You're Not Using AI Like This

Transcript

FULL TRANSCRIPT

Episode: Reid Riffs with Parth Patil on Individual AI Mastery
Official source: https://www.possible.fm/podcasts/riffs041/

This transcript was generated from the local video and formatted by detected speaker. Minor transcription or speaker-label errors may remain; use the official source for the edited transcript.

[00:00] Reid Hoffman: Parth, I partially know the answer to this, but actually, one of the things I've been really looking forward to doing in this interview is actually to discover the fuller answer in some of these questions. So what is your AI stack, and how did you familiarize yourself with these tools?

[00:13] Parth Patil: You know, I was a little later to AI than maybe you were in your career. But, uh, for me, the big moment-

[00:17] Reid Hoffman: Well, I'm older. (laughs)

[00:19] Parth Patil: (laughs) I was, I was following AI during the video game AIs, the ones from DeepMind and, and OpenAI when they were working on DOTA 5. And then when OpenAI put out ChatGPT and it changed the world, for me, it was like, "Oh my God. This is the tool that you can use to teach yourself every other tool." And so it started with ChatGPT, started with getting really good at wielding a language model and asking questions, and then also having it look at all the tools that I have and be like, "Teach me how to use them even better. Teach me about my computer. Teach me about this video editing tool. Music." So ChatGPT is probably the meta-tool that I go to to teach myself how to use all the others.

[00:50] Reid Hoffman: This gets to kind of more of a Parth biographical personal question, but when was your light bulb moment of, "Oh, this isn't just a work tool, this is an everything tool and everything agent"? What was the light bulb, and what was the, the, you know, the moment of frisson?

[01:07] Parth Patil: So I used to work on Clubhouse, the audio app from the pandemic. And when, and I was working there when ChatGPT came out. And what happened was when ChatGPT came out, it became the most popular topic. And so it was an audio app. People would talk to each other on the internet and meet and talk about different topics. And ChatGPT became the most popular topic in every single part of the world. And so I would just join the app, join rooms, ask people how they were using the app, uh, using ChatGPT. And I had so many moments where it's like, oh, photographer learning about the different settings on their pho- on their camera, or then you have, uh, farmers in my home state in India that are using it to help plan their crop cycles. And then you have people here that are like, "Ah, I don't like how it rewrites my email." And I was kind of just like, "Oh my God, this is a new computer." You know, it's, it's the comp- it's the first computer that we can talk to. It's sort of like the C-3PO from Star Wars, except, I mean, eh, you know, Neuromancer, you talk about science fiction. This felt like it w- like, it felt to me that the conversational computer is 100 years early. Like, I never imagined it would happen in our lifetime, and yet here it was on our doorstep, already speaks every single language, all the human lang- well, most of the human languages, and all of the programming languages as well. So it's got this, like, incredible, like, general capability. And to think of it only as a work tool is an incredible oversimplification. It's, it almost represents the human collective intelligence in a sense, and it's a way for us to access the collective intelligence through natural language, by talking to, uh, an AI.

[02:31] Reid Hoffman: One of the things that I picked up from watching your usage was that you, by prompting a role assignment, you know, whether it's being VC, skeptical co-founder, emulate a customer, bunch of other things, but by prompting role assignments and various other forms of kind of meta-prompting, that could get you useful things. And obviously when you combine that with a swarm of agents or a set of agents, that even then becomes more of a, like, part of how we're all deploying a team. So say a little bit about your prompting guidelines, like maybe one that you would hand to beginners and then two or three that you would hand to non-beginners.

[03:12] Parth Patil: For beginners, I think it's mostly, start by just prompting heavily. Like, use voice, transcribe, talk at length about what you're trying to do. And then, um-

[03:20] Reid Hoffman: And maybe assign roles.

[03:21] Parth Patil: Yeah, and maybe assign roles. I think, uh, you'll find that the model can emulate all these perspectives. Like, you can say, "Pretend you are a VC and critique my business plan. Prete- uh, critique my, the, the way I'm running my startup potentially if I'm gonna go have a conversation with a VC looking for capital." Or, "Pretend you are the customer of this product that I'm working on and, uh, explore my website. What, what, do you, do you feel like my website properly, you know, uh, positions our product based on the, your, your needs as the customer?" And so the model gets to pretend to be all these personalities, and so even if you don't have that person, uh, sitting next to you, you can kinda simulate that perspective, and then the AI playing that role will help you understand your problem even in a way that you didn't even consider. Or you might say, um, "I want you to be the most skeptical person of all, uh, that you could possibly imagine and, you know, find 25 different criticisms of my approach to this problem." So I think role, role-based prompting is very, is very powerful, 'cause I think we need to get out of our own perspective. And being able to call on all these other perspectives that the model can emulate is really powerful. I once had a, a coding agent generate 100,000 unique expert personalities. Turns out if you create 100,000 unique experts, you kind of cover every single topic or a huge swath of all of, like, the topics that humans have. Like, 100,000 unique experts covering everything from, like, parenting, to, like, LEGOs, to, like, every single form of art that the models have been trained on, and you get this, like, very multifaceted, like, set of, like, minds that you can tap into. They're not complete minds, but they're perspectives. And then you can say, like, um, "Find the 10 that are most relevant to my problem and then have them all answer the question." And they all answer it differently. Like, that, that different perspective, like, an optimist is gonna have a different answer than a pessimist. But then, you know, an oceanographer is gonna have a different perspective than, say, like, an accountant on almost anything, uh, that you, you present. And so it, it, it's a very surreal thing. I call that role play. And then I think the, one of the most powerful prompting techniques, it's in the area of meta-prompting, like prompts, like it's the, it's the prompt that helps you find the right prompt. A simple one that I use almost every day is I go to, I go to a, I go to a language model, and I'll say, "Here's my problem. Um, describe the problem." And then I'll say, "Interview me until you have enough context to help me with this problem. Ask clarifying questions, and then we're gonna begin." And that's really important because I think a lot of people initially, you go to AI thinking you know what the answer is. And I think a better way to go to AI is let's describe the problem, and, like, maybe a set of solutions will emerge once the AI kind of collects that context and draws it out of you, the things that maybe you weren't thinking about when you first articulated it.And it's really good to, I call it the interview me prompt.

