WHY THIS MATTERS IN BRIEF
Agentic AI is vastly different to traditional Generative AI, and that means in time you’ll be able to give it almost any command and it will carry it out.
Love the Exponential Future? Join our XPotential Community, future proof yourself with courses from XPotential University, read about exponential tech and trends, connect, watch a keynote, or browse my blog.
Could Artificial Intelligence (AI) help you build a TikTok clone in 30 seconds? Maybe in the next couple of years, according to Eric Schmidt, Google’s former CEO, who appeared on Erik Brynjolfsson’s interview series at Stanford the other week.
During the intimate interview Schmidt confessed to revising his AI outlook every six months which is a testament to the field’s volatility. He shared a striking example: “Six months ago, I was convinced that the gap [between frontier AI models and the rest] was getting smaller, so I invested lots of money in the little companies. Now I’m not so sure.”
Now, please don’t focus on the fact that Schmidt thinks the future is in ever-larger models (he does). Rather, consider the nature of his knowledge. He is an insider’s insider, about as well-informed as anyone in this field can be, and unlike some critics, he is also putting his money where his mouth is, backing many AI companies like Mistral, Kyutai and Asari.
AI and the Future of Software and SAAS, by Futurist Keynote Matthew Griffin
Schmidt understands scale and gets neural nets. After all, he ran Google when it acquired Deepmind, developed the transformer architecture and built Tensor Processing Units (TPU), the first chips dedicated to speeding up deep learning. And Google has been about scale since its inception.
Despite this, just six months ago, this tech titan thought smaller models might stand a chance to push the frontier, but he doesn’t believe that anymore. The point is that he was either right then or he is right now. It took just six months for a U-Turn. That is the degree of uncertainty.
Schmidt describes the potent combination of large context windows, AI agents that can learn and improve themselves, and Text-to-Action capabilities. These “will have an impact on the world at a scale that no one understands, yet much bigger than the horrific impact we’ve had by social media.”
Large context windows, the working memory of large language models, are solidly on their way. Claude, Anthropic’s AI for example, can already hold 200k tokens (approximately 150k words) in its working memory and can code for a whole working day straight, and ChatGPT is on its way to “remembering your entire life,” according to Sam Altman.
Agent-based systems, which autonomously execute multi-step tasks and adapt to environmental feedback, are poised to unlock significant economic value next. Building on this foundation, text-to-action capabilities will emerge, translating natural language into specific, working programs. Imagine your own personal programmer at your fingertips, ready to create whatever you envision. A dating app that matches people based on their Netflix viewing history and Spotify playlists? A prototype within minutes.
These forces will converge.
While he doesn’t say it explicitly, it seems that Schmidt expects the next two years to be faster and more turbulent than the previous two.
What does this mean for entrepreneurs? Today, using LLMs already helps achieve a hell of a lot very quickly, even with limited context windows, brittle agents and limited actions. Schmidt’s most provocative example ties these trends together: Say to your LLM, ‘Make me a copy of TikTok. Steal all the users, steal all the music, put my preferences in it, produce this program in the next 30 seconds, release it, and in one hour, if it’s not viral, do something different along the same lines.’
Today, I can accurately and rapidly get an LLM to analyse and deconstruct all manner of academic papers in realtime. LLMs that are currently in testing have much larger contexts, connect to the Internet, and can write code; therefore will be much more capable. Perhaps as capable as rapidly iterating through new digital and, ultimately, physical products as Schmidt provokes.
The implications are simply gargantuan.
This uncertainty extends to the scale of investment required. Schmidt noted that leading companies are discussing needs of “$10 billion, 20 billion, 50 billion, 100 billion.” He even mentioned that OpenAI’s Sam Altman believes it might take “about $300 billion, maybe more to reach Artificial General Intelligence (AGI). These figures aren’t just about money – they translate to enormous energy requirements that could reshape geopolitical alliances. As Schmidt put it, “We as a country do not have enough power to do this.”
Schmidt’s insights reveal a landscape changing so quickly that even he, as a veteran of the industry, is struggling to keep pace. And a lot of what Schmidt said wasn’t new. We’ve been talking about extending context windows, the speed of change and the challenge of scale for years.
But here is why this discussion mattered a lot – Schmidt was speaking in a small, intimate class at Stanford. While not entirely unguarded, he was operating with fewer filters there. You’re more likely getting his genuine thoughts about AI. And his beliefs about the future are not dependent on any scientific breakthrough or klaxon signalling the arrival of AGI. So this is worth taking more at face value than with a grain of salt.
In that sense, some of what is coming soon should already be baked into your expectations and, by extension, your plans for the future, so are they?
