Airtable CEO reveals how the 1% leverages AI | Howie Liu

I am the bottleneck because all the content goes through me. And if I had people who already know my tastes and I said, 'How? Just let him give the feedback.'

People who can do that more effectively will become almost like superhumans, right? You can provide good judgment and feedback like a really effective CEO of humans. You could imagine building a company you never dreamed of, without employees or with very few employees.

So the experimenter mindset is the most important above all. And I think the second is really... my goal is that anyone who has been consuming content about AI, but who wasn't afraid to try it, but thought it was too complicated.

Right, so you realize how powerful the technology has become. Compared to, for example, two years ago, when maybe you tried a single prompt in ChatGPT and thought, 'This is nothing.'

Yes, when you work with this technology correctly, it's actually very transformative. Can you tell me about the current state of AI? We've been in an era of chatbots? You had to initiate the process, correct them, and then the result. Would you really not trust an agent to post for you on LinkedIn? I feel like now we're trying to close the loop with these agents, like teaching them our taste, teaching them how we give feedback, and making them more autonomous. That's what you're also saying.

Yes, I think there has been, you know, like a radical shift in the experience you can create with AI since before ChatGPT, but then I think we entered a more mature era, even of AI chatbots, where the models became smarter. So even thinking about, you know, the year after GPT-4 came out, we could do really interesting things with chatbots, so the maturation of chatbots. And then I think we started generating expectations about agents too soon. So there was a year, like last year, people were saying this was the year of agents. And I think it was when Marc Benioff of Salesforce got up and was talking about how in the future every CEO is going to have thousands of agents as employee agents and human employees. And I think he was right, but it was just too soon. And so what we've finally reached now, I think, and some would say it really was like towards the end of last year with Anthropic's Opus 4.5 and the most advanced generational models, but now you know, you have GPT-5.5 and Kimi 2.6 as these open-source models and, you know, alternative models that I think have all reached this level of intelligence that really allows agents to be almost like humans, right? and autonomous.

Yes, like I'm responsible for that agent.

You know, I think as the models improve more and more, yes, we need to rethink the entire paradigm and product format to reflect it. So I'm going to make a comparison here which is in the world of development agents, you know? You had Copilot, GitHub Copilot, which was like the first, you know, like a chatbot for developers, right? And it really was, you couldn't make it develop really complex programs, it was more like you could autocomplete a few lines of code at a time, so it was more of a practical tool. Then you had Cursor, which introduced its own AI agent called Composer, which is different from how they use that name now for the model they've fine-tuned. But at that time it was a much more autonomous agent. You could tell it, 'Go and create this file for me, write me this simple script or a function.' and it would do something much more autonomously than, you know, the original Copilot experience. I think we've now entered a new era where agents allow you, you know, to go and run in parallel with multiple agents. So in the development world, the best developers, you know, aren't necessarily using Cursor. In fact, now they're using many different agents. Maybe it's Claude Code, for example, running in parallel and it almost feels like they have their own company or team working even overnight, right? In fact, when I'm developing, I use agents that, you know, I try to get them to do something substantial before I go to sleep so that, you know, for several hours at least I'm not wasting resources and they're doing something useful. And so when I wake up they've already finished, right?

So I, you know, I think we're moving towards a world where the form factor will start to look more like managing a fleet of agents or almost like managing, you know, people.

Yes, you still need some of the same techniques, but it's also quite different, so I think we're moving in that direction. But as you say, I think we're not yet, you know, at a point where you literally just let the agents do everything, right? I don't think agents can effectively run an entire company or any significant company autonomously.

I think based on what you said about analytics, you know, the fact that agents can help close the loop, maybe drafting some content. There's still a human review step. You're the one who posts it and then maybe the agents are helping analyze what's working, what's not working and seeing what is, you know, starting to take off, among other people in the content world. You know, I think this creates a circuit that multiplies results immensely for people who know how to deploy agents effectively. So, even if humans are still involved at certain points, I think the part of the process or the percentage of the process that is more agent-driven is going to increase and increase and increase until at some point it's almost like yes there is a human, but that person multiplies their capacity immensely and you could imagine building a company you never dreamed of without employees or with very few employees. Now you can do that with agents.