The curve doesn't care what runs it.

A single technology rides an S-curve: slow start, steep middle, saturation ceiling. The exponential everyone quotes is a cascade of those curves. The dangerous question is which one you are riding -- and whether someone is quietly starting the next one underneath you.

In 1999, in The Age of Spiritual Machines, Ray Kurzweil made a prediction that sounded absurd at the time: human-level artificial intelligence by 2029. He has reaffirmed it for a quarter of a century since, most recently in The Singularity Is Nearer (Viking, June 25, 2024), the New York Times bestseller he wrote as a Principal Researcher and AI Visionary at Google, where he has worked since 2012.

You can take the 2029 date or leave it. That is not the interesting part. The interesting part is the engine he built underneath it -- the Law of Accelerating Returns, an idea he first set out in 1999 and named in a 2001 essay. And one claim inside it is the most useful timing tool a navigator can carry: the exponential trend in computing does not depend on any particular technology. The curve does not care what runs it. The rest of this piece is about why that is true, where Kurzweil's own numbers should be read with a careful eye, and how the same lens reads the curve under your own field.

One technology rides an S-curve. People misread it two opposite ways.

Take any single technology in isolation and its progress traces an S: a slow start while the thing barely works, a steep middle where it improves fast, and a saturation ceiling where it bumps into a physical, economic, or adoption limit and flattens out. That shape is one of the most reliable patterns in the study of technology, formalized by Richard Foster in Innovation: The Attacker's Advantage (1986).

The trouble lives in the steep middle, because from inside it you cannot see the shape. So people make one of two opposite errors. They read the steep part as a straight line and underestimate what is coming, the way nearly everyone under-forecast the early internet and mobile. Or they read the steep part as forever-vertical and price in infinite growth, missing the ceiling that every individual technology eventually hits. Same segment, two wrong conclusions. The ruler does not tell you which one you are looking at.

Both errors come from staring at a single curve. The resolution comes from realizing that the long exponential everyone argues about is not a single curve at all.

The technology underneath changed five times. The trend never bent.

The exponential is a cascade of overlapping S-curves.

Here is Kurzweil's genuine insight, and it is the part worth keeping. Plot the price-performance of computation back to 1939 and you get a straight line on a log scale -- a clean exponential. But the technology under that line changed five times: mechanical relays, then vacuum tubes, then transistors, then integrated circuits, and most recently GPUs and AI accelerators. Each of those is its own S-curve. Each one started slow, climbed steeply, and saturated. And yet the overall line never bent.

It never bent because the next paradigm started climbing before the previous one flattened. The new curve picked up the trend exactly where the old one was running out of room. That overlap is the whole trick. The envelope across the tops of all five curves stays exponential precisely because no single technology has to carry it the whole way. This is what the title means, and it is genuinely Kurzweil's framing in his price-performance paradigm chart: the curve does not care what runs it. When one substrate hits its ceiling, another takes over, and the line keeps climbing as if nothing happened.

So both reading errors dissolve at once. The line is not really straight (so the straight-line reader underestimates), and no single curve is vertical forever (so the forever-vertical reader is wrong too). The exponential is real, but it is a relay race, not a single runner.

The shape is solid. The scorecard is self-reported.

This is where a sophisticated reader has to be careful, because Kurzweil's presentation mixes a robust structural claim with numbers that are his own. The cascade of S-curves is well grounded. The specific figures around it are not independently audited, and it matters to say so plainly.

Start with the famous accuracy claim. The often-quoted “86 percent” comes from Kurzweil's own 2010 review, “How My Predictions Are Faring,” in which he graded his own 147 predictions and counted 115 as “entirely correct” and another 12 as “essentially correct.” By his own count, that is impressive. But it is a self-grading exercise, not an independent audit, and the rubric is generous about what counts as correct. Read it as “by his own scorecard,” never as a neutral fact. (Background: Ray Kurzweil.)

