Lenny's Podcast: Product | Career | Growth
00:16:02 · Tara Seshan
[00:16:02] S0: Yeah, I think there's two aspects to this. One is continuing to work with agents at higher and higher levels of abstraction. So letting the agent do more and more for you independently, coming in, providing that steering, and then letting the agent continue to cook. Let the agent cook and provide details at higher of orders of abstraction feels like the way. People are increasingly thinking about agents that are persistent, that feel like teammates, that feel like coworkers, where you can work with them. The way I might work with someone on my team, which is they do a whole bunch of work, I provide input, and then they do work again. And we sync up at different cadences, look at each other's in progress work and provide more and more feedback. It feels like that coworker model is the way that things are certainly going. It feels like a much more natural interface for us to be able to work with agents, and we already see a lot of that internally as well. The second is that a lot of my work with agents thus far has been one-on-one. I work with my agent, maybe it's spawned some subagents to get some tasks done, but it's me and my agent together, and that is potentially divorced from what my colleagues are doing with their agents. And so there was a time where everyone internally was just sending their Codex threads, screenshots of their Codex threads to each other on Slack. And we're like, "Okay, well, I wanted to share with you how I got to this number. Here's how I got to this number. Here's a screenshot of what I did." But that's also not quite the most natural way for someone to collaborate together. And so as more and more work gets done with our agents, shouldn't we be able to get work done with our agents together? And what is the most natural interface to make that happen? And those are some of the things that we're thinking about.
00:17:50 · Lenny Rachitsky, Tara Seshan
[00:17:50] S1: Chat is what I'm picturing. That makes so much sense. It's like, okay, here's Tara's agent, here's my agent. She did some work on some analysis. I'd be, "Hey, my agent, Lenny's agent, go check, make sure this is legit and connects to the way I think about the world." [00:18:04] S0: Ideally, work feels like a multiplayer game where all of us together are getting stuff done, steering our agents as our agents continue to take care of more and more of those rowing tactical tasks. [00:18:15] S1: It's interesting how it's just been this slow progression of trust and just awareness that this can be how we work. Just this, "Okay, go work for longer, you can take on more." There's been this talk of the slow takeoff, the fast takeoff scenarios, and everyone's afraid of this fast AI takeoff where it's way too smart and now we're in big trouble. It feels very much like we're on the slow takeoff scenario, which is good, where it's just slowly iterating. It doesn't feel that slow, but in a sense, we're not to some 300 IQ AI.
00:18:47 · Tara Seshan
[00:18:47] S0: I mean, the models are incredibly smart. But I think a lot of the things that have enabled us to then work with our agents together or have the agents take care of higher and higher order abstraction things certainly are about the intelligence, their ability to perform long-running tasks, and how long they can stay on task. But also actually, there are very meat and potatoes, tactical things that make this possible. Agents working locally are really convenient because they have access to all the data that's on your machine. To make an agent successful in the cloud, there is a ton of cloud infrastructure that you have to build to make that possible, and just access to your systems. How can agents talk to all these third-party systems that have all of your data? Just like a colleague who you hire, who you lock into a room, never give them access to Google Docs and Slack, and I don't know, the company database would not be that useful to you. Similarly, a cloud agent that is similarly isolated will not be that effective. And so a huge part of making these agents useful and achieving some of these futures are on the intelligence side, certainly, but a lot of it is also just really tactical data access, cloud infrastructure and reliability pieces that feel much more prosaic than some of the broader intelligence questions but matter in some ways just as much for the end effectiveness.
00:20:05 · Lenny Rachitsky
[00:20:05] S1: This touches on something else that has been coming up a bunch on this podcast, this word ambition. I know you think a lot about this too. It feels like not only are we able to be more ambitious because of these AI tools, we almost need to be more ambitious, which is not natural for a lot of people, because everybody can now do all these easy things really easily. The easy stuff is super easy. The hard stuff is easy. And the thing that separates people now and companies now is just how ambitious they can be. Talk about what comes up when I talk about the need and the emergence of this need for ambition.
00:20:42 · Tara Seshan
[00:20:42] S0: Yeah, I think the people that we see who are most effective at using AI tools don't simply use it to automate rote tasks, but use it to expand the set of things that they are capable of doing. Back in the day, before all this AI stuff, the unicorn person was someone who was a really thoughtful product sense person who also happened to be an engineer who may also have been a designer. That person was always the unicorn hire because they were able to really flatten the layers of translation needed between all of these functions and were able to build something or ideate something really quickly and easily themselves and get it up and running, and then were able to work with a team and collaborate with a team on it. And I think the most compelling thing I've found that certainly I try to be able to do with these tools and I've seen some of my most successful colleagues be able to do with these tools is really expand the set of things that are "within their range of possibilities" so that they can start realizing more and more of what's in their head into the reality, the way that someone who was previously jack of all trades was able to do. We kind of all have that superpower now that I can spin up a set of designs on something and I can go build an initial prototype of it and I can figure out the right pricing model for it and model out all the scenarios. Really, the set of possibilities have widened dramatically. And actually what that means in so many ways is that I have the ability to be, to that point earlier about film, be more of an auteur as I try to get something done and realize my vision maybe to higher fidelity. And that to me is part of what can elevate your ambitions while pursuing new ideas and new products, that because all of these things are now within reach, because this new set of capabilities is now within your reach to be able to try and access, you're not really limited. Your ambitions are no longer limited by what you're capable of executing yourself, what you're capable of communicating. It can be so much, so much wider. I think the hardest part about doing this is simply just expanding your thinking. Actually, the capabilities have expanded so dramatically. It is really expanding your thinking of what's possible in an unreasonably short timeframe.
00:22:56 · Tara Seshan, Lenny Rachitsky
And to me, the best way of trying to do that is Patrick Collison has on his website, patrickcallison.com/fast, I think, which is all of these projects that were unreasonably ambitious that were executed in a really, really short time period. And what, for me, is now remarkable about that list of projects is that they all existed before these tools made it possible for you to learn how to build something almost instantly, or ask it with one question, "Hey, can you summarize this very complicated text or this very complicated book for me immediately?" Or can I try to do all of these things that were previously impossible to me, but now I'm able to do? "Can you make for me a CAD model of this idea that I might have?" Really, capabilities that were truly beyond my reach are now in my reach. And so if those fast projects were possible before with the capabilities we used to have, shouldn't we just see an exponential increase of the number of those types of unreasonably quickly and effectively executed things with what AI has given us? [00:23:57] S1: To your point, the hardest part is just remembering to even to try just to be like, "Oh yeah, well let me see if Codex can do this for me." It's just a new habit, a new thing we have to build in our brain.