Airtable CEO: This Is What the Top 1% Do With AI | Howie Liu — Silicon Valley Girl Podcast

Marina Mogilko: I'm the bottleneck because all the content goes through me. What if I have an agent that already knows my taste and I say you just let it give the feedback.

Howie Liu: If you could hire as many people as you wanted for almost zero cost, what other roles would you hire people into?

Marina Mogilko: Hi Howie built one of the biggest software companies on the planet. Airtable is used by more than 80% of the Fortune 100.

Howie Liu: The people who are able to do that most effectively will become almost superhuman. You're able to give good judgment and feedback to a really effective CEO of humans. You could imagine building a company that you never would have dreamed of without any employees or with minimal employees. So the tinker mindset is most important first and foremost. And I think the second is really important.

Marina Mogilko: My goal is that anyone who has been consuming content about AI but was not scared to try it but thought it was too complicated, I want them to actually deploy a few agents after watching this video and realize how powerful the technology has become compared to two years ago when they maybe tried one prompt and ChatGPT and thought this is just nothing, all the hype, I don't understand it.

Howie Liu: When you work with this technology correctly, it's actually really transformative. Can you talk to me about the current state of AI? So we've been in an era of chatbots. You had to initiate the process, correct them, and then the output. You wouldn't really trust an agent to post for you on LinkedIn.

Marina Mogilko: I feel like right now we're trying to close the loop with these agents. Teach them our taste, teach them how we give feedback, and make them more autonomous.

Howie Liu: Is that what you're saying too? I do think there's been a step function change in the experience that you can create with AI from pre-chatbot, but then I think we entered a mature era of even AI chatbots where the models got smarter. Even thinking about the year after GPT-4 came out, we could actually do really interesting things with chatbots. So chatbot maturation and then I think we started to hype agents a little too early. So there was a year like last year people were saying this is the year of the agents and I think that was when Salesforce Mark Benioff went up and was talking about how in the future every CEO is going to have thousands of agents like agent employees and human employees and I think it was correct but just early. What we finally hit now, I think, and some would say it was around the end of last year with the Anthropic Opus 4.5 and beyond generation of models, but now you have GPT-5.5 and Claude 2.6, these open source and alternative models that I think have all reached this level of intelligence that actually enable agents to be almost humanlike and autonomous.

Marina Mogilko: Do you when you see a post on X where somebody says I sleep at night and my company is being run by 50 agents, do you believe that?

Howie Liu: I don't know. I've deployed a lot of agents but I can't sleep. It's still in my head. I'm responsible for that agent. I'd rather have more humans. But do you think we're moving towards that reality?

Marina Mogilko: You know, I think that as the models get better and better, we do need to rethink the entire paradigm and the product form factor to reflect it. So I'll use a parallel here, which is in the development agents world. You had GitHub Copilot that was the first chatbot for developers, and it really was that you couldn't have it go and develop really complex programs. It was more like you could autocomplete a few lines of code at a time. And so it was a convenience. Then you had Cursor which introduced its own AI agent called Composer, which is different from what they're using that name for now, the model they've fine-tuned, but at the time it was a much more autonomous agent. You could tell it to go and build me this file, write me this simple script or function and it would do something a lot more autonomously than the original Copilot experience. I think now we've entered into a new era where the agents do enable you to go and execute in parallel with multiple agents. So in the development world the best developers are already not using necessarily Cursor, they're actually now using many different agents, maybe it's Claude Code for instance, running in parallel and it almost does feel like they have their own company or team running, including overnight. I actually when I'm developing I use agents that I try to have them do something substantial before I go to sleep so that for hours at least I'm not wasting bandwidth cycles and they're doing something useful. And so when I wake up they've already completed. So I think we are moving to a world where the form factor will start to look more like managing a fleet of agents or almost managing people. Everybody when they transition from being an individual contributor to becoming a team manager, it's a different role. You still need some of the same techniques but also it's quite different. So I think we are moving in that direction. But to your point, I think it's not yet at a point where you literally just have the agents do everything. Like I don't think agents can effectively run an entire company or any substantial company autonomously.

