The New Moore's Law: How AI Agents Are Accelerating Beyond Our Expectations
While Moore gave us 24 months, AI agents are doubling their capabilities every 4. Your five-year strategic plan might already be five years behind
The first time I witnessed an AI agent complete a multi-step task without human intervention, I knew we were entering a new era. But what I didn't fully appreciate then was just how rapidly these systems would evolve. Today, as I watch AI agents tackle increasingly complex problems with remarkable competence, I'm reminded that technological progress rarely follows a linear path.
We're now seeing evidence of what could be described as a "New Moore's Law" for AI agents - and the implications for businesses across every sector are profound.
The Exponential Growth of AI Agent Capabilities
For decades, Gordon Moore's prediction that the number of transistors on a microchip would double approximately every two years served as the heartbeat of the computing industry. This exponential growth in computing power underpinned countless technological revolutions.
Today, we're witnessing a similar exponential pattern in AI agents' capabilities, but the pace is even more breathtaking.
Recent research from METR has revealed something extraordinary: the length of tasks that AI agents can successfully complete has doubled approximately every 7 months over the past six years. Let that sink in. While Moore's Law gave us a doubling cycle of 24 months, AI agents are improving at more than three times that rate.
Even more startling, data from AI Digest shows that between 2024 and 2025, this doubling time compressed further to just 4 months. If this accelerated trend continues, we could see AI systems capable of managing month-long complex projects by the early 2030s—possibly even sooner.
This isn't just incremental improvement; it's a fundamental reshaping of what's possible. We’re currently on the orange line - but if we keep accelerating exponentially, there’s a real chance the line shifts even faster.
Measuring Progress: Time Horizons as the New Metric
One of the challenges in assessing AI progress has been finding the right metrics. Traditional benchmarks like accuracy scores or performance on specific tasks don't capture the revolutionary nature of what's happening with AI agents.
That's why the concept of "time horizons" has emerged as such a valuable framework. Rather than asking, "How well can AI perform this specific task?" researchers are now asking, "How long and complex a sequence of tasks can AI successfully complete?"
This shift in measurement reveals the true acceleration curve we're experiencing. In 2019, leading AI systems could reliably handle tasks requiring about 10 minutes of focused work. By early 2025, that horizon had expanded to approximately 24 hours of complex work. If the current pattern holds, we could see AI systems managing multi-day projects with minimal human oversight by 2027.
The implications are staggering. We're not just getting better AI; we're getting AI that can handle fundamentally different categories of work.
What This Means for Your Business
I've spent years advising companies on technology adoption, and I've never seen a capability shift with more transformative potential than this one. Here's what business leaders need to understand:
The compression of task completion time isn't just a technical curiosity—it's a business revolution in waiting. AI agents might soon handle tasks that once required teams of specialists working for weeks in days or hours.
Consider software development, where we're already seeing early evidence of this transformation. In 2023, AI coding assistants could help developers write individual functions or debug specific errors. By early 2025, systems like Gemini and Claude can architect entire applications, manage codebases, and implement complex features with minimal human guidance.
This isn't about replacing human workers—it's about dramatically amplifying what they can accomplish. The organizations that thrive will be those that strategically pair human creativity and judgment with AI's expanding capabilities.
The Acceleration Factors
What's driving this extraordinary pace of improvement? Several factors appear to be working in concert:
First, there's a virtuous cycle between model capabilities and useful applications. As models become more capable, they can be applied to more complex tasks, generating valuable training data and user feedback that further improves future iterations.
Second, the competitive landscape has intensified. With multiple well-funded organizations pushing the boundaries simultaneously, innovations diffuse rapidly throughout the industry.
Third, researchers have discovered that certain architectural improvements yield outsized returns for agent capabilities. The shift from pure language models to agentic systems that can plan, reason, and evaluate their own performance has unlocked new dimensions of capability.
Perhaps most importantly, AI systems have reached a threshold where they can meaningfully participate in their own improvement. We're not yet at the point of full recursive self-improvement, but we're seeing early signs of what that might look like.
Preparing Your Organization
If your strategic planning doesn't account for this acceleration curve, you're likely underestimating the changes coming to your industry. Here's how forward-thinking organizations are responding:
1. They're identifying processes that currently require days or weeks of human effort but fall within the emerging capabilities of AI agents. These represent immediate opportunities for transformation.
2. They invest in the infrastructure and expertise to effectively deploy and manage agent-based systems. This includes technical capabilities and the human skills needed to direct and collaborate with AI agents effectively.
3. They're reimagining workflows and organizational structures based not on current AI capabilities but on the capabilities we can reasonably project 12-24 months ahead.
4. They're developing clear governance frameworks that establish boundaries, oversight mechanisms, and accountability structures for increasingly autonomous systems.
When I advise executives now, I encourage them to think bigger than they're comfortable with.
The natural tendency is to incrementalize our expectations, but the data suggests we should be preparing for step changes in capability.
The Road Ahead
If the current doubling rate holds, by the end of this decade, we'll have AI systems capable of completing projects that would currently require months of specialized human effort. This doesn't mean AI will replace complex human work entirely—rather, it will transform how that work is structured and who can participate in it.
The real question isn't whether these capabilities are coming—the trendlines are clear. The question is whether your organization will be positioned to leverage them effectively when they arrive.
Those who understand and prepare for this exponential curve will find themselves with extraordinary new capabilities at their disposal—those who don't risk finding themselves suddenly years behind more forward-looking competitors.
The new Moore's Law for AI agents isn't just a technical curiosity—it's the heartbeat of the next business revolution. Unlike the original Moore's Law, which proceeded steadily for decades, this one appears to be accelerating. The time to prepare isn't next year or next quarter—it's now.
Steve has worked with hundreds of companies to understand and adopt AI in their organizations. Need a comprehensive AI workshop for your organization? Want highly customized, personal AI training? Grab a 15-minute slot on my calendar to discuss: https://calendar.app.google/3ctoTDUgtg71TQDG7
PS: Some of you have been asking about Magnus. He’s now just over 6 months old and about 100 pounds. Here are two recent shots:
