Welcome to the Jagged Peaks, where AI capabilities rise and fall unpredictably

Close friends and family may be concerned as to why I’ve decided to start a travel company to an imaginary location. I feel like there’s a desperate need to give this experience of large language models some kind of physicality to help confine it and bind it in our minds, because at the moment it can be pretty overwhelming when you don’t have that structure. So Jagged Peaks doesn’t feel imaginary to me, it’s the most honest way I can describe where we actually are with AI right now.

We’re in this fascinating and frustrating phase where AI can achieve olympiad-level mathematics but gets confused when contemplating if a seahorse emoji exists. These valleys and peaks represent key landmarks from which we can all have shared experiences of AI and sort of understand the gradients in between.

The frontier of AI capabilities: some peaks are conquered, others remain out of reach

The Jagged Frontier

In September 2023, researchers from Harvard Business School and Boston Consulting Group published groundbreaking research introducing the concept of the “jagged technological frontier.” They studied 758 consultants and found something remarkable:

• For tasks inside AI’s capabilities: 40% quality improvements and 25% faster completion
• For tasks outside the frontier: 19% more errors despite working faster

As Professor Ethan Mollick explained: “The problem is that the wall is invisible, so some tasks that might logically seem to be the same distance away from the center… are actually on different sides of the wall.”

Unexpected cracks appear where you'd least expect them

Industry Leaders Embrace “Jagged Intelligence”

In July 2024, Andrej Karpathy former OpenAI co-founder and Tesla’s AI Director independently coined the term “Jagged Intelligence” to describe this same phenomenon: “state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.”

This terminology caught fire. By June 2025, Google CEO Sundar Pichai told Lex Fridman we’re in the “AJI phase”, Artificial Jagged Intelligence, where we see “dramatic progress, some things don’t work well, but overall, you’re seeing lots of progress.”

DeepMind CEO Demis Hassabis echoed this, calling current systems “uneven intelligences” or “jagged intelligences”: “Some dimensions, they’re really good; other dimensions, their weaknesses can be exposed quite easily.”

Some peaks require careful scaffolding to reach

Why This Matters for Your Team

This jaggedness isn’t an implementation detail, it’s the central challenge of working with AI today. When I run workshops, I see teams struggle with exactly this: they’ve heard AI can be transformative, but they don’t know which tasks will benefit and which will backfire.

The Harvard research identified two approaches to navigating this terrain:

• Centaurs: Dividing work clearly between human and AI tasks
• Cyborgs: Interweaving AI throughout the entire workflow

Knowing when to use each approach? That’s the skill we’re all building together.

Between the peaks lie valleys where AI capabilities suddenly drop off

The Road Ahead

Beyond the Jagged Peaks route is AGI, where we can expect our experience to smooth out dramatically. Paradoxically, to get to AGI, the capabilities of our current models need to extend to things like multimodality, even robotics. Those are new domains where we can expect more drop-offs and maybe some surprising peaks.

As our use of AI spreads, there’s no clear path to a clear horizon. Instead, we’re left with opportunities and challenges to navigate the terrain in between here and then. And I’m excited for all the lessons and stories to share between then and now.

Hopefully AGI provides a smooth and pleasant ride.

So when people ask why “Jagged Peaks,” this is why. We’re all navigating this uneven terrain together, celebrating the peaks where AI soars, staying cautious around the cracks where it stumbles, and building the shared understanding to tell the difference.

The Knowledge Landscape

Here’s what makes the jagged peaks even more challenging: knowledge about AI is just as unevenly distributed as the technology’s capabilities. Some teams have already mapped certain peaks and valleys through trial and error. Others are still standing at the base, unsure where to even begin their ascent.

This creates a massive opportunity. Teams and individuals who invest in understanding this terrain gain a significant advantage when they learn which peaks are accessible, which tools work for which challenges, and how to navigate the gaps. They move faster, avoid costly missteps, and discover paths others miss entirely.

The jaggedness isn’t just in the technology; it’s in our collective understanding of it. And that’s where the real competitive edge lies: not in waiting for AI to become uniformly capable, but in becoming skilled navigators of the landscape as it exists today.

It’s going to be a bumpy ride. But that’s exactly why having an experienced guide matters.