Bryan, I appreciate your restraint in waiting before weighing in on AI agents and takeoff. So much of the current discourse compresses labor-market shifts, policy moves, and campus debates into hot takes, but your focus on teaching, travel, and higher-ed conversations points to something more gradual: how people actually fold AI into intellectual and creative work. That feels closer to your 'Iterations and intimations' framing than any single takeoff moment. I’ve been testing a similar quieter approach with Seedance 2.5 (https://seedance25.tech/), where the emphasis is on creating through ongoing dialogue—an AI that tries to understand your vision rather than just automate tasks. It’s not an agent-hype platform, but it has made me think about how iterative creative conversation can shape output over time. I’d be curious whether you see that kind of tool as part of the everyday integration story, or as another layer that academia still needs to critically examine.
Bryan, I appreciate your decision to wait a few days before weighing in on AI agents and takeoff. That restraint feels rare right now, especially when so many signals—labor market shifts, policy moves, campus debates—are being compressed into hot takes. Your note about traveling, teaching, and facilitating conversations on AI and higher ed suggests the real story is less about a single 'takeoff' moment and more about how people integrate these tools into daily creative and intellectual work. I've been trying something similar with Gemini Omni (https://gemini-omni.dev/), where the focus is on completing amazing works through creative conversation—literally communicating with your creation. It's not an agent hype platform; it's a quieter way to test how dialogue with AI can shape output over time. That seems closer to the data-rich, reflective approach you're calling for. Looking forward to the series.
Did you guys hear this story? Recently a computer programmer set his AI agent to work on a complex coding task. The agent dutifully worked through the night without supervision until it made a series of breakthroughs that enabled it to finish the project. Excited to share what it had learned with its human, and without being prompted, the AI figured out how to acquire a phone number so it could call in to report its results! Welcome to 2026.
What strikes me is how much of the innovation pie the platform companies themselves own. If you're a startup building, say, a writing tool on top of Claude or GPT, you're in a precarious position, because with each new month it's possible that the platform will just ship a better version of what you built as a native feature.
It makes me wonder how to support genuinely independent innovation in this space. The recursive loop of AI building AI only tightens this. The companies with the models, the compute, and now the self-improvement cycles have advantages that are hard to see anyone outside that loop matching. I'd love to see more people rethink the architecture itself because it means there's more potential vs. building on top of someone else's.
That's a very important point, Stephen. The American giants - OpenAI, Google, MS, rising Anthropic - really aggregate more and more power to themselves.
What do you think of open source as a way around this? And does Chinese AI follow the US pattern?
Some people flag the importance of locking down access to advanced AI models as a national security concern. And because that's so far outside my expertise, I really can't speak to whether that's good or bad. But, I feel like open source must be the direction to head in, because it has proven to be so universally a good call in all other areas of software.
I think there's also a good case to be made for, like, a 'third front' that is independent from US/China - companies or foundations with an HQ in Europe, India, SEA, etc.
Bryan, I appreciate your restraint in waiting before weighing in on AI agents and takeoff. So much of the current discourse compresses labor-market shifts, policy moves, and campus debates into hot takes, but your focus on teaching, travel, and higher-ed conversations points to something more gradual: how people actually fold AI into intellectual and creative work. That feels closer to your 'Iterations and intimations' framing than any single takeoff moment. I’ve been testing a similar quieter approach with Seedance 2.5 (https://seedance25.tech/), where the emphasis is on creating through ongoing dialogue—an AI that tries to understand your vision rather than just automate tasks. It’s not an agent-hype platform, but it has made me think about how iterative creative conversation can shape output over time. I’d be curious whether you see that kind of tool as part of the everyday integration story, or as another layer that academia still needs to critically examine.
Bryan, I appreciate your decision to wait a few days before weighing in on AI agents and takeoff. That restraint feels rare right now, especially when so many signals—labor market shifts, policy moves, campus debates—are being compressed into hot takes. Your note about traveling, teaching, and facilitating conversations on AI and higher ed suggests the real story is less about a single 'takeoff' moment and more about how people integrate these tools into daily creative and intellectual work. I've been trying something similar with Gemini Omni (https://gemini-omni.dev/), where the focus is on completing amazing works through creative conversation—literally communicating with your creation. It's not an agent hype platform; it's a quieter way to test how dialogue with AI can shape output over time. That seems closer to the data-rich, reflective approach you're calling for. Looking forward to the series.
Did you guys hear this story? Recently a computer programmer set his AI agent to work on a complex coding task. The agent dutifully worked through the night without supervision until it made a series of breakthroughs that enabled it to finish the project. Excited to share what it had learned with its human, and without being prompted, the AI figured out how to acquire a phone number so it could call in to report its results! Welcome to 2026.
What strikes me is how much of the innovation pie the platform companies themselves own. If you're a startup building, say, a writing tool on top of Claude or GPT, you're in a precarious position, because with each new month it's possible that the platform will just ship a better version of what you built as a native feature.
It makes me wonder how to support genuinely independent innovation in this space. The recursive loop of AI building AI only tightens this. The companies with the models, the compute, and now the self-improvement cycles have advantages that are hard to see anyone outside that loop matching. I'd love to see more people rethink the architecture itself because it means there's more potential vs. building on top of someone else's.
That's a very important point, Stephen. The American giants - OpenAI, Google, MS, rising Anthropic - really aggregate more and more power to themselves.
What do you think of open source as a way around this? And does Chinese AI follow the US pattern?
Some people flag the importance of locking down access to advanced AI models as a national security concern. And because that's so far outside my expertise, I really can't speak to whether that's good or bad. But, I feel like open source must be the direction to head in, because it has proven to be so universally a good call in all other areas of software.
I think there's also a good case to be made for, like, a 'third front' that is independent from US/China - companies or foundations with an HQ in Europe, India, SEA, etc.