Howie Liu: When do you think this flip is going to happen? Do you think by the end of this year we're going to see more output that's actually deployed right away by agents?

Marina Mogilko: Picture this. You have something clear in your head—how a banner, a logo, a thumbnail should look—and you're building a personal brand and you need fast execution. Design.com is a perfect tool for that. You describe what you want, the AI builds it. You refine in chat, it adjusts on the spot. Logo, website, banners, social graphics, business cards, presentations, you can use the same tool for all of this. Say if you're growing on LinkedIn, and a lot of people say that video banners actually work best right now. Design.com can help you with that. I'm seeing more and more people who post graphics and it's hard to stay consistent if you have to redesign everything from scratch every time. Or maybe you need a logo for your personal brand. Design.com is a perfect tool for all of that. You describe what you want, you get something usable in a couple of minutes. If you want to change the layout, swap a color, try a different font, you just ask in chat, it updates. No revision cycles, no delay. Same flow for Instagram posts, stories, YouTube thumbnails, anything you're posting that week. Set your brand once, the platform carries it across every new asset. Over 400,000 logos built so far. A million designs total, and what you make is yours to use commercially. If design has been the thing you've always wanted to work on but just couldn't, try Design.com yourself. Link is in the description. It's free to start.

Howie Liu: I think that on analytics, the fact that agents can help close the loop with maybe drafting some content, there's still a human review step. You're posting it and then maybe the agents are helping to analyze what's working, what's not working, and seeing what's catching fire out there from other people in the content world. I think it does create this really high leverage circuit for people who know how to effectively deploy agents. Even if humans are still in the loop at points, I think the amount of the loop or the percentage of the loop that is more agent-driven is going to increase and increase and increase until at some point it's almost like yeah, there's a human, but that human is extremely leveraged and you could imagine building a company that you never would have dreamed of without any employees or with minimal employees. Now you can do that with agents.

Marina Mogilko: Absolutely. And I see it with my content. We haven't really expanded the team. We're still hiring but we are producing I think three times more content just because it's so much easier to generate that content. Of course we have to check and everything but one person can now handle SEO and newsletter and also threads and another person can do three other outlets and maybe even researching ideas. So all the substantive work that goes into going deep on a topic, a lot of that can be enabled by agents. So for someone who's excited about all of this but also they don't know, I am also confused. Every week there's something new—there's Claude with Code, oh my god, Perplexity Computer amazing, Hypic Agent. How do you choose? I saw you use Claude as well. It's not like you're just choosing one tool. What's the best setup?

Howie Liu: What's the best setup?

Marina Mogilko: So I think maybe going back to our generational metaphor there was the chatbot era and there were a lot of products that were built as chatbots. Now we have agents that are much more fully autonomous. I first would separate products that fall into the chatbot era from the agent era. Even within the same company like Anthropic, they have Claude, the vanilla version which is much more close to a chatbot. It is actually a weakly agentic product. So it can do a few turns at once but you're not using Claude to go and work on something for five hours without human intervention. And this is Claude, the end-user product. Confusingly they also use that name for their models and as a prefix to their other products like Claude Code and Claude Computer, but the plain end-user Claude experience I think of as more of a chatbot. Likewise with ChatGPT and that can still be very valuable but when you think about frontier agents that are actually able to perform maybe hours of human equivalent work autonomously, you put into that category OpenClaw. Everybody got excited about OpenClaw on Twitter because of what it could do. I would put into that category Claude Computer. And products like Perplexity Computer and of course Hyper Agent. And within that bucket there's different options depending on what you want. What Hyper Agent gives you and what we focused on is both the fact that you can do this in a much more team-enabled setting as you refine their skills and memory. We put a lot of emphasis into closing the loop for self-improvement which I think is key because even though agents are out of the box very smart and they're only getting smarter as the models underneath them get better and better.

Howie Liu: Most people use Claude as a search engine. They type in a question, they get an answer, most times they're not really satisfied with it and they close the tab. I did the same thing for months and I was looking at people who were saying AI is changing their life and I was confused. Then I spent one afternoon setting it up properly, uploaded a few files about how I think and how I work and it completely changed. I wrote the whole process up step by step. You get it when you subscribe to my newsletter, Future Proof. It's free. The link is in the description. This is what I'm very excited to try. When you mentioned the team collaboration, we don't use Slack, we use Telegram, but I saw you have this Telegram integration. And the thing is, the industry is moving at the same pace and thinking about the same problems.

