In a Sequoia general partners panel, three speakers discuss the massive AI opportunity and the urgency to move quickly. Speaker A frames distribution as three simple physics—awareness, desire, and purchase—and notes that unlike the cloud era, AI now commands global attention after ChatGPT’s launch. He argues that while value currently resides in the application layer, foundation models are encroaching via test‑time compute, reasoning tools, and inter‑agent communication, creating a race to capture value. Speaker B highlights coding as the breakout AI product that achieved product‑market fit, and points to innovations at the research‑product boundary such as deep research and Notebook LM. He emphasizes vertical agents—domain‑specific AI assistants trained end‑to‑end with reinforcement learning on synthetic and user data—as a promising startup opportunity. Speaker C expands the vision to an emerging agent economy where agents not only exchange information but also transfer resources and conduct transactions. He outlines critical technical challenges: persistent agent identity, seamless communication protocols, and robust security and trust. He foresees a future of nested neural networks forming swarms that will fundamentally reshape business.
Chapters
If you think about the physics of distribution, you just need three things. People have to know about your thing, they have to want your thing, and they have to be able to buy your thing. That's it. When the cloud transition got going, nobody was paying attention. AI is very different. November 30, 2022, ChatGPT comes out. The entire world is paying attention to AI. People are starting to break out. But for the most part, the opportunity is still wide open. We have believed, and we continue to believe, the value is at the application layer. But guess what, you got competition. We got a second scaling law, we got test time, compute, we got reasoning tool use and interagent communication that allows the foundation models to pretty far into the application layer. But this is the race. This is where the value is going to be. And if you don't get in front of it, somebody else will. You are in a run like heck business right now. Now is the time to go at maximum velocity. All of the time.
More and more of us are getting value out of AI and we're all climbing the ladder together on how to weave AI into our daily lives. The breakout application category of the year was coding, which reached screaming product market fit. A lot of the most exciting technology innovation in AI right now is happening at that blurry boundary between research and product. And I think the two breakthrough examples of this over the last year were deep research and Notebook lm. Let's talk about where value will accrue in the AI stack. I remember debating this question with my wonderful partners at Sequoia at the time. My partners, Pat in particular, were adamant that the value was going to accrue to the application layer. I think you were right, Pat. Vertical agents are a wonderful opportunity for startup founders who deeply understand the domain. And we see companies creating agents that are trained end to end to excel in a very specific workflow using techniques including reinforcement learning on synthetic data and user data to make AI systems very performant at very specific tasks. And the evidence we've seen so far makes us really optimistic.
A year ago, AI Ascent was all about agents. We were talking about agents and they were just beginning to form into businesses. The topic was really these machine assistants that eventually we predicted would come together as agent swarms. In the years to come, we think that this matures even further into an agent economy. An agent economy is one in which agents don't just communicate information, they transfer resources, they can make transactions. But in order to achieve that very big important next wave, we have a lot of important technical challenges, persistent identity the agent itself needs to be persistent. If you're doing business with someone and they change day to day, you probably won't be doing business with them for very long. Seamless communication protocols. Imagine personal computing without seamless communication protocols. No tcp, ip, no Internet. We're just now building that protocol layer security. If you can't meet the person you're doing business with palm to palm, face to face, that importance of security and trust is even further elevated. Eventually, these processes and agents are going to merge. You're going to have neural networks within very large, complex neural networks, a network of these neural networks, and this is going to change everything.