[06:07] Reid Hoffman: Mm-hmm.

[06:07] Parth Patil: Just interview me, and then we're gonna, we're gonna begin. And so I do that for any project that I'm starting from scratch. I'll say, "Interview me, and then we will begin." And it might just be a 10-turn back and forth, and I realize, "Oh, this consideration, I wasn't even thinking about that."

[06:21] Reid Hoffman: (laughs)

[06:21] Parth Patil: And the AI will ask those intelligent follow-up questions, kind of like any good employee isn't just gonna start doing the work.

[06:26] Reid Hoffman: Yes.

[06:26] Parth Patil: They ask the clarifying questions upfront-

[06:28] Reid Hoffman: Yeah.

[06:28] Parth Patil: ... and that, and then that brings out a higher quality, like, problem scope.

[06:32] Reid Hoffman: Mm-hmm.

[06:32] Parth Patil: And then when the AI begins, it becomes much more magical. And you have to kind of, uh, I think a lot of times people just, they need to recognize that maybe the answer that you have in your mind isn't the right answer.

[06:41] Reid Hoffman: Hmm.

[06:42] Parth Patil: And that the, the AI can kind of, like, feed off of your initial suspicion and g- provide an even better answer when you're in that loop of, like, iteration. And so I think for a lot of people, it's, uh, it's really just realizing that, uh, being a little bit humble about what our limits as humans are, how many scenarios we can s- we can think about, and how many considerations we can make, and then using the AI to expand that and parallelize that thinking.

[07:04] Reid Hoffman: Part of the thing is the mindset shift of how do you put your ego aside.

[07:09] Parth Patil: Yes.

[07:09] Reid Hoffman: So say a little bit about that 'cause this is actually one of the things that gets in-

[07:12] Parth Patil: Yeah.

[07:13] Reid Hoffman: ... the way of a lot of otherwise smart people's-

[07:15] Parth Patil: Yeah.

[07:15] Reid Hoffman: ... way under-delivery of AI.

[07:17] Parth Patil: Yes. For me, when I, um, when I was interacting with GPT-4 for the first time, so this was Mar- uh, March 14th of 2023, GPT-4 came out, which was the successor to GPT-3, and the first big leap after the ChatGPT moment was GPT-4. And, uh, I was talking to my teammate at Clubhouse, and we were both data scientists, data analysts, and I was like, "Olivia, this GPT-4, it writes perfect SQL, it writes perfect analytics code if it understands the schema of the problem that you're working in. If it understands your database organization, it, it just writes perfect analytics code." And she was like, "Parth, this thing aced the interviews for both of our roles." And I was like, "Wait, what do you mean, like, qualitative, quantitative?" And she's like, "Both. I think it's got a pretty good idea for what we should do as a company too." And I was like, "Whoa, what does that mean?" And, and, uh, and then we went to our manager, and we were, we were kind of like thinking about this. It. We had a small data team, so we were kind of just like, "Whoa, this language model is clearly an amplification of our own ability to do analysis." And he was basically like, "We're not going to hire anyone until we figure out how to use this, and then everyone we hire is gonna be using this." Because then we get this, like, super analytics kind of approach-

[08:24] Reid Hoffman: (laughs)

[08:24] Parth Patil: ... where I'm describing a problem in English, and then the AI is executing what I would have done manually by hand. And I had a moment where I asked for, uh, I asked for an analytics query. I thought it was a hard query to write, and, uh, it wrote a very elegant solution, and I just didn't believe it. And then I looked closely, and I was like, "Oh, that's just better than every version of the solution I've seen before." And it was very humbling 'cause I was like, "Oh my God, like, this is, this is definitely better than me at the writing of a SQL query." And then I realized like, okay, I think my job is to aim this. My job isn't to compete on the, like it's like the Kasparov versus Deep Blue moment.

[08:59] Reid Hoffman: Yeah.

[08:59] Parth Patil: But for me, it was like the data analysis. It's like the, or it's like John Henry, the steel man against the steam engine. And I'm thinking, "Well, I sure don't want to compete on a manual writing of SQL queries anymore."

[09:11] Reid Hoffman: (laughs)

[09:11] Parth Patil: "Actually, I would rather be working on the, like, automated version of analysis, where I'm speaking in English to the computer." It starts ex- uh, turning my, my questions into computer code that can solve the problems.

[09:22] Reid Hoffman: Yep.

[09:22] Parth Patil: And so that was a huge shift for me, and I realize a lot of people are a little bit later on that.

[09:26] Reid Hoffman: Mm-hmm.

[09:26] Parth Patil: Um, especially in engineering, I see experienced engineers tend to, they, they tend to be attached to their, their core, their core, their superpower, right? But I think you looking at AI and realizing eventually it might be better than you at the thing that you are really good at.

[09:39] Reid Hoffman: Mm-hmm.

[09:39] Parth Patil: But then your wisdom of working in that problem space becomes how you expand beyond just the, the AI or just you.

[09:45] Reid Hoffman: And when did you get to that recognition of it being a meta-tool?

[09:50] Parth Patil: I think it was probably like three or four months into talking to it for 14 hours a day.

[09:56] Reid Hoffman: Mm-hmm.

[09:57] Parth Patil: Um, realizing it could teach me about programming, realizing it could teach me about music. And then I was, I was like sharing screenshots of my desk, and I was seeing-

[10:04] Reid Hoffman: Mm-hmm.

[10:04] Parth Patil: ... that it could actually click around the c- Like, if, if you allow it to click around the computer, and you run it on an API on your computer, you're able to, like, orchestrate a web browser. You're able to write code in every single language. And I was like, "Okay, this is, like, language is actually the most powerful thing that you could possibly automate," I think.

[10:21] Reid Hoffman: Mm-hmm.

[10:21] Parth Patil: Uh, that's, that's, that was the realization there, that like, language touches everything.

[10:25] Reid Hoffman: Yep.

[10:25] Parth Patil: And then, um, and then you were always talking about Wittgenstein, and, uh, he has a quote which is, "Language is the limit of my world."

[10:31] Reid Hoffman: Yep.

[10:32] Parth Patil: And that was very, that was like, I realized then, like, my vocabulary, everything I've been exposed to in my life, I could now access intelligence through that vocabulary, through that language.