Marina Mogilko: Last week I was talking to my COO and I said I'm giving so much feedback to my team in Telegram which is inaccessible to anyone. So I have to download the conversations. How do we close this? How does the agent learn my taste? So what we're going to do is download the conversation, train the agent, but then we're going to add the bot to all the team chats that I have so it learns my feedback and can give feedback to me instead of me giving feedback to my team.

Howie Liu: Exactly. One thing, as an aside, I think Telegram has a really interesting opportunity to become maybe the dominant platform in messaging for agents because they have the best ergonomics of all these messaging platforms to deploy bots. Hyper Agent has a really first-class Telegram integration. Even if you're not a current Telegram user, I might recommend setting up a Telegram account because it's an easy and free way to create both individual and group chats with your agents. It's very cool to see in a group chat setting. You can add even multiple agents that you've created through Hyper Agent and have them jump into the conversation as relevant and even start to talk to each other.

Marina Mogilko: Even give feedback because what I realize in my company is I'm the bottleneck because all the content goes through me and sometimes I'm doing this and I can't get back to my team. What if I have an agent that already knows my taste and I say on busy days you just let it give the feedback?

Howie Liu: Absolutely. I think that is one of the interesting emergent phenomena here. As people have gotten really into their agent building, you see all these OpenClaw fans who have created a virtual twin of themselves in their OpenClaw instance. This is my virtual Howie agent that has really learned so much about me and actually has real-time access to the same context that I do so it knows what my schedule is. They talked about the singularity and how we're going to put ourselves into a computer and it's actually happening with agents.

Marina Mogilko: What's the most unique use case with Hyper Agent that you've experienced?

Howie Liu: I personally like multimedia-related use cases. Sometimes for fun and sometimes for actual functional marketing purposes, I like to have Hyper Agent create marketing ideas. At one point, we did a billboard campaign for Hyper Agent, announcing the launch of it. I had it go end-to-end sourcing the actual billboard location. We worked with a vendor that had basically a list of all the billboards that were available. We grabbed all the inventory and then it had all the specific locations like what street it was on. Then Hyper Agent was able to process all of those locations, cross-reference them with Google Street View and Google Maps so it could literally put them on a map and then show the street view point of view of what that billboard looks like on the street. Because it's very versatile in terms of chaining together different tools, it was able to take those street view images and pass them to leading image models like Midjourney and then use that to generate a really high-fidelity mockup of what our billboards would actually look like. It took real-world location shots, mashed that up with our actual campaign imagery which it created based on the concept and then could help us visualize what that billboard on Sunset would look like. I thought that was very cool. We ended up not being around long enough to do a Super Bowl ad because we just started Hyper Agent basically around the holidays. But had we done a Super Bowl ad or maybe for next year, I would use Hyper Agent to generate actual video concepts. It's able to come up with concepts combining real product marketing understanding of what does this product do. We gave it access to our documentation for Hyper Agent. So Hyper Agent was able to learn about all of the capabilities and differentiation of Hyper Agent and then come up with really good marketing concepts that it could cut up into different scenes like a screenwriter or director of a Super Bowl ad script and then actually generate really high-quality production-grade videos using voiceover from Google so that it actually looked like a real Super Bowl ad. I like the creative use cases because they're immediate and visceral. But of course I also have many functional use cases, a chief-of-staff use case where it's reading my emails and Slack all the time and pushing me anything that needs my attention, saying "Hey, you just got this really important email from a customer, partner, or investor. You should probably respond to this and here's even a drafted reply."

Marina Mogilko: Can you show me some of your productivity use cases because this is where I see a lot of my time being saved?

