The minimum size of a functioning company keeps falling, and most people are still describing it with numbers from 2021.
I watch this from an uncomfortably good vantage point. I run an agency with a team, and I also build products where the team is mostly me and a set of systems.
The gap between those two experiences has become the most interesting thing in my working week, and it is the same split I described when AI split your team in half.
What actually collapsed
The functions that disappeared first were not the creative ones. They were the translation ones.
For fifteen years, a large part of every project was moving an idea between specialists. A concept became a brief, the brief became a design, the design became a ticket, the ticket became code, the code became a deployment. Each handoff cost time, money, and precision.
That chain is where AI removed the most work. Not by being brilliant, but by being competent at translation. A sketch becomes a rendered concept. A description becomes a working page. A rough spec becomes something you can click on the same afternoon.
I built the first working version of a product last year across two evenings. In 2019 that same scope meant a developer, three weeks, and a conversation about priorities I would have lost.
The output was not better than what a good engineer would have produced. It was faster, cheaper, and mine, and for a version nobody had validated yet, those three things mattered more than quality.
The constraint moved, it did not disappear
Here is where the solo founder story usually goes wrong, and where I want to be precise, because I have made this mistake with real money.
Removing the execution bottleneck does not remove the bottleneck. It relocates it, and the new location is worse, because it is inside your own head.
When you could only ship one thing a month, the scarcity did the prioritization for you. Bad ideas died in the queue. Now every idea is buildable in a day, which means every idea gets built, and you end up with six half-alive products and no distribution for any of them.
I have a folder of these. Genuinely working software, all of it, and most of it never met a customer. It is the solo version of why most AI pilots die in the middle.
The second failure is subtler. AI is confidently mediocre in exactly the places where mediocre is expensive. It will produce a serviceable brand voice, a serviceable pricing page, a serviceable onboarding flow. Serviceable is fine for version one and fatal by version three, because serviceable does not get chosen over an incumbent.
Somebody with sixteen years of pattern recognition can tell when the output is serviceable rather than good. Somebody without that cannot, and they ship it, and they wonder why conversion sits at one percent.
What this means if you run a real company
The strategic read is not “fire everyone and buy tools”. That version of the story sells newsletters and destroys businesses.
The read is that your competitive set just widened in a specific direction. You are no longer only competing with companies your size. You are competing with individual operators who have no overhead, no approval chain, and no legacy commitments, moving on the specific slice of your market they find most interesting.
They cannot do what you do at scale. They do not need to. They need to do one part of it better, for a segment small enough that you never bothered to defend it.
The defensive move is not more tooling. It is deciding which parts of your business genuinely require the organization you have built, and being honest about the parts that exist because they always have.
Every agency has a process that made sense when the work took three people. Some of those processes now consume more time than the task.
The offensive move is to run your own small unit as if it were one of those solo operators. Give one person the whole stack, the mandate, and no approval chain, and watch what ships. It is uncomfortable, and it is the clearest diagnostic you will get about your own overhead.
The part that still cannot be compressed
What has not changed is the thing nobody wants to hear, because there is no tool to buy for it.
Knowing which problem is worth solving. Knowing that a customer’s stated objection is not their real one. Knowing when the numbers are right and the decision is still wrong.
That is accumulated judgment, and it is still priced in years.
A solo operator with taste and a full stack of systems is now a serious competitor. A solo operator with systems and no taste is just producing more output, faster, in the wrong direction.
The tools got cheap. Knowing what to point them at did not.