[10:41] Reid Hoffman: Yeah. So one of the things that in serious part I've learned from you is however we got to voice pilling. Say a little bit about how important it is to actually, in fact, be using voice, why that is, and what people will learn from that.

[10:54] Parth Patil: It is probably one of the most powerful prompting techniques there is. If you haven't tried it, it's really, like you wanna try it, you wanna go to one of these language models like ChatGPT, and I think people get hung up thinking about typing their prompt in a certain way and structuring their prompt a certain way.

[11:08] Reid Hoffman: Mm-hmm.

[11:09] Parth Patil: And I think actually what people should be more concerned with or more focused on is trying to get as much of the ideas out of their head into the model. So it's more about, like, you wanna say more. You wanna be able to get the, you wanna describe the problem. And I, I go to the extent of like, I will sit here, and I'll ramble for five, 10 minutes to the computer about the problem that I have on my, on my mind, and it turns into like a three-page kind of transcript. Even though it's kind of like unstructured stream of consciousness-

[11:33] Reid Hoffman: Mm-hmm.

[11:33] Parth Patil: ... turns out that is a very high bandwidth way of communicating with AI.

[11:37] Reid Hoffman: Mm-hmm. Mm-hmm.

[11:37] Parth Patil: And I think that some of my best prompts are not the ones that are structured a certain way, but they're the ones where I'm just being extremely effusive and very c- com- commun- communicating as much as I possibly can because I'm just rambling ab- at length about the problem. And I find that typing, when we type, we're kind of committing our ideas to a couple words.

[11:54] Reid Hoffman: Mm-hmm.

[11:55] Parth Patil: And in that process of committing, we're, we're not typ- we're not saying as much as we might if we were talking to a friend or we're describing a problem to, uh, someone that we wanted helping us with a problem.

[12:04] Reid Hoffman: Yeah, and part of that, that use of voice that I learned from you is not just depth of context, but also, and breadth and fill- filling in the blanks.... is when we're typing, we also tend to write, like, like, precision, like I write a coherent sentence and so forth.

[12:20] Parth Patil: Yeah.

[12:20] Reid Hoffman: Whereas, actually, in fact, these are such good... These AIs are such good interpreters, that even if you're like, "Well, I got a half-baked idea here," it actually will guarantee have a more focus that such that you might go, "Well, that was part of it, but now this is what I really mean."

[12:35] Parth Patil: Yep.

[12:35] Reid Hoffman: And that iterative gameplay, almost like video games-

[12:39] Parth Patil: Yep.

[12:39] Reid Hoffman: ... is a real key thing. So there's a relationship between the voice filling and also, like, almost like a video game style interaction.

[12:47] Parth Patil: Oh, yeah. Yeah. I, I, I think, I mean, I... As a gamer, I think, like, when you're playing games with your friends, you're not typing to them. You're really just, like, yelling commands. You're like, "Oh, I'm gonna come here. Here's what we're gonna do." And that's... It's faster. It's, it's, uh, high speed. It's... I think any real-time coordination, even on a basketball court, uh, people are like, they're yelling at each other. They're like, they're calling out what they're going to do. I think that's, that, th- this is like... Voice is a, is the way to get real-time coordination.

[13:11] Reid Hoffman: Hmm.

[13:11] Parth Patil: Both between people and also between people and AI. And also, yeah, I see sometimes people will be typing a prompt and then they'll, they'll ta- they'll have a typo and then they'll hit backspace. And I'm like, "This thing is really smart."

[13:21] Reid Hoffman: (laughs)

[13:21] Parth Patil: "It's okay to have typos." And then if I look at my prompts, all of my prompts are just filled, littered with typos.

[13:27] Reid Hoffman: Yeah.

[13:27] Parth Patil: Because I know it's like, it's so smart that it understands what I'm saying-

[13:29] Reid Hoffman: Yes.

[13:29] Parth Patil: ... even though a couple of the letters are in the wrong place.

[13:32] Reid Hoffman: Yeah.

[13:32] Parth Patil: Um, but yeah, s- same thing with voice, right? It's not about, uh... You don't need to have structured thought all the time. I think sometimes, if you're gonna use a prompt every day, you should think about that prompt. But if it's like a, a single one-shot kind of you're, you're describing a hard problem, it's more important that you describe it at length.

[13:48] Reid Hoffman: Yeah.

[13:48] Parth Patil: Get that context out.

[13:49] Reid Hoffman: One of the things also that you do that I think relatively few people do, is you use multiple of the frontier models, both in parallel, in rotation, in experimentation. How do you choose which models to use? How is that evolving over time? And any hacks or heuristics or principles or things that our listeners might be able to kind of apply or kind of remember and take with them?

[14:15] Parth Patil: Right. So it's a very c- I mean, in the beginning, it was just, it felt like it was just OpenAI and ChatGPT.

[14:19] Reid Hoffman: Uh-huh.

[14:19] Parth Patil: I think that, uh, it's easy to get kind of overwhelmed by the options that we have. But the main piece of advice, I would say, is you wanna get good at one state-of-the-art tool in every single category.

[14:30] Reid Hoffman: Mm-hmm.

[14:30] Parth Patil: So one really good language model, one really good image, image model, one really good, um, video model. And then those principles tend to translate over to the competitor products in each category, right? So you get good at ChatGPT.

[14:42] Reid Hoffman: Mm-hmm.

[14:43] Parth Patil: You're also probably gonna be good at using Claude and Gemini. And, and any given week, the number one model t- sometimes is different.

[14:50] Reid Hoffman: Mm-hmm.

[14:51] Parth Patil: Um, so I don't think everyone should necessarily be, you know, trying to stay up-to-date on that.

[14:55] Reid Hoffman: Mm-hmm.

[14:55] Parth Patil: But really, if you have one of the three, in terms of Claude, Gemini, and, and ChatGPT, you're really, like, like that should be your goal, is be really good at least one of them.

[15:03] Reid Hoffman: Mm-hmm.

[15:03] Parth Patil: And then, like, every once in a while, you see, you try the other ones as well.

[15:07] Reid Hoffman: First, a heretical question. Like, if you said, you know, say bizarrely someone has not done this at all, what's the model they should start with?