Howie Liu: Yes, I optimize stuff with agents. This is a demo account of mine that's realistically recreated based on my actual usage. My actual usage has everything like all of my calendar, my emails, everything. It's a little sensitive. Either you can have it go on a recurring schedule or similar to OpenClaw you can have a heartbeat mode where it's always waking up and checking new stuff and then pushing you messages either via Telegram, Slack, or email, but you can set it up how you want. I like Telegram because then it just feels like it's almost a personal assistant messaging you all the time. You can respond to it and if you have the bot living in your group channels already with your team, you can just mention the bot and say "Hey, go research this thing on behalf of the team" or "Go create a caption for this post."

Reed Hoffman: Exactly. Okay. So here's one that is one of my personal favorites. Obviously you have to be on top of X like 24/7 now to keep up with all the latest news. It's all happening so fast and it's overwhelming. Sometimes I feel like I forget to check X and I miss some really important stuff. So I have one agent whose entire job is just to constantly watch X and literally do this all the time 24/7 and then it's making a judgment call. It knows enough about me and what I care about and Airtable and Hyper Agent that it can push to messages alerting me when there's something interesting, only when it's actually relevant to me, and it can even frame this around why is this relevant to you, to Howie.

Marina Mogilko: Interesting how social media is transforming because I see a lot of people doing this because once you enter X you forget your life.

Reed Hoffman: Totally, you go down an agent consuming content for you. Exactly, it's an interesting shift for social media in general.

Reed Hoffman: Well, it's agents consuming the content, but then also probably agents helping to create the content. And so there's this funny loop of what happens when it's really just our agents posting and consuming all the content on our behalf.

Marina Mogilko: Exactly.

Reed Hoffman: So this is a favorite one of mine. And one thing you can do here is if you want to set up a live mode agent, you can actually click that dialogue button or just ask it in the feed or in the thread. Turn this into an always on agent and push me messages via Telegram when you see something interesting for me.

Marina Mogilko: And then it will set it up. It will walk you through the flow. You can set up a Telegram bot. You're doing this now and I'm realizing a few months ago we built this system that gives us notifications about videos that are underperforming.

Reed Hoffman: Okay, took us a few days.

Marina Mogilko: Yeah. And by the way, even a few days is far better than a few months or a few years. But it's still this is such a way how technology progresses these days. You have to be always on, always testing, because there's new products coming.

Reed Hoffman: Yeah. And I think one of the cool things if I take a step back and think about the state of tech right now is I've always believed that you can choose to make tech an enabling tool. In fact, we spent a lot of time and studies scoffed at the idea of a personal computer. They literally laughed at it. They were saying nobody would need a personal computer on their own desk, let alone they didn't even think about smartphones. And the idea that obviously Apple and Microsoft had was, "Wow, these processors are getting cheaper and faster." And in fact, instead of using that to do even more arcane and specialized work, what if we made this even more accessible? What if we chewed up all of that computing power to make computers something that anyone can use and get value out of? And I think there's a very strong parallel to models. As the models get smarter, you could apply them towards very complicated use cases, hire Palantir for $300 million to do a very complex government deployment of AI, but also the alternate point of view is you can use it to enable people to do even more broad and interesting and ubiquitous things. So I'm personally a fan of the democratization angle. That's the entire founding premise of Airtable was democratizing app creation, and I think now we have a very similar opportunity to do that for agents. I really think there's going to be two paths: agents that are more and more powerful but inaccessible, and then agents that are actually more and more friendly and accessible and useful to as many people as possible.

Marina Mogilko: Yeah, I just saw you speak on stage and you said builders are going to win in the next few years. How do you adopt the builder mindset?

Reed Hoffman: Yeah, I think what really being a builder comes down to is having the appetite to tinker. That's the most important thing, because I think it's a humility to say nobody knows all the answers. If somebody tells you here's exactly how to build the perfect agent for every single use case and every single company out there, they're full of it, because there's no way to prescribe that so perfectly. It's constantly changing. And what's really cool about the builder community out there already on X or on Reddit is you're finding that people are almost accidentally discovering the best ways to use agents. They're learning, for instance, one way to make agents very good at content production is to give it this kind of skill with this kind of guidance, or maybe create a skill from the Mr. Beast leaked handbook on how to create content. When you feed that into the agent and have it develop its own skill informed by that, it does really well. And so there are all of these emergent practices that actually make these agents work better and better, and the only way to really get good at it is to try it out. The great thing is there's no master class, there's no PhD that you need for it. I'm sure you could take some great classes out there, but you don't really have to. If you have nights and weekend time to go and just play around with the agents, and it's quite fun to use, I think that's the best way to actually get into it. So the tinker mindset is most important first and foremost, and I think the second is really a willingness to distill down what is the work in order to generate the output you want. Some people have an easier time imagining what is the ultimate purpose of software engineering or sales or marketing or content creation. But if you are stuck in thinking about it as the activities that you currently do, software engineering—the activity of writing code by hand line by line—is now going to be obsolete. Same as script writing.

