Back Research Notes The Carousel of Progress Is Broken: AI Agents and the New Economics of Speed Published on July 28, 2025 By Jordi Visser When there’s a great big beautiful tomorrow Shining at the end of every day There’s a great big beautiful tomorrow Just a dream away There’s something permanently etched into my memory from family trips to Disney World: the Carousel of Progress. It wasn’t the thrill rides or the fireworks that stuck with me. It was a slow-turning, outdated attraction with animatronic families and retro optimism. But for my family, it served a very practical purpose. Every time the sky opened up with an Orlando lightning storm, and it always did, we’d duck into the Carousel to wait out the rain. Inevitably, we’d stay through all four acts of the show, watching time progress in 20-year leaps. And just as inevitably, by the time we emerged, the rain would stop, and that song, “There’s a great big beautiful tomorrow…” , would be permanently lodged in my head for the rest of the trip. That ride didn’t just attempt to show us the future. It showed us how we used to think about the future. Time progressed in orderly fashion. Change was generational, technological innovation was a shared family moment, and each step forward was manageable. The Carousel made it clear that tomorrow was always “just a dream away.” But what if tomorrow arrives before today ends? The Carousel of Progress is a relic of a world that believed in linear innovation. Its very structure, rotating through scenes every couple of decades, was comforting. In that world, the washing machine arrived, then the television, then the computer. Each came with enough breathing room to adapt, absorb, and still preserve the past. But today, the carousel isn’t just turning. It’s spinning faster than we can follow. And it’s flinging the present, past, and future into a blur. This weekend, during one of my regular podcast binges, I stumbled upon an episode featuring Daniel Kokotajlo. He’s a former OpenAI researcher and now the executive director of the AI Futures Project. He’s also one of the authors of a paper called AI 2027, which lays out a vision of a near-future world that feels like science fiction, except it’s only two years away. In it, the authors describe a world where AI agents are no longer just tools but autonomous actors: conducting research, building businesses, managing workflows, even writing code. Cycles that once took generations now collapse into quarters or months. Rather than a single breakthrough defining an era, we’re facing layers of compounding breakthroughs month after month. The feedback loop of AI building better AI is already underway. I hadn’t heard of AI 2027 before listening to Daniel, but a few podcasts later, it popped up again, this time on the Moonshots episode featuring Emad Mostaque. Emad, a former hedge fund manager and the co-founder of Stability AI (the company behind Stable Diffusion), mentioned the paper as well and emphasized many times, it is coming by 2027. Hearing both Daniel and Emad independently reference AI 2027 in the same weekend sparked a deeper dive. It became clear they were both pointing to the same urgent theme: the unprecedented impact AI is set to unleash within just the next two years. This is what it means when the Carousel of Progress spins too fast: we stop being able to place ourselves in time. Our frame of reference shatters. A product launched this week can be obsolete in six months. Careers are reinvented overnight. What felt like science fiction last year is commoditized today. And unlike the safe, air-conditioned nostalgia of Disney’s ride, this world doesn’t offer a slow fade into the next scene. It hits like a system update mid-sentence. The speed of AI’s advancement, and its effect on productivity, will likely leave economists and market analysts consistently underestimating the future trajectory of GDP and corporate earnings through expanding profit margins. The AI 2027 paper also signals a collapse of traditional GDP models. In the past, economic growth followed a simple recipe: more labor, more capital, more productivity. But now, AI is rewriting that equation. Total factor productivity, once a slow-moving force, becomes the central character. AI agents accelerate R&D, compress product cycles, and remove the lag between discovery and deployment. In macro terms, we’re watching the Cobb-Douglas production function break. Growth no longer depends on adding workers or machines. Instead, it’s powered by recursive intelligence, AIs improving AIs, compounding output without compounding input. GDP, once a reflection of human labor and capital, is becoming a function of how rapidly intelligence can iterate on itself. Emad Mostaque talked about this shift in his Moonshots with Peter Diamandis interview. He put it bluntly: “Capital no longer needs labor.” That’s not a theory, it’s a reflection of what AI infrastructure is already doing. Data centers, LLMs, and autonomous agents are becoming standalone economic engines. And that leads to a deeper question: If capital no longer needs labor to generate productivity, how does labor gain capital? The AI economy isn’t just about speed, it’s about redefining participation. We’re transitioning from a jobs economy to a skills economy, one where adaptability and fluency with AI matter more than titles. But not everyone is ready to make that leap. And in the meantime, GDP will grow, just not in ways that feel inclusive. In Situational Awareness , Leopold Aschenbrenner outlines a roadmap in which the trajectory of AI development, driven by exponential scaling in compute and breakthroughs in algorithmic efficiency, puts us on a clear path toward artificial general intelligence by 2027. He emphasizes that the trendlines are intact: increasingly capable systems are arriving faster, and the infrastructure behind them suggests world-changing breakthroughs are not decades away, but merely a few training runs out. This aligns closely with the AI 2027 paper, which forecasts that autonomous AI agents, recursive productivity, and a breakdown of traditional economic models could emerge within the same time frame. Both pieces point to the same reality: we are accelerating toward a future where the rate of AI progress will outpace institutional, regulatory, and even cognitive adaptation and the world of forecasting that we knew for anyone involved in economics or forecasting company earnings will be fundamentally disrupted. Historical valuation frameworks that once justified “cheap” or “expensive” labels now mean little in this new paradigm. It’s garbage in, garbage out. At the center of this disruption are AI agents , not humanoid robots, but digital employees who work around the clock, learn independently, and scale instantly. These autonomous agents are rapidly becoming the most important drivers of economic output, capable of conducting research, writing software, managing operations, and even launching businesses. As the AI 2027 paper makes clear, this shift toward recursive, agent-driven productivity is already underway. Emad Mostaque underscores the same point in his Moonshots interview, noting that data centers and AI models are turning into standalone productivity engines where capital no longer needs labor in the traditional sense. Aschenbrenner, too, suggests that the gap between our current systems and those with world-altering capability may be just a few iterations away. The common thread is unmistakable: the most valuable workers of the near future won’t be human or humanoid, they’ll be intelligent, adaptable agents accessible through an API. In short, the dynamics of growth are being fundamentally redefined. Capital now scales instantly through AI deployed on-demand via APIs, while labor becomes increasingly optional as autonomous systems replace human workflows. Most importantly, productivity becomes recursive, AI improving AI, driven by rapid discovery and compounding inno