then help the 60 plus percent of Americans who are struggling and and part of that is because look, we're going to get all this productivity, what are the kinds of things we can do?
11:57And and as people begin to get experience of how AI can help my life even as it changes things and changes, most people are getting a little nervous and change and you know, they unlike entrepreneurs like yourself, they don't they don't go oh, jumping off the cliff and working it out. You know, most people are like no, I want a stable job and I just want to know what everyone. So, times of change are stressful.
12:18But they also create great opportunity for you.
12:22And, you know, how do we kind of, you know, get engaged into using the AI?
12:28Use it for medical, use it for legal, use it for education. Like like a personal use that I do and I recommend to people is go to one of the kind of top models. Um could be Soul, could be Astra, could be um Fable, other. And and take something that you really want to learn about. It could be something in a career or job, it could be something in a, you know, like a piece of like um uh like a piece of intellectual knowledge, like, you know, quantum mechanics or something,
12:56and say, please ask me the questions that um would allow you to make a personalized course for me to learn and understand this.
13:06Right? And um and it works. It is amazing. I use it myself personally.
13:11And so if people start doing that, they go, oh, this is how it helps me. This is how it helps me get agency. This is how it helps me navigate my life, navigate even work.
13:21And that kind of thing I think is very important.
13:23You know, Reed, you've mentioned something very, very important and I I'm also in the same advocate for like, you know, now information is so easy for us to find, personal information, like what used to take us to find a mentor or somebody, like the knowledge which was only in like, you know, certain people's head, now it's accessible using AI for so many of us. It has democratized the the playing field for all of us and be like, okay, like I can be what I want to be. I
13:52can learn even if I'm sitting in like a remote place in the world. But with this information, one part that I've seen is also used in not so good way is I was actually talking about something and the reason like we have gone completely when we are interviewing people is we want to look at their thought process rather than their resumes because every single person's resume looks the same.
14:16Uh, LinkedIn had been my favorite platform, like since forever, but lately what I've started seeing is there's just so much content on the platform and it's not just LinkedIn, right? Like every other platform is just getting flooded with AI generated content and video and everything um that somewhere distinguishing between what is a true human skill versus Mhm.
14:41at one point in time writing was a human skill, right? Like there were very few good writers, but now good writing is not as much of a human skill anymore. Like anybody who has a $20 subscription can write amazing things.
14:54So how do you really like keep that good human skill? How do you judge like what's a good human skill?
15:00Well, it's interesting you brought up writing because I've been reflecting on it the last couple of months and obviously part of the thing that these models do is they they a lot of people were just not good writers, felt stressed about it, couldn't do stuff. And now with you know, ChatGPT, Claude, you know, Copilot, Gemini, you can do stuff.
15:23None of that is really great writing.
15:26It's good writing, not great writing. Uh, it's not great writing. I mean the obvious one I mentioned before is like fiction and other kinds of things.
15:33Um, it can come up with interesting like plot ideas or something, but it's not it's not like the the the the the edge of creating, you know, kind of interesting characters and dialogue and so forth is well beyond even these most advanced models that are doing things like solving Navier-Stokes problems so far. It doesn't say forever.
15:54But like the notion of like how do you have like when when I'm engaging in writing, like the only time I've ever put like AI writing just in was when I did the book before in Prompto and I deliberately put it in to kind of show how how it was working. The other time to get the energy, the verve, the the human connection and I was I actually I'll use AI a lot in researching and suggesting even in editing, suggesting and so on. But at the end of the day, I
16:24end up like that tuning to make it a you know, an effort I'm you know, I'm not trying you know, to to make it make the writing better. And that's part of it is you can still get that and still have the skill of making it better, being the director of it you know, happening.
16:39And one of the things that I think is particularly interesting when it gets to this kind of use of you know, kind of AI is that we don't realize like, oh, it did these other you know, let's let's even be aggressive, 80% of the stuff that I was doing Yeah.
17:00Great. Your 20% is really important and then you learn other things about how to make it valuable. So like for example, like another skill that AI is still not very good at is making investment decisioning. It's good at like due diligence list. It's good it's it's reasonable at competitive analysis you know, etc.
17:19Um especially for things that are kind of public, although when you say this is the kind of the classic and this gets back to the great writing you say like if you were kind of putting in a PowerPoint deck or a or a um business plan and you say what's the one thing that most matters in this investment or the three things, it doesn't actually get that right.
17:40Right? And so similarly, that's part of what like great writing is because it's like what's the thing, not only is the expression of it and the and the um uh the the emotional connection and going on the journey with you, but it's also like what's the thing what's the one thing here that most matters? It's really good at giving you the seven things of which the one thing will be in it, right? But it won't be like it's like frequently like when I get a you know, like a due diligence list, it gives me like 15 things and I go, well these three matter Yeah.