Marina Mogilko: Exactly, yeah.

Howie Liu: But if you can rethink the work as being how do I generate great software? How do I output great code and great applications and go up in the abstraction to have a team of agents that actually do that work for me? Or the same for script writing or content, I think then you're able to really effectively leverage agents because you're thinking about the outcome first and then how do you use these awesome capabilities to generate that output, versus being stuck in the activities of before. I think a lot of it is just getting that starting point right. Sometimes it feels very daunting to get started. I personally, every time I try a new generation of AI or products, it always feels a little intimidating. How do I get started? Over the holidays I personally got back into software development in a really big way. I had done some tinkering before, but with the latest family of models, I actually started building with code again using development agents. At first, I just picked some low stakes projects to do for fun and got so immersed into it that by the time I personally started working on Hyper Agent in code, it was already very exciting and I had built up some confidence around what I could do with the agent. So I think the really empowering thing is you can start with anything—low stakes personal use cases—but that helps you develop the confidence and the fluency to then go and bite off something even bigger.

Marina Mogilko: Yeah, you could have a whole agent just dedicated to helping you prepare for that. And you could even have a meta agent that itself is running 24/7 to discover new opportunities for additional agents.

Howie Liu: Oh, that's amazing. Do you have that agent? No, I think it's genius. I don't have one right now, other than my chief of staff, which is constantly looking through all my stuff and suggesting follow-ups, and sometimes it will suggest additional agent use cases. But I should—I want to have a weekly review, look at everything I did, and then come up with additional agents that you should hire, basically, onto your team to automate even more of what happened over the week. So this one I'm just setting up—the final parts of the Telegram flow. But in short, it will help me walk through the configuration to actually turn this agent into an always on Telegram agent that will then push messages on.

Marina Mogilko: So I gave it some feedback on what I wanted to do. I should set up an agent for a kids school email. Now that I'm thinking...

Howie Liu: Well, now that we're talking about personal use cases, I have some fun ones. I'm in the market for a used car right now. I have an agent that's monitoring all of the used car listing sites and finding cars of my spec. I want a convertible here—we're here in LA, and it's a great place to have a convertible. It will do extensive research on the price point I want and whether the cars have the specs that I want, see locations, click through and see where it is and get more details.

Marina Mogilko: You know, just fun personal use cases that I wouldn't have otherwise hired a person to go and monitor. Now I'm thinking I need to book my flights to Canada. I could basically build an agent that goes through all my points that I have on different credit cards and find the right one.

Howie Liu: Oh yeah, a points optimizer to use the points effectively, even help you plan out with a map all the different places you want to go.

Marina Mogilko: So yeah, it's a very fun time.

Howie Liu: I saw this term referenced in one of the AI forums I'm in. One way to think about agents is obviously you can use it in your core job, but it also enables you to make all these luxury hires that you otherwise wouldn't. Maybe you wouldn't hire a full-time travel concierge for your team. That wouldn't be a full-time hire. But now with an agent, you can literally have an agent whose full-time job, even running 24/7, is to find ways to optimize your points, find new itineraries that are surfacing from your email. It knows, maybe you need to go out and do a promotion or an interview, and it will automatically propose here's the flight itinerary, here's the way that you can optimize your points for this trip. It allows you to imagine all these scenarios: if you could hire as many people as you wanted for almost zero cost, what other roles would you hire people into? I think that's a very cool thing. A lot of people are worried, rightfully, about what jobs will look like in the coming years, but I think some of the most interesting use cases of agents are going to be not necessarily replacing existing human jobs outright, but actually enabling completely new jobs that weren't being done because it wasn't worth it to have a full-time hire to do something. You're able to do incrementally so much more that puts you ahead, personally and as a business.