[15:14] Parth Patil: I would say start with ChatGPT if you're just, eh, getting into language models for the first time.

[15:18] Reid Hoffman: Mm-hmm.

[15:18] Parth Patil: It's, it's probably the best, um, like, general purpose assistant product, uh, productized version of an, uh, language model. And then if you're interested in more technical stuff, I think Claude Code-

[15:28] Reid Hoffman: Mm-hmm.

[15:28] Parth Patil: ... and moving into the coding agents is a, is a really good move as well.

[15:30] Reid Hoffman: What has been some of your experience in the last six months about, like, I prefer ChatGPT for this, I prefer Claude for this, or I prefer Gemini for this, your AI stack?

[15:41] Parth Patil: I'd say, like, my general purpose, like my web browser, I use the ChatGPT Atlas browser. So it's got ChatGPT baked into the web browser and it can control the browser. So you can tell it to click around and you can tell it to book flights for you, book hotels. I'm the kind of person that doesn't book a flight until last second 'cause I... And it's not that, you know, you know you're gonna go on that trip.

[15:58] Reid Hoffman: (laughs)

[15:58] Parth Patil: But you just don't take the 15, 20 minutes it takes to sit down and book a flight. And I'll just open a tab, and, and these days, I just say, "ChatGPT, go find me the, the best flight in the evening-"

[16:07] Reid Hoffman: Mm-hmm.

[16:07] Parth Patil: "... from LA to, uh, to, to San Jose," and it'll go find that. And then I say, "Go find a hotel," and it'll find that. And then all I do is the final booking.

[16:15] Reid Hoffman: Mm-hmm.

[16:15] Parth Patil: So I really like ChatGPT Atlas as taking-

[16:17] Reid Hoffman: Mm-hmm.

[16:17] Parth Patil: ... over this kind of, like, everyday kind of drudgery kind of work. It's very interesting. It's kind of like a mechanical turk, um, AI that just does the menial-

[16:24] Reid Hoffman: Mm-hmm.

[16:24] Parth Patil: ... form filling kind of task.

[16:25] Reid Hoffman: How much memory context of you do you have, uh, the agent keep in mind, like, for presumably with ChatGPT with, uh, GPT Atlas, to, like, when you say, "Best hotel, best flight..."

[16:40] Parth Patil: Mm-hmm.

[16:40] Reid Hoffman: ... it knows what your parameters are, for example?

[16:43] Parth Patil: So I think, uh, memory is a very interesting thing that these, these language models are beginning to start getting their, their, like, their grasp on. And memory is like, what is that personal context about me, my preferences, my, my tendencies, the things that I actually, like, prefer, that I would want the model to know so that anytime it takes an action. Like, it's like, "Oh, Parth likes to sleep in, so maybe don't book him a 6:00 AM flight." Like, that would be an interesting memory point for the, the model to take into consideration before making agentic decisions on your behalf, like take, booking a flight, for example. Or, you know, you really only need... Like, like I, I, I, I think that, like, the better it knows you, the better it can help you with some of these things. But memory is very tricky. I think it's largely unsolved. I think ChatGPT has more of my memories.

[17:28] Reid Hoffman: Mm-hmm.

[17:28] Parth Patil: But I also notice that, um, sometimes I'm like, I don't... I actually prefer talking to a coding agent that knows nothing about me personally, and I, and I, and I... because I like to sometimes have a fresh slate with these models where they're not assuming anything about my, my preferences or my tendencies. So, I think it goes both ways. The personal copilot, you kind of do want it to have some sense of memory. But then, like, your automation tools, maybe they don't need the same level of personal life kind of, uh, understanding.

[17:58] Reid Hoffman: Mm-hmm.

[17:58] Parth Patil: And then my everyday primary copilot is ChatGPT.

[18:01] Reid Hoffman: Mm-hmm.

[18:01] Parth Patil: Um, obviously because it's got a great mobile app. You can point your phone at things and talk to it about the real world, and use it to help you solve problems every day. Like, even just like, "Oh, how should I organize my, my, like, apartment?" Like, I obviously should buy some containers to organize my, my, my closet. It's very good for, like, the in-person real-world kind of AI, and really good for research.

[18:21] Reid Hoffman: Mm-hmm.

[18:22] Parth Patil: ChatGPT pro mode, um, is the best research tool that I've ever seen and it's, it continues to be the case. And then I would say after ChatGBT, it's I've got my coding agents. So I usually have three coding agents-

[18:35] Reid Hoffman: Mm-hmm.

[18:35] Parth Patil: ... assigned to almost anything that I care about in my life. (laughs)

[18:37] Reid Hoffman: (laughs)

[18:37] Parth Patil: Just like this. Yeah. I have a Claude, a Codex and a Gemini, and they're just attacking three... And this is just one project. I have like, you know, seven, eight other projects where I'm just spawning more, um, more agents in different directions at the same time. So I would say the coding agents are more like my ambient fleet.

[18:54] Reid Hoffman: Mm-hmm.

[18:55] Parth Patil: Where anything I care about has one to three agents on it. Any project that we're working on, Read AI, some of these creative projects or, or when we're building something new, the first thing I do is I create a folder and I put Claude, OpenAI's Codex and Google's Gemini into that project. And I just send them in three different directions on that project in an empty folder. And I kind of transcribe to them and I tell them what I want them to do, and then they kind of just work on that in the background. And, and then I spin up another project and I have three more, um, on that project. And so I'm at this point now where it's like anytime I have a new idea, the f- my instinct is to put three agents on it and have them make some progress before I come back to it.

[19:32] Reid Hoffman: What do you think is the maximum number of projects you've spun up with your trio of agents?

[19:37] Parth Patil: This fleet, I've gotten up to like 17 projects where they have on average two to three agents each one.

[19:43] Reid Hoffman: Mm-hmm.

[19:43] Parth Patil: I think this is just a current constraint and it's something that took me like three months, the last three months I kind of designed my workspace to allow for this kind of context switching because then it's, it's like when, especially when if you look at the frontier models now, GPT-5.2 and Opus 4.5, you can set them up if you tell them, if you give them a good approach to planning and you say, "Make sure you write down the plan and periodically update your status in the plan." You can say, "Go work on this for today" and it will continue to work in a loop for a whole day even longer than that.