18:08and these five fine and the rest of them no, right? But it doesn't have any particular order. Anyway, so that's part of the reason why I think and like for example, like you can tell frequently when you know, in a especially medium and long form, when it's actually been done by somebody with some skill and care.
18:27I agree. I've I I probably noticed this that with AI you can do really good mechanical work, which is where coding and math, all of this comes in, which is not really like human readable, right?
18:40But the part where we're talking about good writing, it's not about how technically accurate it is or like how grammatically correct it is. It is about like does it make my eyes spark? And that's not something that you can codify the essence of it.
18:53Yeah, it's actually it's funny. The first thing I I think I've got in parallel is is AI writing's a lot like Wikipedia. And Wikipedia is okay writing. It's fine.
19:03But you don't use Wikipedia as your example of good or great writing.
19:07Yeah, absolutely. Absolutely.
19:09So Read, I was talking to Parth just outside, and he was talking me up he was telling me about these crazy things that you have been working on. Parth is like every time I go meet Read, I come back with 40 ideas and I'm like, oh my god, how are we going to do this? So you are like clearly a AI tourist or AI tinkerer, whatever you could want to like call that. You're just like all of it, right?
19:34What are some of the things that you would say and you have been one of the first people out there to use the Read AI avatar? Probably I would tell you like I got really fascinated to use Hajan because of your avatar. So you have been doing this for like quite some time and I want to get your honest feedback as a user that hey, these are the things which I feel like AI has done a great job at. It's got it, but I don't think the frontier models are there yet for these sort of use cases.
20:04Well, there's a lot of use cases the frontier models are not there for. Um it's still and that's part of the reason it's not I'm not just optimistic cuz I'm optimistic about people and I'm optimistic about human beings' ability to learn and adapt, which I am optimistic about. But it's also because I actually see what are the things that AI is great at and what are the things that AI's mediocre at.
20:25And part of the reason why I do these things like you know read AI, you know they kind of I've had to do everything here at Masters of Scale Summit. It's doing podcasting interviewing for people coming by and Okay, excellent. [laughter] Thank you. I'm I'm very curious to get your feedback on it.
20:41Um but like you know for example like one of the questions that I ask call it every couple months is would read AI be capable of standing in for me giving the kind of entrepreneurship advice that I would like to get and in some things it can and you know we'll try to we'll try to then democratize that and make that much more available to to many more people and in other things it can't, right? And so it's a little bit like I was talking
21:10with the investment memo it's like part of the thing that I do in my in my work on boards and everything else I go like what's the one thing that most matters Mhm.
21:20to this person, to this entrepreneur, to this CEO, whatever she or he is kind of engaging in in order to help and yeah it's it's hit rate on that is you know at best moderate.
21:34Right? So so that's one kind of instance. The other thing is like um you know we have with again read AI we have a lot of um prompting uh to get it to ask questions um but like you know anyone who's been interacting with AI agents realizes that it's kind of it's been prompted to ask questions so it asks kind of like blah blah blah. So how are you feeling today? Is there anything else I can help you with? It's like that's a kind of a
22:01wooden thing. That's not seeing what that's not seeing the person being in the dialogue, seeing the dialogue. So the questions are still in in frankly a kind of a amateurist almost like kind of quasi like not quite socially full fully present kind of interactions and those are just some things that run across everything, right? Um you know I think some of the stuff like it it still lacks some kind of metacognition and context awareness. It's one of the things that
22:31I'm actually really you know, if you ask me to predict and you say, well, these these AI is going to get better year after year, 10 years from now, what's the role for human beings? It's like, well, I would guess that actually they think one way, we think another way. Part of the things that we add in is you know, kind of context awareness, metacognition, are we playing the right game? Does that look like the right thing to do? And my guess is we will still be very substantially doing it.
22:57It's one of the reasons I think about, you know, your use of AI is you're you're a conductor, you're a director, you know, human on the loop as well as in the loop and kind of his way doing it. And I think that that will persist because I think these limitations in the AI models will likely persist.
23:18Again, we're all going into this wondrous and like just strange the I frequently say the future is sooner and stranger than you think. We're going into those soon and strange times, which I think is we should be optimistic about, but it's going to be a very you know, confusing journey.
23:39So Read, one of the things I was telling you about this Anthropic report that I was reading, right, where it spoke about how the economic gap is going to look like and whatever kind of future they were trying to predict, Mhm.
23:51even though we could see the entire country's GDP go up, we could still see the middle class or like the people who are working class not get the same benefits as the foundational model companies or like people who are truly entrepreneurs, I would say. So it was the model owners who were getting that shift in profits versus the model users.
24:16And somewhere that also shows up in how jobs have started transforming and I barely see any companies hiring for entry-level folks and that's that's seems like a challenge, right? Like any parent that I speak with or like any 20-year-old which is going to going to college, they're like, I don't even know what job role should I apply for because no none of the companies seem to be hiring for these roles anymore. So how do we like really prepare ourselves for that shift?