The same caution applies to the big numbers in The Singularity Is Nearer: a roughly 75-quadrillion-fold gain in hardware price-performance since 1939, around 11,200 times more compute per dollar since 2005, and a roughly millionfold gain in software efficiency. These come from Kurzweil's own proprietary chart, not from an independent measurement. The frequently repeated “about 10x improvement per year” is an approximation, not a precise figure; independent estimates of effective AI-compute growth land in a wide band rather than on a single clean multiple. Treat all of these as Kurzweil's own measure, and do not quote the 10x as exact.

One more correction, because the legend tends to inflate it: Kurzweil was a genuine prodigy young -- he appeared on television in 1965, at 17, with a computer he built that composed music -- but he did not formulate the exponential-growth thesis as a teenager. That came later, as an adult, across 1990 and 1999. The prodigy story is true. The thesis-as-a-teenager story is not. None of this weakens the structural point. It just keeps the structural point separate from the salesmanship around it.

How to read the curve under your own field.

Timing is the gap between Nvidia and Kodak.

The cascade is not just a chart about the past. It is happening right now, in public. Across 2024 to 2026, the major AI labs converged on near-term timelines for human-level systems -- the kind of cluster of expert opinion that, in S-curve terms, marks the steep part of a new paradigm climbing fast. Whether you believe the dates, the convergence itself is a timing signal: a new curve is carrying the trend, and the old assumptions about how slowly capability improves are getting overtaken.

That is exactly the moment where reading the curve correctly separates winners from casualties. Nvidia read the handoff -- general-purpose compute giving way to accelerated, parallel compute -- and positioned at the base of the new curve. Kodak, which had invented the digital sensor in 1975, read its own film curve as permanent and optimized it toward the ceiling while the new curve climbed underneath. Same era, same information available, opposite timing reads. One rode the next S-curve; the other defended the last one.

The structure here is the one Utterback and Abernathy described as the era of ferment before a dominant designsettles: the moment a new paradigm starts climbing is precisely when incumbents are most tempted to keep polishing the old one. The navigator's advantage is not predicting which technology wins. It is asking the prior question: is my field on a single S-curve near its ceiling, or at the base of a new one?

This one is not a metaphor. It is a clock.

The S-curve is not a story MindrianOS tells. It is a first-class move it runs. /mos:analyze-timingplaces your field's core technology on the S-curve clock -- early signal, inflection, rapid diffusion, or approaching ceiling -- and surfaces the position as a decision gate, so you argue about where you sit on the curve before you bet a single dollar on its continuation.

For the same timing lens applied to a domain that dissolved rather than handed off, see the great horse manure crisis. That piece is about a curve that ended; this one is about a curve that keeps getting handed to the next runner. Both come down to the same question the whole category keeps skipping: which curve am I actually on, before I bet on it continuing?

Copy this. Fill in your field. Paste it into MindrianOS.

I want to understand the timing in my domain, and I am not sure whether I am riding a curve that is about to flatten or sitting at the base of a new one.

My domain: [DESCRIBE YOUR FIELD IN ONE OR TWO LINES -- what you research, build, or operate, and the core technology or capability everything else depends on].

Here is what I notice. For as long as anyone in my field can remember, the core capability has improved on a steady, almost straight upward line. People plan as if that line continues forever. But a single technology does not climb forever -- it follows an S-curve: slow start, steep middle, then a saturation ceiling. I cannot tell from inside whether the steep part I am in is still climbing, or whether I am approaching the ceiling while a different paradigm quietly starts its own S-curve underneath me.

It reminds me of the computing story. The straight exponential line of price-performance runs back to 1939, but the technology underneath it changed five times -- relays, vacuum tubes, transistors, integrated circuits, then GPUs and AI accelerators. Each was its own S-curve that saturated. The trend never bent because the next paradigm started climbing before the last one flattened. The curve did not care what ran it.

Help me think through this properly. First: is my field on a single S-curve approaching its ceiling, or at the base of a new one? Do not jump to a strategy. Place it on the curve, show me which paradigm is actually carrying the trend right now and which one might carry it next, and tell me where the real opening is -- the bet that rides the next curve rather than optimizing the one that is about to flatten.