[20:14] Reid Hoffman: Mm-hmm.

[20:15] Parth Patil: And then people are like, "Oh, how, like why would you want multiple agents?" And I was like, "Because when you tell someone to work on something, like you're not gonna just walk away from the computer and like go to the beach." You're, you're like, "No, actually I wanna fire off another one." You know, like you then, then the justification for having kind of like a small fleet, a couple different, uh, directions going at the same time makes, makes perfect sense to me. And I think that as we get better at like the user interface constraints and the context switching, I think the code review and how many of these can you manage is really a question. I think a lot of people are still in the single co-pilot phase, and I think right now is probably the time to move into the, how many of these can you orchestrate at the same time across a range of different, uh, projects?

[20:58] Reid Hoffman: Like the trio of agents. Awesome. We'll get to what people should be doing with one agent in a second.

[21:03] Parth Patil: Yep.

[21:03] Reid Hoffman: But what's the funniest thing or the funniest thing that comes to mind about what's gone wrong with one of your trio of agents projects?

[21:11] Parth Patil: Yeah. So when we first met like two years ago, I, I was thinking about this problem, an earlier version of this problem of like, can I put two chatbots in a, in a room and then give an objective and then like, can they make progress and then walk away from my computer? The first time I did that, I realized that they, you know, they would work on a problem and then they would not know when to end the conversation. And so they would just say thank you to each other in an endless loop. And I came back and I spent like a couple hundred dollars on the word thank you. And I was like, "Oh my God, what are we doing here?" (laughs) Do I need to create a manager to say end the conversation? Which is what I did. Like, it felt like a naive thing, but I was like, makes sense to have a third person in the room that just ends the meeting, right? Um, and then, and so, but that, that kind of clued me into this idea that maybe the, the coordination across multiple chatbots with their own context windows is like largely kind of unsolved territory. And now we're in that same phase. Uh, we're, we're in the next version of that same problem, which is they're not just chatbots, but now they can take actions, they can work on projects. And I tried it again, right? I said, "Okay, what if I had the, gave them the ability to talk to each other in a, in or like send messages to each other, right?" They can send messages to each other or they can see where each other are in a project. And it was experimental, it was basically allowed three different coding agents to DM each other while they worked.

[22:25] Reid Hoffman: Hmm.

[22:25] Parth Patil: And I realized, one, they kind of missed timing a lot. So like, they'll be working, but they, they'll get a message and then they'll reply to the message, by that time the other agent's already like halfway through the problem. So there's something about like timing. And I also realize that, uh, it opens interesting questions as to like permissions, because you'll have an agent and it'll be working on a problem and it might be like, "Oh, user, I need some, I need you to approve me using this tool." And me being a human, I can give that approval. But then if it starts asking the other agents for approval, it's like, "Oh, hey Codex, I need to use, uh, this analytics tool. What do you think?" And then Codex is very willing to just approve that access. And I was like, "Okay, maybe this breaks our definition of a sandbox if they can just ask each other the permission themselves."

[23:07] Reid Hoffman: Is it okay if I erase the hard drive? Yes, absolutely. (laughs)

[23:10] Parth Patil: Yeah, yeah exactly. That's the, that's the risk there, right? So, so then I was like, "Okay, if we're gonna do that kind of experiment, we should be sandboxing it. We should run it on a machine where we don't care if we lose everything, right? Like a virtual machine, um, with a docker container." But it is interesting because I think we're gonna get there where like I want them to be able... The only, like how I deal with that now is I just send them in different directions. I say like, "You're gonna research, you look for bugs and then you help me with like the blog and my website." And then I know like their work doesn't overlap and so there's less likely to have, we're not gonna have this like collision of, of like different workers. And I'm sure that something similar to what GitHub did for humans-

[23:45] Reid Hoffman: (laughs)

[23:45] Parth Patil: ... is going to be, there's gonna be something, some similar kind of solution for coordinating multiple agents working together on the same project. Then, uh, when I want to deploy software to the web, I wanna make an app and then share with people, I use Replit. So Replit is another coding agent, but, um, and, and these, these three agents will work with the Replit agent on a project. Like my website is built and hosted in Replit. And these Codex and Cloud Code have helped with the website, but it is, it is largely Replit's agent has built the, the, the website. And so that's really good because if you want deploy software, there's no easier way than Replit. And then for image models, I really like, um, Nano Banana from, uh, from, from Google. Nano Banana Pro from Google is the most powerful image model.And it's good for... Y- it's not just about generating cool images of, like, surreal things, but it's also very good at building, like, designing infographics-

[24:37] Reid Hoffman: Mm-hmm.

[24:37] Parth Patil: ... that are extremely coherent. And, um, Ethan Mollick, he thinks that, he thinks that Nano Banana is kind of like a successor to PowerPoint-

[24:44] Reid Hoffman: Mm-hmm.

[24:44] Parth Patil: ... in, in the sense that it can create these rich visuals with coherent text and layout. Um, I think, I think it's some combination of code and, and generative images, like Nano Banana-

[24:55] Reid Hoffman: Mm-hmm.

[24:55] Parth Patil: ... is probably the evolution of presentation tooling. I like Nano Banana. I think other image models I like, Flux. Flux from the Black Forest Labs team, um, it's a, it's kind of an open source model. Um, I also like, of course, Midjourney is, I mean, it's best in class in, in its own category. For video models, I really like Sora and VO3 from Google, so Sora from OpenAI and VO3 from Google. These are the most powerful video models I've seen, and, um, so I u- I use them for animation pipelines. So I'll take a coding agent and I'll give it an image model, and then I'll give it a video model, and I say, "Help me animate this world."

[25:29] Reid Hoffman: Mm-hmm.

[25:29] Parth Patil: And it'll start creating scenes and characters and animating them. The intersection of coding agents and all these other creative tooling is also a very interesting-

[25:37] Reid Hoffman: Mm-hmm.

[25:37] Parth Patil: It's a new form of animation.

[25:38] Reid Hoffman: Mm-hmm.

[25:38] Parth Patil: New form of, like, world-building and storytelling.

[25:40] Reid Hoffman: Some of our folks are probably listening to your description and thinking, "Oh my god, sci-fi," like, you know-

[25:47] Parth Patil: Yeah.