24:43So first, Hard Corps at it's challenging times for the for the you know graduating classes, you know both last couple, you know going into them.
24:54And you know part of the thing is everyone goes, oh it's all AI. It's actually it's turbulent times. It's it's wars, it's tariff battles, it's COVID correction. There's a whole bunch of things. Anytime business get uncertain and AI is one of those uncertainties The they they just kind of pause on hiring and that that's legitimately anxiety producing for young folks. So one of the things I always tell businesses say, look you're wasting a great opportunity. You
25:21want to be hiring the AI native folks because that's the only way your organization is going to become AI native. And who is more AI native than the kids who have grown up with it?
25:32It's the same thing like the internet, it's the same thing with mobile. You know, so so that's the thing. And so part of the thing that I would say to to the to the the younger so the to the companies it's you must be doing this because it's critical. You're right, your your mid-level mid-career person is going to be much slower to figure out how AI can adapt your business than the next generation. You've got to hire at least some of them. Like bring them in.
26:00Then for you know you know graduates and you know recent graduates and folks to graduate don't just follow generic direction about like you know I went to chat GPT and it told me to write the following cover letter and I just wrote the following cover letter and like figure out how to show some AI nativity in what you're doing to do to do something that's interesting. Like for example, if I was a college grad
26:28you know I might go to you know Google's VEO or or ByteDance's C-Dance and say, I'm going to create a little video thing to go along with my thing to show you kind of what I'm doing and how that's helpful. Or I'm going to create a you know, kind of a an a research report, an AI document on something that that that would be good for the business and show that I'm I'm kind of like AI native and forward.
26:51And that kind of thing as a way of going, okay, like you you you desperately need AI native, you know, graduates, you know, new new workforce entrance to help you. Let me show you a glimpse of that. And I think that's that's the navigate. But it's it's it's not people tend to go to, well, you know, what I did is I, you know, worked hard, got A's, good grades, I did well my test and da da da. It's like, no, no,
27:21this is where you have to start being more out of the box thinking entrepreneurial. And this is part of the reason I wrote my first book, The Startup of You. Because it's like how how do you have the skills of an entrepreneur in your own career path? And it's those set of things applied.
27:37Read, I'm going to tell you something and I want to like prepare you for this like, you know, embarrassment that I'm going to give you because I absolutely absolutely love your personality. And I feel it's really really hard to achieve everything that you have and still have that hunger of a 20-year-old of being like, you know, I want to try something new. I want to go do something.
28:00You have absolutely no reason for hosting this amazing conference. And I was telling this to this your team outside as well that this is the only conference that I find the most value in because nobody is here to give me a product pitch. It's pure value and knowledge that I get out of this place.
28:18And this this capability that I'm talking to about, right? Like your urge to like always keep learning, I feel that's such an important skill to have in the AI world. How do you like keep up your energy? Like every single day once you like look back and you're like, I've done all of these amazing things. You've you're now like advisor at Manas, which is like another like you know completely different area that you're getting into.
28:40How do you keep up yourself with with that motivation to like just switch that trajectory, pivot into something new and like start from scratch?
28:48Well, I think the thing that I feel of many things that I feel fortunate about in my life, it's the people that I'm going through it with cuz I these magical people and it's and it's not you know it's like part of the LinkedIn thing is life's a team sport, right? We we go through with our allies and you can call it network.
29:05It seems like a little bit of a dry term, but like people we go through life together and the folks that I go through life help me see the magic of what we're doing and and when I see what they're inspired by what we're doing like I want to help too and that that that's what keeps me kind of young not just at heart, but also in mind and kind of like oh there's new great things to do and let's do that. And so I think that's that's the thing is like think about like who are the folks that help give you energy, you want to go on the journey with, you want to learn with,
29:34you learn from, like all of that is the is the kind of thing and that's what I I think that that that's the answer for me.
29:41Love that. Maybe just one rapid fire question for you. What is that one thing which you know that AI can do it better for you, but you're like no this is my secret thing. I'm going to like do it myself.
29:52Um um it's uh you know I've been uh a number of people who know me know that over a decades I've been thinking about writing a book on friendship and I think that would be the thing. I I wouldn't mind having an AI research assistant, I wouldn't mind having it edit it, but like friendship it's a very human thing.
30:13I I would want you know me yes.
30:16[laughter] Beautiful.
30:18We should probably get you a like the old school typewriter which makes those like cranking noises. I I love that. Well, thank you so much Reed. Thank you so much for sitting down with me again. I feel like compared to the 18 months ago podcast that we recorded like there's so many things that have changed and like your perspective still is so fresh and I'd loved loved love speaking with you.
30:38Yes, a great pleasure. I look forward to the next one.
30:41Yes, absolutely. Thank you. Thank you so much, Reed.