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

Short version of part one: I graded the people writing the future-casting material, found they land about one in three on dated claims, found the pattern in the misses is useful, and concluded that you should take their mechanisms and set your own dates. I also said their material is written from the epicentre, and needs converting before most people can use it.

This is the conversion. Not a rebuttal, and not a rival theory. It is what happened when I spent about five years building the thing their essays describe, at the smallest possible scale, and then read the essays.

The thesis I am converting

Salim Ismail’s argument goes roughly like this. Ronald Coase explained in 1937 why firms exist at all: because coordinating work inside a company is cheaper than transacting for it in the market. Agentic AI, the argument goes, has inverted that. Execution is becoming nearly free while coordination stays expensive, so the reason to hold a large organisation together is dissolving. Firms therefore collapse into lean entities holding intellectual property, data and agents, run by a fraction of their current headcount, with humans on governance rather than production.

Wissner-Gross’s Solve Everything runs alongside it with a maturation curve: domains move from ill-posed, to measurable, to repeatable, to automated, to industrialised where buyers stop paying for effort and start buying verified outcomes, to commoditised.

I think the mechanism in both is right. I have very little confidence in the dates, for reasons part one covers in detail.

But notice what that thesis is describing. It is a story about subtraction. A firm of four hundred people works out how to be a firm of ninety. Almost all the material, the case studies, the sprints, the certifications, the transformation programmes, exists to help large organisations shed.

I have never had much to shed, like many other lean businesses.

The same destination, reached by addition

For the last 5 years I have been running my own businesses with effectively no headcount. A exec search firm, a lead-generation marketplace, a job board and a consultancy. Those are the company-of-one experiment, and they are where most of what follows was learned.

Alongside that I work fractionally inside a bakery in a Northumberland village. That one is different in every way that matters. It has staff, ovens, a lease, food hygiene, weather, and a culture built over years that is worth more than anything I could automate. I am not running it as a company of one and I would not try. I am one person inside it, using AI to do my own work at a level I could not otherwise sustain, while taking a very traditional business through digital transformation slowly and deliberately.

I keep coming back to it in this piece because it is the honest test. The company-of-one experiment proves the thesis can work when conditions are perfect. The bakery is what happens when they are not, and most businesses are the bakery.

I did not shed anything to get here. I added. I started with a person and a laptop and put agents around them one at a time, each because something specific was broken. That is the same destination the thesis describes, arrived at from the opposite direction, and the interesting thing I found is that the additive route is enormously easier.

There is no change programme when there is nobody to change. No middle management to have a difficult conversation with. No sunk cost in an existing process, no political defender of the old way, no procurement cycle, no board to convince. The entire cost of the transformation their material describes is the cost of moving people, and if you have no people to move, that cost is zero.

Which means the group best placed to run their thesis is the group their material was never written for. That is not a criticism of them. It is just an observation about who is standing where.

The honest caveat is that this only holds while you genuinely have nobody. The moment there are people, all of the cost their material describes comes back, and it comes back as the hardest kind of cost. Which is the other half of what I have been doing, and the more instructive half.

What their model got right about mine

Reading Solve Everything and the singularity material after building, rather than before, was a strange experience. Most of it described things I had already built without having a name for them. I want to be specific about that, because vague agreement is worthless.

Five of their ideas, and what each one turned out to be once I was living in it.

Coordination is now the expensive part. The single truest thing in the whole corpus. My constraint has never been the ability to do work. It is knowing what state everything is in. Almost everything I have built is coordination infrastructure wearing a different hat.

Build the harness, not the hero. I wasted months building agents that solved one problem brilliantly and taught me nothing reusable. The ones that survived are the ones that made a class of problems easier. I would not have framed it that way without their language.

A lean entity holding IP, data and agents. This is a fair description of what I ended up with, and I did not set out to build it. It is what you converge on when you keep asking what actually has to sit inside the business.

Humans on governance, machines on production. Right, and harder than it sounds. The governance job is real work and it does not shrink. It changes shape from doing to checking, and checking well is a skill I had to learn badly, in public, at some cost.

The maturation curve. The most useful single tool I took from any of it. Knowing which rung something sits on, and which rung the category is heading for, turns a vague unease into a specific question.

So the honest verdict is that their thesis described my experience better than anything else I have read, and gave me vocabulary for things I had built blind. That is worth a lot and I am not going to be coy about it.

What it got wrong, or at least what it missed

Four things, all of which come from the same place: their model is written for organisations, and a company of one is not a small organisation. It is a different object.