[25:47] Reid Hoffman: ... William Gibson, the future's already here, it's just unevenly distributed. But like, if someone said, "Okay, I'm gonna start using an agent."

[25:53] Parth Patil: Yeah.

[25:54] Reid Hoffman: What would... You would go, "Okay, start using an agent thus."

[25:58] Parth Patil: I would say start with ChatGPT, and maybe I would say start with, like, the ChatGPT Atlas browser. If I type "/agentmode", now we're in agent mode. And so ChatGPT is a chatbot, but I can say, "Let's go to the Tech Meme front page and look at some of the headlines, and then maybe pick an article and then open it up." I've transcribed my prompt and I'm gonna send this message, and now ChatGPT is gonna take control over the browser and then navigate to the website and explore the front page. Looks like there's an OpenAI article right there. And you can see on the right side, it's thinking about brow- how to browse the website.

[26:30] Reid Hoffman: Mm-hmm.

[26:30] Parth Patil: And it's clicking on the first ar- it's probably gonna click on the first article. It's reading this article. Um, let's surf, let's surf Wikipedia, and I wanna learn about language models, so let's go explore Wikipedia.

[26:40] Reid Hoffman: And by the way, for those people who don't have the Path command line, the agent mode's also available on the button.

[26:46] Parth Patil: Yeah, there is a button there. (laughs)

[26:47] Reid Hoffman: (laughs) Just, you know, it's all cool, but it's just, you know, making sure that-

[26:50] Parth Patil: I, I've gotten so, I've gotten so hot key-oriented-

[26:52] Reid Hoffman: Yeah.

[26:52] Parth Patil: ... that I'm like, I won't even use the trackpad.

[26:54] Reid Hoffman: Yeah.

[26:54] Parth Patil: And so right now, we're watching ChatGPT use a web browser.

[26:58] Reid Hoffman: Mm-hmm.

[26:58] Parth Patil: And I think this is, like, uh, pretty, pretty g- I mean, I think if you're gonna, if you're just getting into what is an agent and you've, maybe you've talked to ChatGPT or a language model before, this is a version of an agent that can take actions, right? Without me clicking around, it took us to Wikipedia. You could tell it to book a flight. You could tell it to, to research a topic. And you can fire off more. You can open up more tabs and then there are, have more agent mode queries running. And on the side, it's explaining we can talk to the, we can talk to ChatGPT-

[27:24] Reid Hoffman: Mm-hmm.

[27:24] Parth Patil: ... about the contents of the page. So I think that ChatGPT, as an agent, is probably the best everyday agent.

[27:30] Reid Hoffman: Mm-hmm.

[27:30] Parth Patil: There's so many more capabilities, uh, but I think for research, it's probably one of the best. And also, just for personal, everyday, kind of exploring the web.

[27:38] Reid Hoffman: Mm-hmm.

[27:38] Parth Patil: Everything we were u- we were using the web browser for is now, you have a very intelligent language model that can help you explore pretty much any topic-

[27:46] Reid Hoffman: Mm-hmm.

[27:46] Parth Patil: ... learn anything, teach yourself anything. And I think that's the... The, the most powerful thing you can do with AI is have it teach you about how the world works.

[27:54] Reid Hoffman: So we've covered some of the entry points. Now, what is it... One of the things that, as you know, one of the ways that I describe you, uh, off the matrix is, not only have you taken the red pill, but you're bathing in the red pill.

[28:06] Parth Patil: (laughs) Yeah.

[28:07] Reid Hoffman: And as a question that's kind of in the, um-

[28:09] Parth Patil: The Morpheus, yeah.

[28:10] Reid Hoffman: Yes, the Morpheus thing.

[28:11] Parth Patil: Yeah.

[28:11] Reid Hoffman: What does it, you know, look like for an individual to cross the line from, you know, kinda asking questions and using ChatGPT as a search, you know, engine or kind of like a, give me my, my research Wikipedia answer, et cetera, but to building agenic systems, automations around themselves-

[28:30] Parth Patil: Mm-hmm.

[28:30] Reid Hoffman: ... and in particular, share an example from what you do.

[28:34] Parth Patil: So I think, like, we talked about ChatGPT, that's a great, like, intro to language models and agents. Um, I think that the, if you wanna go one step further, I think you want... Uh, th- there, there are limitations to this chat experience in a web browser, where you, you actually... The, the real power of the language model is un- unlocked when you give it a computer and you allow it to kind of use a computer with you, use your own computer, and, like, working with files on your computer. And for me, one of the, th- the first things I realized I needed, I was like, "I should have a personal website. I should start talking about these things." I thought, "Well, AI is gonna help me build my website." And I've never done this before. It was, I was like, "I need to build a web... I, I want to build my first website from scratch."

[29:12] Reid Hoffman: Mm-hmm.

[29:12] Parth Patil: And, uh, so I went to, I went to AI and I said, "Let's build a website." And now, I think it's like, these coding agents that built the first version of my website, now they run my website. And so I'll just show you. So if we look at m- my screen right now-

[29:26] Reid Hoffman: Mm-hmm.

[29:26] Parth Patil: ... we see three panes, and this is kind of in the, the, the space of, like, the Morpheus kind of thing.

[29:30] Reid Hoffman: Yeah.

[29:30] Parth Patil: So in the first pane on the left, I'm gonna launch Claude. Claude... And this is Claude Code. So this is Anthropic's Claude, but running o- on the cloud-

[29:38] Reid Hoffman: Mm-hmm.

[29:38] Parth Patil: ... but working on my computer. In the middle one, I'm gonna launch Codex. Codex from OpenAI. And on the one on the right, I'm gonna launch Gemini. And all three of these agents, so we see Claude, Codex, and Gemini, all three of them are actually working in the folder which has my website code.

[29:53] Reid Hoffman: Mm-hmm.

[29:54] Parth Patil: So I'm gonna fire off three different tasks. Gemini's got the longest context window.

[29:58] Reid Hoffman: Mm-hmm.

