I Stopped Planning Five Years Out. That's Too Long.

AI token prices dropped 280x in 24 months. That number is from the Stanford AI Index, cited in a piece Peter Diamandis published this week called "The Next 5 Years: A Supersonic Tsunami". The piece is good. Dense with data. Frontier model prices are falling 10x per year — from $20 per million tokens down to roughly $0.40. Dario Amodei thinks AGI could arrive as early as late 2026 or 2027. Humanoid robots are already shipping. It's a lot to absorb.

I read it twice. And I kept catching on one thing.

The article's entire argument is that the pace of change is so extreme, so unprecedented, that most people aren't prepared for what's coming. I find that argument convincing. And yet the piece is structured as a 5-year plan. If you take the data seriously — really seriously — I think you land somewhere else entirely. You land at the conclusion that a 5-year plan is the wrong instrument for this moment.

The value of long-horizon planning was always based on an assumption: that the future is roughly legible. You could reason about it. The macroeconomic environment might shift, competitors might move, but the underlying structure of industries, cost curves, and capabilities changed slowly enough that a 5-year map was useful. It wasn't a crystal ball. It was a way of making present decisions coherent with a plausible future.

That assumption is broken now.

When cost curves move 10x per year, your year-3 assumptions are already fiction by the time you hit year 2. I'm not being glib about this. I've lived it. I made bets three years ago that looked obviously right at the time and were clearly wrong 18 months later — not because I had bad judgment, but because the environment shifted faster than the timeframe of the original bet. I've also had the reverse happen: decisions I wasn't fully confident in turned out to be early and correct, which felt like luck as much as skill. The planning frameworks I was using five years ago were built for a world that moved at a different speed.

That's where this gets interesting to me. The question isn't whether to plan. The question is what kind of planning still works.

My take: the horizon has compressed, but strategy hasn't disappeared. It's just changed shape.

What I actually do now looks something like this. I make near-term moves with high conviction — decisions where I have enough signal that I can act with confidence across a 6-to-12 month window. Those decisions I hold firmly. Beyond that, I try to preserve optionality rather than commit to a fixed path. The question I ask isn't "where will I be in five years?" It's "what positions do I want to be in so that I can move well when the next shift happens?"

That's a different orientation. It means I favor reversible decisions over irreversible ones wherever I can. It means I make directional bets — I think AI is eating software costs, so I'm betting in that direction — without pretending I know exactly how that plays out by 2029. The direction has high confidence. The shape of year 4 doesn't.

It also means I pay a lot of attention to what I'm learning, not just what I'm executing. In a slow-moving environment, information compounds slowly enough that you can make a plan and then mostly execute it. In a fast-moving one, information becomes the primary input to strategy. You're updating more than you're executing. That's uncomfortable for people who were trained to think of learning as the precursor to planning and planning as the precursor to action. Now they're all happening at once.

I should say: I don't think this means nothing matters beyond next quarter. That's the overcorrection. Diamandis cites Kurzweil's projection of Longevity Escape Velocity by 2033 — a point where we're adding more than a year of life expectancy per year. If that's even roughly right, the time horizons for human potential extend dramatically, which is one of the stranger and more interesting implications buried in the piece. The right response to uncertainty isn't to shrink your ambitions. It's to hold them more loosely, and stay positioned to move.

Five years used to be the right unit of analysis for a strategic plan. I think it's now closer to the outer edge of where reasoning stays honest. And even then, the plan needs to be written in pencil.

The Diamandis piece is worth reading. The data is genuinely useful. I just think its own evidence argues against the format it's presented in — and the more useful output isn't a 5-year forecast, but a sharper set of near-term convictions and a clear sense of what kind of future you're directionally betting on.

That's actually harder to write. It requires committing to a view without the comfort of a roadmap. But that's the actual work now.