1. The bottleneck moves to you, immediately and permanently. Their model assumes coordination cost falls when you remove people. In a company of one it does not fall, it concentrates. Every decision, every check, every piece of judgement routes through one head, and that head has a hard ceiling that no amount of capability raises. The failure mode of the large firm is politics. The failure mode of the company of one is the operator becoming the single point of failure for a system that now runs faster than they can supervise it. Nothing in their material prepares you for that, because in their world you hire.

2. Governance is not a layer, it is most of the job. The diagrams put governance in a box at the side. In practice it is the majority of my week. Deciding what is allowed to happen without me, checking the things that did, noticing when something has been quietly broken for six weeks. I once had a scheduled system stop running and did not notice for seven weeks, because nothing was watching the watcher. That is the actual texture of this work and it does not appear in any framework I have read.

3. Trust is earned per-task, not granted per-system. The models talk about autonomy as a setting. It is not. It is a per-task judgement built up slowly from evidence, and it is not transferable. I trust some of my systems completely and some not at all, and the difference is not sophistication, it is track record. Anyone who tells you to switch on autonomy has not run it.

4. It has almost nothing to say about a business with people in it. This is the one that matters most, and the bakery is where I learned it.

Their model treats headcount as the thing to be reduced. Inside a traditional business with staff who have been there years, headcount is not a cost line, it is the business. The recipes, the standards, the way a regular is greeted, the reason people come back. Take an axe to that in the name of an operating model and you will hit your efficiency target and destroy the asset.

So the work there looks nothing like the thesis. It is digital transformation in the old sense, done at a pace people can absorb. One process at a time. Get the numbers visible before changing anything. Prove a change works on something small before asking anyone to trust it. Leave the things that are working alone, even when they are unfashionable, because working is the highest standard there is.

AI is fundamental to that story, but not where you would expect. It is not replacing anyone. It is how one fractional person does the work of a much larger engagement: seeing what the numbers actually say, modelling a pricing change before proposing it, working out which lines earn their shelf space, preparing properly for every conversation. The leverage lands on me, and what the business gets is a level of attention it could not otherwise afford. That is a genuinely different shape from anything in their material, and I suspect it is the shape most small businesses will actually encounter first: not an agent in the business, but a person with agents helping the business think.

The step change there came from business fundamentals applied with more thinking capacity than one person would normally have. Not from automating anything. Their thesis, applied literally to that business, would have had nothing useful to offer and quite a lot to damage.

The correction that fell out of the bakery

Their model treats agents as an execution technology, and measures success by how much execution you can remove from humans. In every business with people in it that I have touched, the value showed up somewhere else entirely: as thinking capacity applied to decisions that were previously made on instinct, because nobody had the time to look properly.

That reframing survives every timeline. It is true if the singularity arrives in 2031 and it is true if it never arrives at all, which is exactly the property you want in something you are going to build a business on. It also has the advantage of being something you can do next month without asking anyone to change who they are.

What I would tell you if you are heading this way

A lot of people are about to try this. The material available to them is either written for companies a hundred times their size or written by someone selling a certification. So here is the practitioner version, and it is deliberately unglamorous.

Where this leaves me

I did not write these two pieces to argue with anybody. I wrote them because I took the material seriously, checked it properly, and then found that the checking and the building together produced something neither would have produced alone.

What I have ended up with is a sentence I did not have six months ago, and it is the only real thing this whole exercise gave me that I did not already own.

The sentence = my MTP

Prove the one-person, many-agent company, then teach every £0-20M founder-operator to run their business the same way.

The order matters. The proof comes first, because an opinion about this is worth nothing and a working example is worth a great deal. The teaching is what makes it worth doing at all.

Their vision is a large organisation working out how to become small. Mine is two things running side by side: a company of one working out how to stay small while doing considerably more, and a traditional business with real people in it being taken forward slowly, with its culture intact, by someone using the same tools to think better rather than to remove anybody. Same direction of travel. Two completely different speeds, and both of them legitimate.

I think the second route is the one most people reading this will actually take, and I think it is currently the worse-documented of the two by a distance. That seems worth fixing, so I am going to keep writing it down as I go, including the parts that did not work. Especially those.

Part one, I Graded the Futurists, has the full scorecard and the sources. If you are running something small and heading down this road, I would like to hear what you are finding, particularly where it disagrees with the above.

Enjoyed this? Join the newsletter.

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