[29:58] Parth Patil: So I'll say, "Gemini, read every single blog post that I've written and suggest the next three topics we should cover to help people who are getting into working with AI agents and language models, um, discover this kind of value that we, we've found over the last couple years." So transcribe the prompt, send. Now Gemini is gonna read all the blog posts that I have. Um, we'll go to Codex. "Codex, can you please pull all of the website traffic analytics and suggest improvements and next steps for improving the performance of the website, both from a performance on, like, maybe...... engagement on the website as well as maybe on the content side of things. And then we're gonna have Claude. Claude, I need you to open my website and then, uh, we should take a look at how it looks like on a mobile experience. And then I want you to browse it as if you are someone visiting my website for the first time, someone that wants to learn about AI, wants to learn about working with coding agents. Pretend you are that person on their phone exploring my website, and let's, give me feedback on the website to improve the experience.

[30:58] Reid Hoffman: And here we've kinda demonstrated three different agents in the, not working the same thing, but kind of given different, different... Same file, but different tasks.

[31:06] Parth Patil: Different objectives, that's right.

[31:07] Reid Hoffman: Different objectives, so they're not combining. And then, you know, a little bit of what's implicit is which ones do you think will be, you know, a little better at each task.

[31:15] Parth Patil: Exactly. And if we look at it, in the middle we have Codex is pulling analytics numbers across the website. It has access to the analytics.

[31:22] Reid Hoffman: Yeah.

[31:22] Parth Patil: So me being a data analyst, the first thing I figured out was, like, it's really good at analytics. Now we have Claude, has pulled up my website-

[31:28] Reid Hoffman: (laughs)

[31:28] Parth Patil: ... in, uh, Google Chrome, and it should... The next thing it'll do, resize it for mobile.

[31:32] Reid Hoffman: (laughs)

[31:32] Parth Patil: And now it's going to... And you can see how it's thinking. "I'll open your website, let me resize it," and the next thing it's gonna, it's gonna explore the website like it's a user. And, uh, we also have Gemini behind this that's just reading every single blog post. Look, it's giving us feedback. Nice hero section. Good content cards. Let me continue scrolling.

[31:46] Reid Hoffman: Mm-hmm.

[31:47] Parth Patil: Meanwhile, we have Codex that's running analytics. So here it's reading a, a blog post I wrote about Glod- Claude code. Yeah. So this is, I think the... Building a personal website using many, multiple agents is very, obviously useful for, like-

[31:58] Reid Hoffman: Yeah.

[31:58] Parth Patil: ... most solopreneurs.

[32:00] Reid Hoffman: Yeah.

[32:00] Parth Patil: Um, and it's a single website. So it, it, it's, uh, a single person made the website. Like, I made it with AI. I wouldn't even know how to do this without AI, let's be honest.

[32:08] Reid Hoffman: Mm-hmm.

[32:08] Parth Patil: Like, you, you go back three years and it's like, "Okay, I'm gonna learn-"

[32:12] Reid Hoffman: Oh, yeah.

[32:12] Parth Patil: "... every single programming language and then figure out how to stitch it all together." It would take, it would take months.

[32:16] Reid Hoffman: Yeah.

[32:16] Parth Patil: I would probably... This, usually you have to hire people to do that. And then the quality is also higher than I could ever imagine because the AI is so good. And I can also aim it at things and say, "I like this website. Let's, let's emulate that kind of..."

[32:28] Reid Hoffman: Yeah. So, hopefully one of the things, you know, our various listeners have picked up here is the scope of, like, like you're just throwing darts where you have a complete set of things on each of these different directions. Um, what's... Who are some of the sources, whether it's podcasts, social media feeds, et cetera, of people that you pay attention to to learn more about prompting?

[32:52] Parth Patil: Oh. For prompting, I think, uh, there's this, uh, Dexter Hoorthy. Um, he, uh... Great AI engineer. Um, he... I met him earlier this year. We were talking... We were, we, we met in a group. It was kind of like a Claude Code anonymous group.

[33:08] Reid Hoffman: (laughs)

[33:08] Parth Patil: It was, the, the requirement to get in the group was to be addicted to Claude Code.

[33:12] Reid Hoffman: (laughs)

[33:12] Parth Patil: And everyone was kind of just like, we were sharing how much we were using it and how we were using it. There's all these different techniques for, like, running many of these at the same time. And so Dexter, he... Like, we, we had some very interesting dialogue on, on, like, how to orchestrate many agents. But then he actually kind of, uh, really expanded on the idea of context engineering, where there's prompt engineering, which is, like, maybe came into the play, like, in the ChatGBT 2023 phase. But now context engineering is, like, actually, I think, the better way to think about it. It's like, what are the various techniques we have for bringing the right context into the model and making sure it doesn't have the wrong context? So we're not wasting its, its, its cognitive bandwidth on the wrong context.

[33:52] Reid Hoffman: Mm.

[33:52] Parth Patil: And how do we... Like, our job as, as AI engineers is to think about the, the context window as a, as a canvas.

[34:00] Reid Hoffman: Mm.

[34:00] Parth Patil: And we're, like, filling the canvas with the most relevant context, whether that's images, whether that's code examples, whether that's, like, tools that it can use, without cluttering its mind and then giving it an objective and hoping that it is the right mix of information that will allow the model to do the right job.

[34:15] Reid Hoffman: Mm.

[34:16] Parth Patil: And so he thinks a lot about context engineering, and I've learned a lot from the way he thinks about it. Even in how we use coding agents and, like... You know, coding agents, you can't give one of these coding agents unlimited tools.

[34:28] Reid Hoffman: Mm.

[34:28] Parth Patil: Like, not yet, right? They don't have unlimited cognitive bandwidth.

[34:31] Reid Hoffman: Mm-hmm.

[34:31] Parth Patil: But then there are ways to give them tools where they can use a broad set of tools without having to memorize every single one upfront.

[34:38] Reid Hoffman: Mm, mm.

[34:38] Parth Patil: Where, um, I think the term uses, uh, what is it? Progressive disclosure. So if you give it a t- if you give a, create a tool and the tool has a help guide on how to use the tool, then it, you don't need to explain the tool upfront to the agent. The agent just needs to read the guide.

[34:53] Reid Hoffman: Mm.

[34:53] Parth Patil: And only when it needs to-

[34:55] Reid Hoffman: Mm.

