For a decade now, the conversation around artificial intelligence has been dominated by two dramatic ideas: AGI (Artificial General Intelligence) as some discrete finish line we will one day cross, and the Singularity — the moment after which AI improves itself so fast that human understanding and control become irrelevant. Both ideas share a common assumption: that AI progress is building toward a single, sudden, world-altering event.

It’s time to retire that assumption.

Every Slice of an Exponential Looks Like a Hockey Stick

Here’s the trick the Singularity narrative plays on us: in any exponential growth curve, if you zoom into almost any section of it, that section itself looks like a hockey stick — a slow start followed by a sharp, dramatic bend upward. That’s just what exponential curves look like at every scale. Zoom into last year’s slice and it looks like a hockey stick. Zoom into this year’s slice and it looks like a hockey stick too. That doesn’t mean we’re approaching some unique, final, world-ending inflection point — it means exponential growth is made of hockey sticks, one after another, for as long as it continues.

This is exactly what’s happening with AI. Each year’s jump feels like the dramatic bend — “look how much better this model is!” — and it genuinely is dramatic when compared to the year before. But that’s not evidence of an approaching Singularity; it’s just what steady, compounding progress feels like from the inside, year after year. What we’ve seen, release after release, is a predictable rhythm: new model generations arrive, they’re noticeably better than the last, and the rate of improvement itself doesn’t accelerate without bound. It compounds — impressively — but it doesn’t explode into the one final, uncontrollable hockey stick that never ends.

Every year, AI models get meaningfully better at reasoning, writing, coding, and problem-solving. But “meaningfully better than last year” is a very different claim from “improving so fast humanity is losing control.” The Singularity, as originally conceived, requires an intelligence explosion with no ceiling in sight. In practice, every technology humans have ever built — from transistors to airplanes to solar panels — has followed an S-curve: slow start, rapid middle, and eventually a plateau as physical, economic, or informational limits kick in. There is no compelling reason AI will be the one exception to this pattern. Compute has physical and financial limits. Data has a ceiling. Algorithms yield diminishing returns as they approach the limits of what a given architecture can express. AI will plateau eventually — it’s just that we are nowhere near that point yet.

AGI Is Already Here, By Certain Standards

The other myth is that AGI is some binary switch: one day AI is “narrow,” and the next day it flips to “general,” instantly transforming the world. This framing made sense when AI could only play chess or recognize cats in photos. It makes much less sense today.

Look at where we actually are. Modern AI systems can:

There is no single human alive who is better than today’s best AI models at all of these tasks simultaneously. A brilliant surgeon can’t write flawless Python. A world-class novelist can’t necessarily pass the bar exam. A single AI model can credibly attempt all of it, often at a level that meets or exceeds an average professional.

If “general intelligence” means breadth of competence across a huge range of cognitive tasks — rather than superhuman mastery of literally everything, or human-identical consciousness — then a reasonable case can be made that we’ve already crossed into AGI territory. Not as a single dramatic event, but quietly, incrementally, one model release at a time. There was no headline that said “AGI achieved today.” There was just a slow accumulation of “wait, it can do that now too?” moments — and one day we looked up and realized the sum of those moments looked a lot like general intelligence.

Why This Reframing Matters

This isn’t just semantics. It changes how we should think about AI’s trajectory:

1. Stop waiting for a moment that isn’t coming. If you’re waiting for a clear, unmistakable “AGI achieved” announcement before you take AI seriously, you’ll be waiting forever — because that moment has already blurred past us. The capability leap already happened, gradually enough that it didn’t feel like a leap.

2. Stop fearing a runaway explosion that also isn’t coming. The Singularity as a sudden loss-of-control event is a much less likely scenario than a continued, manageable, if fast, curve of improvement. That doesn’t mean AI progress is risk-free — it means the risks are the ones that come with powerful, widely deployed tools, not with an unstoppable self-improving super-mind appearing overnight.

3. Expect the pattern to continue: better, not done. The fact that AI already outperforms humans at most individual tasks doesn’t mean progress stops here. Reasoning will get more reliable. Memory and context will get longer and more coherent. Models will make fewer errors, hallucinate less, and handle more complex, multi-step, real-world tasks with less supervision. Costs will keep falling, making today’s frontier capability available to everyone tomorrow. Each year will keep producing a new “best model yet” — probably for a good while longer — even as the headline-grabbing nature of each jump gradually fades, precisely because we’ll have grown used to expecting it.

The Plateau Is Real, Just Not Yet

None of this means infinite improvement forever. Every S-curve eventually flattens. At some point, gains from scaling compute and data will shrink, architectures will hit fundamental limits, and progress will slow to the kind of incremental refinement we see in mature technologies like commercial aviation or automobile engines — better, safer, more efficient, but no longer astonishing.

We’re just not there yet. The curve is still climbing. The right way to think about AI isn’t “are we approaching a singularity?” or “will AGI arrive tomorrow?” It’s this: AGI-level breadth of competence has already quietly arrived, progress will keep compounding for years, and eventually — like every technology before it — it will plateau. Just not yet, and not suddenly.