Every few months a new AI model appears seemingly out of nowhere, matching systems that took years and fortunes to build, and the same explanation races around the industry: they just copied a smarter model. It is a tidy story, and it is the wrong one. Copying is real, and it does happen, but it plays a much smaller role than people assume, and it stops mattering far sooner than anyone expects. Mistaking it for the whole answer is how you misread your competition.
The apprentice who couldn’t start
Imagine you hire a brilliant but totally inexperienced apprentice. She is sharp, eager, and has read every book in the library. But she has never actually done the job. Now imagine the only way she is allowed to learn is this: she has to complete a real task successfully, and only then does she get to keep the lesson. Fail, and she learns nothing. Succeed, and she gets a little better forever. That, in plain terms, is how the best AI models are trained today. The industry calls it reinforcement learning. I will just call it learning by doing.
It sounds simple, and at heart it is. But hidden inside that one rule; learn only from what you finish, is the problem that quietly decides which companies pull ahead and which ones stall out. Almost nobody outside the labs talks about it, and almost everybody misreads it.
The trap nobody sees coming
Here is the catch. Our apprentice can only learn from tasks she manages to finish. If a job is too hard and she fails it every single time, there is no lesson to keep, so she never improves at it. She is trapped in a vicious circle: she cannot get good at the hard work because she cannot finish it, and she cannot finish it because she is not good enough yet.
Engineers have a name for this circle. They call it the cold start problem. It is not a minor technical footnote. It is the single biggest wall standing between a mediocre model and a great one, and it is the reason raw talent and endless reading are never enough on their own. Without a way through this wall, even the most promising system sits still, spinning its wheels on exactly the problems that matter most.
If you have ever watched a talented new hire drown on their first real project, or a capable team fail to break into a market they clearly understand, you have already seen this dynamic in the wild. Ability is not the bottleneck. The first success is the bottleneck.
Bringing in the mentor
So how do you break the circle? You bring in a mentor. For a little while, you let a far more experienced expert sit beside the apprentice and walk her through the tasks she keeps failing. She imitates the expert, and for the first time she completes a few of those impossible jobs.
This borrowing of skill from a stronger model is the part everyone has heard about. In AI it is called distillation, and it is the entire basis of the “they just copied us” headline. When a newcomer suddenly rivals the leaders, distillation is the thing critics point to. It sounds damning. It sounds like the whole story. It is neither.
What the copying story gets wrong
The mentor does not make the apprentice great. The mentor simply gets her unstuck. Once she has completed a few hard jobs, even by imitation, she finally has real successes to learn from. The vicious circle is broken. From that moment on, she does not need the mentor at all. She practices on her own, keeps the lesson from every win, and climbs week after week entirely under her own power.
Look closely at what the expert actually contributed. Not mastery. Not the years of compounding practice that follow. Just a ladder over one specific wall, used once and then discarded. Distillation is the on-ramp, not the engine. Everything that made the model genuinely good happened after the copying stopped.
Why this matters in the boardroom
This is where a comforting story becomes a dangerous one. “They just copied us” frames a competitor’s progress as a single act of theft: something that can be blocked, licensed, litigated, or regulated away. It suggests the problem has a legal solution and a clean end date.
But distillation is not a moat, and preventing it is not a strategy. It is a temporary ladder that any serious, well-funded team can build once and then throw away. The durable advantage lives somewhere else entirely: in the ability to keep generating your own successes and compounding them, quarter after quarter, without anyone’s help. If your competitive position depends on rivals never getting their first win, you do not have a competitive position. You have a delay.
The lesson that reaches beyond AI
Strip away the technical vocabulary and this is a story about how hard things get started. Getting a person, a team, or a company to its very first taste of success at something genuinely difficult is the expensive, painful, uncertain part. It is where most efforts quietly die. Not from lack of talent, but from never getting the one win that makes the next win learnable.
But once you are winning even a little, and you have built a system that captures the lesson from every win, momentum switches sides and starts working for you. The climb from good to great is a grind, but it is a grind you control. The climb from zero to your first win is the part that breaks people.
So the next time a rival seems to leap forward overnight, resist the easy explanation. The overnight part was never really about copying anyone. It was about escaping the cold start, and then quietly compounding every win that followed. The mentor gets forgotten. The momentum is what you have to worry about.