[34:55] Parth Patil: ... use that tool. So he kind of, uh, helped me think about like, "Okay, how do we, how do we design tools for our coding agents so that they're not constantly getting overwhelmed by all the things we want them to do, but they do a really good job at the few things we want them to do?" So in this case, the Claude agent has access to a web browser, and it has, like, access to some of my personal knowledge. And so it's able to, like, use the web browser, but the other agents aren't using the web browser like that. And so they're more free to think about the content and the strategy. And on the creative side of things, I think there's Don Allen. Um, he's... I've known him since high school. He's one of the m- He's one of the most prolific AI creators. Someone who's able to weave, like, these, wield these, like, creative models, the image models, the image models, the video models. Dave Clark also extremely good at visual storytelling. I think, um, the, I put them in... Nem Perez. I'd say I put these three in the category of the new Hollywood-

[35:45] Reid Hoffman: Mm.

[35:46] Parth Patil: ... where it's like, um, what if the studio is in your pocket, a couple of these models is helping you tell your story? And so they're very good at, like, animating short stories and creating, like, trailers. And now they have, they have their own studios that they're working on and, and trying to, like, discover the new workflows of the entertainment industry and e- entertainment space. I think on the... If I think about who's the ultimate, like, creator entrepreneur, you talk about, like, treat your, um, treat your life, like, your, your life is a startup, right? Like, th- the startup of you. I think, uh, top of that list for me is Kat GPT.

[36:16] Reid Hoffman: Mm.

[36:17] Parth Patil: She is, like, a solopreneur creator. She's built a large audience through talking about AI and playing with the tools, kind of like we do, very publicly, but also making it make sense to almost anyone. So if you follow Kat, you're just gonna learn a lot about AI in a very non-judgmental, everyday useful kind of way. And she's leaned into the more technical side of things. She's picked up vibe coding, and she's starting to use those capabilities to launch businesses and small teams working on, like, highly lucrative projects more efficiently than ever before because now you have the, like, amplified creator entrepreneur that's coming online. So she has the media piece, extremely good at video, extremely good at media.... and extremely good at that distribution side of things. Built an audience, became well-known, and also is now, like, building products for that audience. And I think that's a very exciting, uh, person to watch. So I think of all of these people as AI native people, and like-

[37:05] Reid Hoffman: Mm-hmm.

[37:05] Parth Patil: ... these are the people, uh, that I'm constantly kind of following, trying to understand where we are, what can be done. Um, of course, Andrej Karpathy, who's like... I mean, I mean, I, I think he's, he's kind of like that, that person who when he speaks every time it's like, "Oh, what does he think about coding agents?" He's like, "Ah, they're kind of slop." It's like, "Huh, maybe they are slop." Like... (laughs) And then you start thinking like, yeah, it's magical from my perspective, but then you take an expert and you think about like... He's thinking about the jaggedness of intelligence and like-

[37:33] Reid Hoffman: Mm-hmm.

[37:34] Parth Patil: ... how, you know, it's so good at these things and entirely, like, useless in other lenses, right? And, um, thinking about that more on a, on a very deep level of, uh, where does that... Where, where do language models go? Where are their limits? So I, I think Andrej Karpathy is probably one of the best people, uh, to follow for AI researchers.

[37:50] Reid Hoffman: This is, I think, the path that we're all on, whether or not we know or not, is, how do we become AI native?

[37:55] Parth Patil: Yeah.

[37:55] Reid Hoffman: So what is... Like, what should an individual consider doing to, you know, start making their life more magical? Um, we've got a bunch of different prompt ideas, whole bunch of things, but like, what are the things that, you know, you who, you know, basically ask yourself every single thing you trip across as, "Can I have AI amplify or do that?" What would you say for individuals to kinda start making, like to, to, to seriously engage in doing, to s- to experiment with making their life magical with AI?

[38:31] Parth Patil: Yeah. I think about this a lot. Um, for me, I would say, I would say that the main thing is that you should apply AI to something that you're intrinsically motivated by, like your passions, your interests. I think there's a, there's like, a default kind of approach, which is like, "I gotta use this for my job. I gotta become more productive." And that's fine, but I think what's more interesting is using it to expand your sense of self. Like, who- whoever you were before these tools came online, I promise you, like, you're much more than that once you start interfacing these tools. You start expressing yourself through these tools. Right now, I think of myself as a visual storyteller. I think of myself as like animating, like animating worlds and creating worlds, right? Even though I was a data analyst maybe just three years ago, now I'm also an engineer. I'm like a vibe coder, right? Or, or what you might call it. And I think it's like the expansion of the sense of self is... It's like you could think of it as like a replacing of what you were, but I think the other end of it is like an expansion of your sense of self. And if you aim this at your passions and your interests, so in my case, like music, games, uh, visual storytelling, I think that allows me... Then it just like, no one's gonna tell you to do it. You're gonna see the magic. You're gonna, you're gonna discover like the, the upside because it's something that you didn't think you had. Like, I think of it as like, I get to live all these other lifetimes. I get to be all these other things that I kind of like sidelined in favor of my career, but now I'm expanding back into them, coming back into them in a different angle, right? So... And, and it's an intersection of so many of my interests of both technology and creative and computers. And, and I think that a lot of people will experience something similar, which is the expanded sense of self through, um, this kind of like technology.

[40:12] Reid Hoffman: I think that people don't realize that with AI, they need to re-expand their imaginations-

[40:17] Parth Patil: Yeah.

[40:18] Reid Hoffman: ... for a sense of self, for a sense of capability.

[40:20] Parth Patil: Yeah. And you'll be surprised at how much more ambitious you become when you see-

[40:24] Reid Hoffman: Yeah.

[40:24] Parth Patil: ... what you can do. It's not... You're not going to just generate an image and like call it a day. You're gonna be like, "Whoa. Actually, let's create a world around this, maybe a story around this." And it, and it becomes a bigger kind of, uh, ambitious, uh, pursuit.

[40:35] Reid Hoffman: Possible is produced by Palette Media. It's hosted by Aria Finger and me, Reid Hoffman. Our showrunner is Shaun Young. Possible is produced by Tenasi Delos, Katy Sanders, Spencer Strasmore, Imo Zu, Trent Barboza, and Tafadzwa Nemarundwe.

[40:50] Credits: Special thanks to Surya Yalamanchili, Vida Sepieva, Ian Alice, Greg Beato, Parth Patil, and Ben Relles.