Over the weekend, I nearly bought a subscription to somebody else’s thinking.

You might have heard of an essay called Solve Everything, written by Alexander Wissner-Gross with a licensed chapter from Peter Diamandis. Alongside it, Salim Ismail has been on the podcast circuit arguing that most companies will run on about a quarter of their current headcount, and that you have 5-7 years to make the change or you are gone. The playbook ships as a continuously updated skill you can load into your AI, because the tactics apparently move too fast to print. That skill used to be free and I went searching for that but it’s no longer free.

I read the essay properly, and it was useful. It made me notice something about a programme I run that I had genuinely not seen. The programme measures return on AI spend, which is what the tooling cost, what it saved and where it is redeployed, with a North star of EBITDA and enterprise value uplift. Reading their material I realised that is a tool-era measure, and the thing that actually gets scarce as execution gets cheap is not tool budget, it is cognitive budget. Whose attention, on what, and how much of a business’s thinking capacity gets burned producing a given outcome. Nobody is teaching that yet, including me.

Good. That single realisation was worth more than most paid material I have read this year, and I want to be straight about that before I say anything else. I love their work. It has had a real and positive effect on how I think about what I am building, and this piece exists because I took it seriously enough to check it, not because I wanted to take it down.

Then I noticed the second thing, which is that I had accepted their timeline without checking it. Not adopted it, exactly. Just absorbed it. 5-7 years had lodged in my head, and I could not have told you where the number came from or whether the person saying it had ever been right about anything.

So instead of buying the skill, I went and graded them, using my system to help, of course.

What follows is not a takedown. It is a translation. These are careful, generous, mostly-right people writing from a very specific place, and the further you sit from that place the more their material needs converting before you can use it. I’m based in Newcastle, UK, and feel far from the epicentre in San Francisco. I’ve not seen many conversion pieces, so I have had a go.

The method

You cannot grade a prediction about 2035. You can only grade one whose date has already passed. So that is all I looked at: specific, dated, falsifiable claims, made or prominently relayed, where the deadline has been and gone. Everything else got logged and set aside.

That rule does most of the work, because it strips out the part that feels like insight and leaves the part that can be checked. It also means the sample is smaller than you would like, and skewed toward the older material. Both true. It is still infinitely better than the alternative, which is grading nobody and quoting everybody.

The record

On Diamandis, whose corpus is by far the largest and the easiest to check, I could grade thirty domains. Nine hit. Four partial. Seventeen missed.

Roughly one in three. Before anyone reaches for that as a gotcha, one in three on genuinely hard forecasts over fifteen years is not a disgrace. Most people who make public predictions do not publish enough dated ones to be graded at all, and he does, which is to his credit and is the only reason this exercise was possible.

And some of the hits are excellent. Not lucky. Excellent.

Now the misses, and this is where it gets useful.

The pattern

Look at those two lists again and the shape jumps out.

Every single hit is either a cost curve in an information technology, or a prize he designed and funded himself. Every single miss is a claim about deployment, adoption or commercialisation of something bounded by physics, biology, regulation, capital stock or unit economics.

That is not the usual summary. The usual summary is that these people are directionally right and temporally early, and you should trust the arrow while discounting the clock. I went in expecting to find that. It is too generous, and it hides the thing you can actually use.

The failure is not optimism about timing. It is category confusion. Where the subject is information, he is not just right, he is conservative. Where the subject is bounded by the physical world, the curve goes linear and he applies exponential framing anyway.

The one line worth keeping from all of this

A cost falling 90% tells you almost nothing about adoption share within a decade.

Solar is the cleanest example. The cost call was exactly right. The system-share call, majority displacement of fossil fuels by 2030, was not close: solar supplied about 2.9% of global primary energy in 2024 and about 8.7% of electricity in 2025. Deployment is bounded by grids, land, permitting, capital and planning cycles. None of those are information technologies, so none of them compound.

There is a second pattern too, and it is more reliable than the first. Right category, wrong vehicle. Watson, Hyperloop One, Lilium, Volocopter, Plenty, bioprinting. The capability often arrived. The named company almost never delivered it. Which means that when somebody offers you a company as proof of a trend, the company is the weakest part of the argument.

Considering proximity

Here is where I think the record above comes from, and it is not carelessness. These are careful people. It is position.

All three of them operate at the epicentre. Not a criticism, a description. Their rooms are venture-funded deep-tech founders, their peers are running companies with balance sheets between ten million and ten billion dollars, their talent pool is the densest on earth, their customers are early adopters by temperament, and their capital is patient enough to fund something for six years before it works. From inside that room, the observations are accurate. Execution really is nearly free. Coordination really is the expensive part. Firms really are running leaner. They are reporting the weather where they stand, and they are reporting it honestly.

The error is not in the observation. It is in the assumption that the view generalises without adjustment.

Distance from that epicentre is not measured in miles. It is measured in five things, and most businesses are a long way out on all five.

Apply those five and the five-to-seven-year window stops being a countdown and becomes a gradient. The people closest to the epicentre are already living in the world the essay describes. The people furthest out will get there, but later, more slowly, and by a different route. Neither group is behind or ahead. They are at different distances from the same weather.

The temperance this needs, said plainly

If you run a business outside a major tech hub, in a sector with physical or regulatory drag, funded out of your own cashflow, with no spare person to run a change programme, then the honest translation of “you have five to seven years” is this: the direction is real, the deadline is not yours, and the constraint you will actually hit is not the technology.

It is that you and your people have to learn to work differently, and there is no version of that which moves at the speed of a model release.

A live example, which is my own

I said at the top that reading this material made me notice my programme measures the wrong thing eventually. That is true. What I did about it is the more useful part.