Marina Mogilko: If I'm running all these agents, I'm getting all that information and learning how to live in that environment where yes, they source a lot of stuff for you, but then you're still the end decision maker. So it's a lot of decisions now, which is actually a founder's job. You're not building, you have to be making those big decisions.

Howie Liu: That's the thing. I think the common thread is that the thing that scales the most and gives you the most leverage in the agent world—and it was true in the human world as well—is good judgment. How do you apply good judgment as efficiently and to as many different threads as possible? I think the people who are able to do that most effectively will become almost superhuman. You will have this team of agents that you're able to give good judgment and feedback to, just like a really effective CEO of humans. What discerns a good CEO of humans versus not is ability to scale good judgment.

Marina Mogilko: And you scale it by making more decisions and seeing how they work for you?

Howie Liu: Absolutely. And maybe iterating, experimenting more, having agents try things that you otherwise wouldn't have tried. Maybe you can do new content experiments or new business model experiments. Hey, what if we built our own brand or micro sites around these different opportunities for Airtable or for Hyperagent? We can actually execute on way more marketing programs or product features than we would before. Not all of them will work, but we just get way more at-bats.

Marina Mogilko: Okay, my last question for anyone who wants to start with agents today: first three steps.

Howie Liu: One is obviously sign up for Hyperagent. We try to make Hyperagent really non-intimidating. It's got a very great GUI experience. You don't have to go in and do technical setup. It's like using the Mac as opposed to setting up Linux. The second is just come up with a few personal use cases or low stakes work use cases. I think problem hunting—figuring out what are the problems you want to solve—is 80% of the battle, because the actual building of the solution is no longer so hard. Compare this to the prior era where if you wanted to build custom software, even before no code, and you wanted to solve a problem with a custom app, well, you could pick the right problem. Let's say you want to track all your brand deals. But then actually writing the code to build the app is hard, and you need to hire someone.

Marina Mogilko: And so now I think it's inverted, where the most interesting part and actually the part that you should spend the most time on up front is just coming up with a list of what are all the problems that I want to solve? And then third is have fun with it.

Howie Liu: I think that's the most important part—or that's a really important part. This has to become a passion. The best agent users and builders I see really enjoy it. Just like if you were an early internet user, you had to enjoy it. There's a functional purpose to it, but it's also fun to go online and see what else was out there—shopping sites, news sites, games, and so on. There's this interactive and dynamic nature to agents that I think, to become truly fluent, you have to have passion and technology fit. Try to enjoy it. That's why I think picking the lower stakes, fun use cases where you get to really experiment is better than putting all this pressure on yourself up front to deliver some ROI. It's just like any other new and disruptive tech—computers, the internet, the iPhone. To really understand how to apply it in an efficient and effective business way, you have to first immerse yourself as a consumer of it.

Marina Mogilko: And two skills everyone should be working on from what you're saying: the skill to figure out the problem to solve.

Howie Liu: And the skill to make the right call or judgment.

Marina Mogilko: Absolutely.

Howie Liu: I think those will be the two most important things, and effectively mastering them I think makes you superhuman in this.

Marina Mogilko: Exactly, and we're living in an age where everyone has the potential to become superhuman in less than a couple of weeks.

Howie Liu: Seriously, I think we'll see a lot more entrepreneurship. The era of one person building a great company—whether it's the literal billion dollar revenue company that was forecasted with AI or it doesn't have to be a billion in revenue to still be successful—I think a lot more people are going to be able to get off the ground with their own ideas and build a business. Whether it's an online retail business that before they couldn't have afforded to hire the team to market it, source the inventory, build a site—now you can do that all with agents—or it could be a content business or a software business. All of these businesses I think are now much more possible for anyone out there who has the idea and sees an opportunity.

Marina Mogilko: It's an amazing time to be alive. Thank you so much, Howie, and thank you so much for the product.

Howie Liu: Of course.