AI’s Plateau Is Just a Launching Pad for Next-Gen AI Agents

Dear Subscriber,

Recent developments in artificial intelligence have revealed an unexpected twist in the industry's trajectory.

In the past, new breakthroughs in what AI can do have been the result of a simple strategy — building bigger models.

However, that may not be the most practical approach going forward.

Until now, Moore’s law held firm. Named for Intel founder Gordon Moore based on his observations, the rule states that the number of transistors in a chip will double every two years, which increases processing speed and power.

That means, every two years our technology gets better. And that is fundamental for consistently improving our AI capabilities.

But for the first time since its inception, Moore’s law may be reaching its limit.

The first Intel microprocessor, Intel 4004, had 2,300 transistors, each 10 microns in size.

As of 2019, a single transistor on the mass market is, on average, 14 nanometers, though even smaller 10 nm models were available as early as 2018. The very smallest we know of is just 1 nm.

It doesn’t get much smaller than that.

Moore’s law is running out of room.

This shift has profound implications for both AI development and the crypto industry, where AI integration has become increasingly crucial for innovation and market growth.

Leading AI organizations are facing unexpected challenges that suggest a fundamental shift in the development landscape.

OpenAI's forthcoming model, Orion, has demonstrated only moderate enhancements over its previous benchmark. Particularly when it comes to coding tasks, despite substantial investments in computational resources and training data.

Similarly, Google's Gemini project has encountered performance issues, with users reporting catastrophic inaccuracies and limitations that contrast with Google's claims of the model's advanced architecture.

These developments stand in stark contrast to earlier, more optimistic predictions. Like those made by industry leader Sam Altman, who suggested that superintelligence might be achievable within a few thousand days.

The reality appears more complex, with raw computational power alone proving insufficient to bridge the gap to advanced AI capabilities.

But I don’t think this is the end of AI. Rather it could usher in an era of new opportunity.

Because it reveals a truth about technology that isn’t always understood: Breakthroughs often come not from raw power, but from reimagining how existing capabilities can be used more effectively.

Specifically, I believe this plateau will spark an exciting pivot away from new, large models to agentic AI — frameworks that enhance existing models with decision-making capabilities and real-world interactions.

Companies can no longer justify exponentially increasing computational resources for only marginally better results. Instead, development teams are exploring how existing models can become more capable through AI agency — the ability to make decisions, interact with tools and execute complex tasks autonomously.

It’s a fundamental shift in how we believe AI systems should operate.

And it may already be underway. Consider …

This new approach creates AI systems rather than just models.

Implications for Crypto AI Projects

The AI scaling plateau has significant implications for crypto AI. Like with TradFi AI, projects that previously established themselves on breakthrough AI capabilities through pure scaling must now revise their roadmaps and value propositions.

Infrastructure requirements are shifting away from amassing enormous computing resources — things projects like Render (RENDER, “B”) and Bittensor (TAO, Not Yet Rated) help with — to developing efficient processing solutions and specialized applications.

Projects that can demonstrate practical applications and efficiency will likely gain advantages over those promising revolutionary but resource-intensive AI capabilities.

That’s why we’re seeing agent-based approaches emerge as a particularly promising direction. They allow projects to improve their capability significantly … without requiring exponential increases in computational resources.

This aligns well with the crypto industry's needs for automated decision-making and real-time response capabilities.

Projects that use AI agents are shaping up to be particularly promising.

Three AI subcategories in particular stand to benefit the most:

AI Orchestrators. Fueled by this shift, we're likely to see a proliferation of "AI orchestrators." These are platforms that coordinate multiple AI models and tools to achieve complex goals.

Think of orchestrators as the operating systems of the AI age, managing resources and directing specialized models much like a conductor leads an orchestra.

For crypto projects, this could look like systems that seamlessly combine trading algorithms, risk assessment models and market analysis tools under a unified AI agent.

Local AI. We’re also likely headed for a renaissance of local AI.

As models become more efficient at working with limited data and computing power, I expect local-first AI applications — which can operate effectively on personal devices — will surge.

This trend, in particular, could have a lot of promise in the crypto sphere. It aligns perfectly with crypto's core values of decentralization and individual sovereignty. And with how clearly governments and big companies want to control access to AI resources, a crypto-driven push toward local AI just makes sense.

A Blend of Old and New. But the most successful projects moving forward will likely be those that sit at the crossroads between the AI infrastructure of old and the AI agents of the future.

Projects like this could navigate the complexities of DeFi protocols, optimize yield strategies across multiple chains or even participate in DAO governance with sophisticated voting strategies based on real-time market conditions.

Real value will likely emerge from projects that can effectively harness existing AI capabilities through AI agency.

AI is still expected to be a powerhouse of a sector in the coming years, both in TradFi and crypto. Statista projects its market size will grow at a CAGR of 28.45% through 2030.

That would result in a market volume of $826.70 billion in just six years!

The only difference now is what opportunities you target in that sector.

The winners in this new AI landscape won't be those with the biggest models or the most computing power.

Instead, you’ll want to look for projects that can effectively orchestrate AI capabilities to solve real-world problems in the crypto ecosystem.

Best,

Jurica Dujmovic

P.S. Researching individual projects — while necessary when choosing an investment — can be time consuming and confusing. Not to mention that’s just the start! Then, price and market analysis has to come into play … all before you invest.

Or you could watch my colleague Juan Villaverde’s latest briefing. In it, he reveals one crypto he expects will outperform this cycle.