The programme teaches business fundamentals applied to current tools, and it measures return on AI spend, because that is the measure the people in the room can actually move this quarter. Return on cognitive spend is a better measure and it is where this is heading. I am not switching to it, and I want to explain why, because the reasoning generalises.

Most businesses have no baseline for it. You cannot measure how much of your organisation’s thinking capacity a given outcome consumes if nobody has ever measured anything about how work moves through the place. Asking them to make that jump is not ambitious teaching, it is skipping a rung. And the rung you skipped is the one that holds the weight.

So the fundamentals are not a lesser version of the future measure. They are the thing that makes the future measure possible. You learn where work actually goes, you get a scoreboard on something, you get comfortable seeing your own process honestly. Do that and the cognitive-spend measure becomes obvious later, almost trivial. Skip it and the measure is meaningless, because you have nothing to compare it to and no habit of looking.

What changed, in the end, was one slide. The programme now names the measure it becomes and shows people where the scoreboard is heading, while still grading them on the one they can move now. That is the whole adjustment. It cost me an afternoon and it means nobody leaves thinking today’s measure is the last word.

If you teach, sell to, or advise businesses that are not at the epicentre, I think that is the shape of the right response to all of this material. Not a rebuild. Not a pivot to the future state. Name the destination, teach the rung in front of them, and be honest that the two are different things.

The provenance of claims

Three of the most-repeated claims in this field have travelled a long way from where they started, and picked up authority nobody ever granted them. That is not something done to us. It is what happens to any claim repeated often enough, and I did it myself: 5-7 years was sitting in my head as a fact and I could not have told you where it came from.

The 86% accuracy figure. It gets repeated as though it were an audit of Diamandis. It comes from a blog post he wrote in September 2017 which opens: “Ray Kurzweil has a documented 86% accuracy rate in his technological predictions.” It is Kurzweil’s assessment of Kurzweil, relayed once and repeated ever since. It was never a claim about Diamandis, and he never said it was. As far as I can find, no independent assessment of his own record existed until this one, and I would genuinely like to hear if I have missed it.

40% of the Fortune 500. Usually credited to Ismail. It is not his claim, and he does not present it as one. Exponential Organizations attributes it to Babson’s Olin school in 2011; the endnotes in Diamandis and Kotler’s BOLD trace that figure to a Fast Company advertisement. Both books cited it accurately. The distortion happened entirely in the retelling. Worth knowing too: the clock ran 2011 to 2021, so it is already due, and “no longer survive” is a much higher bar than dropping off a list.

The evidence behind the framework. There is one peer-reviewed paper supporting the Exponential Organizations model. It compares the top ten against the bottom ten of the 2014 Fortune 100 over 2014 to 2021, using an instrument built after the outcome window. Four citations, no replications, authors affiliated to the model’s own research arm. That is thinner than the confidence the model is usually quoted with, and it is also completely ordinary. Almost every management framework in print has the same problem, including ones I use and recommend. An independent review of the underlying concept published last year names the gap explicitly.

None of that makes the ideas useless. It makes them ideas rather than evidence, which is a different thing and should be priced differently. It matters because at some point you are going to repeat one of these in a room, and it is better to know which kind of thing you are holding.

What I actually do with them now

I did not throw the material out. I use it more than I did before, and more carefully. Four rules came out of the exercise and they work on anyone, not just these three.

On that last one, I want to be careful, because I sell things too. A paid offer does not invalidate the thinking behind it, and I would not accept that argument about my own work. The point is narrower. Urgency is the product. The frameworks are given away because they are the top of the funnel, and the clock is what converts. So read the frameworks freely and hold the deadline loosely.

What I actually think of them

Higher than before I started.

They publish enough dated, specific claims to be graded at all, which almost nobody in this field does. The mechanisms they describe are sound and given away free. Wissner-Gross’s maturation curve is the most useful analytical tool I have picked up this year and I now use it weekly. Ismail’s reading of why firms exist and why that logic is inverting is, I think, basically right, and he had the intellectual honesty this year to say out loud that his own twelve-year-old framework no longer holds. That is not a small thing to say when you have books in print.

What I would not do is take the clock from them. The clock is the part that is being sold, and it is the part with the worst record. Take the mechanisms, which are free and mostly right, and set your own dates from your own position.

Which brings me to the reason I bothered with any of this.

There is going to be an enormous number of people trying to run a business the way I have been running mine: small, agent-heavy, no headcount to speak of. The material that exists for them right now is either written for companies a hundred times their size, or it is written by people selling a course. Neither is translation. Both assume a proximity to the epicentre that most of the audience does not have.

I have spent about 5 years building the thing their essays describe (18 months with AI supporting) at the smallest possible scale, without a transformation budget or a team to transform. I got plenty of it wrong. Some of it works well enough that I would recommend it. The sister piece to this one is that account: what their vision looks like from inside a company of one, where it matched what I found, where it did not, and what I would tell someone setting out down the same path now.

This piece was about whether to trust the map. That one is about what the territory actually looks like when you walk it.

The one-line version, if you take nothing else

Read them. The mechanisms are good and they cost nothing.

Then set your own timeline from your own position, because the deadline in the essay was written from a room you are probably not standing in, and the constraint you will actually hit is not the technology. It is how fast the people around you are willing to work differently, and nothing in any essay changes that.

Sources for the graded claims are all public: diamandis.com’s blog archive, the TED2012 transcript, Exponential Organizations, solveeverything.org, and the usual reality checks from IRENA, ITU, Ember, ONS, ClinicalTrials.gov and company filings. Every figure here was fact-checked adversarially before publication and one claim in my first draft was struck as a result. If you find an error, tell me and I will correct it in place, and I would rather hear it from you than not hear it.

Enjoyed this? Join the newsletter.

One email a week. What I'm building, learning, and what's actually working. No fluff.